(PDF) Pre-cognitive semantic information

Noname manuscript No. (will be inserted by the editor) Pre-cognitive Semantic Information Orlin Vakarelov the date of receipt and acceptance should be inserted later Abstract This paper addresses one of the fundamen- 1 Introduction tal problems of the philosophy of information: How does semantic information emerge within the underlying dy- This paper addresses one of the fundamental problems namics of the world? dynamical semantic informa- of the philosophy of information: How does semantic tion problem. It suggests that the canonical approach to information emerge within the underlying dynamics of semantic information that denes data before meaning the world? Let us call this the dynamical semantic in- and meaning before use is inadequate for pre-cognitive formation (DSI) problem. DSI is related to problems information media. Instead, we should follow a prag- #2 and #4 in the list of 18 fundamental problems of matic approach to information where one denes the the philosophy of information that Luciano Floridi has notion of information system as a special kind of pur- compiled (2010; 2004; 2008b). Problem #2 is The I/O poseful system emerging within the underlying dynam- problem: what are the dynamics of information? Prob- ics of the world, and dene semantic information as the lem #4 is the data grounding problem: how can data ac- currency of the system. In this way, systems operating quire their meaning? DSI is also related to a third prob- with semantic information can be viewed as patterns lem, not explicitly enumerated in Floridi's list. This in the dynamics semantic information is a dynamical is the problem of the pragmatics of information: how system phenomenon of highly organized systems. In the can information-using systems exist and what makes simplest information systems the syntax, semantics and them such. I tackle three problems of information the- pragmatics of the information medium are co-dened. ory not out of ambition but out of necessity. Indeed, It proposes a new more general theory of information it is one of the central claims of this paper that, in semantics that focuses on the interface role of the in- the simplest cases of semantic information, these prob- formation states in the information system the inter- lems must be approached simultaneously. In the sim- face theory of meaning. Finally, with the new frame- plest information systems those systems that uti- work it addresses the debate between weakly semantic lize semantic information the pragmatic, semantic and strongly semantic accounts of information, siding and structural/syntactic aspects of information are co- with the strongly semantic view because the pragmatic determined and must be investigated simultaneously. account developed here is a better generalization of it. The paper is related to another kind of problem: can we provide a foundation of cognitive science with the Keywords Semantic information · Pragmatic Informa- notion of (semantic) information? It is my conviction tion · Dynamical Systems · Information Systems · Pre- that we can, and I suggest how elsewhere. (Vakarelov cognitive systems · WSI vs. SSI problem forthcoming) The project of the foundation of cognitive science places a negative constraint on a general theory of semantic information. At least some information sys- tems must be pre-cognitive. Of course, we are cognitive Department of Philosophy, Social Science Bldg. Rm 213, systems, and the kinds of informational media that are University of Arizona, Tucson, AZ 85721, USA. e-mail: interesting for us, such as languages, maps, etc., are

allowed to depend on cognitive tools. The gimmicks of 2 cognition make information a powerful phenomenon, no of organized complex dynamical systems. Section 6 of- doubt. Still, the project demands, information systems fers a general theory of meaning the interface theory and thus semantic information must be able to exist of meaning appropriate for the general information without cognition. systems discussed in the earlier sections. Section 7 ad- My strategy for addressing DSI is this: Start with a dresses the SSI vs. WSI debate. notion of Information System (IS) that is a special kind of autonomous dynamical system interacting with an 2 Canonical Views of Semantic Information environment. Describe semantic information as a cur- rency of the information system. That is, treat infor- Few things are canonical about semantic information, mation for the system not as a primitive but as a de- but for the purposes of this paper I will assume that rived notion, similar to the way currency is a derived canonical views of semantic information have the fol- notion of an economic system. Take a decomposition ap- lowing form: (1) semantic information = data + mean- proach to analyzing the components of semantic infor- ing (+ truthfulness), (2) the data is conceptually pri- mation that is, regard notions such as data, meaning, mary. For the moment I will bracket the debate about and source, as depicting aspects of informational pro- whether semantic information must be true informa- cesses with the information system. Provide a theory of tion. I will return to the issue in Section 7. Floridi meaning, the interface theory of meaning, for the infor- (2003; 2010) describes this as the General Denition mational states (data states) of an information medium of semantic Information (GDI). According to this view, within the information system. to provide an account of information is to provide an Finally, I address a current debate about whether account of the structure of the data, and furthermore to truthfulness should be required as a condition for se- provide an account of what makes the data meaningful. mantic information. The contrast is between a theory The data is a non-empty set of distinctions and each of strongly semantic information (SSI), which demands datum is well-formed. The well-formedness condition, that truthfulness be part of information, and a theory minimally, assumes that it is possible to distinguish be- of weakly semantic information (WSI), which regards tween the dierent datums and to separate data from semantic information as simply meaningful data. My non-data. Some data systems may have further syntac- account of semantic information does not t directly tic structure. For example, languages may have (com- in either of the accounts. It is more general than both. positional) grammar, etc. The question is: of which theory is it a more natural Given the data, one can provide an account of how generalization, SSI or WSI? I will argue that DSI is meaning is determined. There can be dierent theories a more natural generalization of SSI, and thus it pro- of meaning, i.e. dierent ways to specifying content to vides indirect support to the SSI view. However, WSI a datum. For a language or a map, content may be is a theoretically useful notion because it can be inter- provided by a reference relation or by specication of preted as the currency of semantically decoupleable a functional role. It is usually assumed that in order information systems. A unied theory of information, for the question of meaning to arise, one must have if possible, would require both SSI and WSI notions of the system of data on hand. Conceptually, data is more information. primitive than semantics. Data is, or so the assumption The paper is organized as follows: Section 2 intro- goes, a necessary condition for meaning. For, if there are duces the canonical approach to semantic information no meaning vehicles, how can we talk about meaning? and points out some diculties related to the gener- Meaning of what? ality of the approach. Section 3 oers an alternative Naturally, semantic information is interesting to an approach suggesting that the most general kind of in- informee because it can be useful to satisfy a goal. Hav- formation is pragmatic information and that semantic ing the information that a tiger is hiding in the bush information must be investigated within the framework changes the behavior of the informee to avoid or hunt of a pragmatic theory of information. Section 4 intro- the tiger. This aspect of information is the problem duces the notion of information system and explains of the pragmatics of information. It is widely acknowl- why it oers a non-circular basis for dening semantic edged that pragmatics places important constraints on information. Up to this point the paper develops the semantics and syntax.(Bar-Hillel 1964; Barwise & Selig- conceptual footing of the project. The following sec- man 1997; Dretske 1981; Floridi 2010) Nevertheless, tions address the specic problem of DSI. Section 5 one further, third, assumption of the canonical view outlines the strategy for analyzing information systems, of semantic information is that one can provide an ac- and thus semantic information, as a dynamical system count of semantic information independently, and prior phenomenon information systems are a special class to providing an account of how information is used. 3 Thus, one implicit feature of the canonical view of 1 gorithmic theory of information , or the Barwise & Selig- semantic information is the conceptual priority of data man's theory of information ow, etc., one begins with (or syntax) over meaning (or semantics), and of mean- the assumption that the data set is given in advance. A ing over use (or pragmatics). This idea suggests how to language comes with a xed alphabet (and a probability interpret the + in the composite expression data + distribution); a set of numeric sequences is well-dened meaning (+ truthfulness) + use. The + is regarded (in algorithmic information theory); or, one can dene a as an amendment operation. Therefore, I will call the network of classications with a xed set of tokens and view where data is conceptually primary, meaning sec- types (in Barwise and Seligman's theory). In all these ondary, and use third-ary, an amendment view of se- cases one does not provide an account of how the data mantic information. The amendment view goes back system is dened, except formally; one simply takes for to Shannon and Weaver, who quite explicitly acknowl- granted that it is. edge the three levels of problems of information, only But, when can the data set be specied? Formal me- to isolate the syntactic level, which is the focus of dia, where the data structures can be manipulated in- Shannon's theory.(Shannon 1948; Weaver & Shannon dependently of meaning or use, can be dened provided 1963) Carnap and Bar-Hillel (1952) also quite explic- underlying stable structures can be created, and mech- itly make the same acknowledgment, only to focus on anisms for reproduction or transformation of the struc- semantic information. And so do Dretske (1981); Fet- tures can be oered. Purely symbolic media depend on: zer (2004); Floridi (2003, 2010, 2004, 2005, 2007), and the stability of the ink on paper, the ability of the hu- many others, see (Floridi 2010: 3.4) for extensive refer- man cognitive system to recognize reliably the symbols ences. and how they enter in expressions, the encoding conven- tion of ASCII systems and the ability of devices to copy, In many respects, adopting an amendment view is transmit, and convert the codes, including converting irrelevant to the structure of a formal theory of semantic them to forms readable by a human, etc. Mixed media, information. One is interested in describing the struc- like maps or diagrams, have similar requirements. Gen- ture of the data, the nature of meaning, etc. Which is eral digital data systems depend on carefully crafted conceptually prior is lost in the nal product. This is physical devices, such as RAM cells, CPUs, CDs, Hard important because it implies that even if one questions Drives, etc. All of these devices are crafted to maintain the amendment view, as I will do in a moment, one reliable states, to interface with other devices, and to does not thereby aect the formal theories of syntac- transmit digital signals correctly. That is, the devices tic or semantic information. However, the amendment meet engineering specications in virtue of which they view aects meta-theoretical judgments. count as data and information media. We can safely say that all common information media with clearly One important meta-theoretical judgment aected speciable data sets are media that result ultimately of by the amendment view is when one theory of informa- human construction or interpretation. tion is more general than another theory. Within the How about natural information media? Are they amendment view it is natural to obtain a more gen- not independent of human construction or interpreta- eral notion of information by obtaining a more general tion? We can regard the tree rings, it is claimed, as a notion of data. Thus, one may start with a paradigm ex- data system that contains semantic information about ample of an information medium, (e.g.) language, and the age of the tree. Note, however, that isolating the relax some of its characteristics. One may move from rings of a tree as a datum requires ignoring some varia- language, to map, to continuous signal systems, to ab- tions of the tree and focusing on others. The same tree stract category theoretic system of classications and could have had its tree rings slightly dierently, yet of info-morphisms.(Barwise & Seligman 1997) The more the same number and same average thickness. In such abstract and general the theory, the weaker the con- a case, the dierent rings would count as the same data straints from semantics or pragmatics. type (similar to the way the letter `a' can be written with dierent fonts). Specication of the data system When is an amendment view appropriate? Obvi- requires specication of what variations are signicant ously, only when the structure of the data can be speci- and what must be ignored. This idea is captured by the ed independently. In many cases it can. In formal sys- notion of level of abstraction. (Floridi & Sanders 2004; tems trivially it can. This is partly what makes them Floridi 2008a). Specication of a data system always formal. Thus, in mathematical theories of information, such as Shannon's probabilistic theory of communica- 1 For a technical introduction see (Li & Vitanyi 1997), for a tion, the related Carnap and Bar-Hillel theory of se- philosophical introduction see (Adriaans 2008; Grunwald & Vi- mantic information, Kolmogorov/Chaitin/Solomono al- tanyi 2008) 4 requires a specication of a level of abstraction. In the interaction. The meaning vehicle obtains its semantic case of the tree rings, there are nomic dependencies that characteristics in virtue of the intermediary role that support the semantic relation between the rings and it plays in the interaction. Here I have purposefully the age, but the specication of this level of abstrac- avoided some of the technical language used by prag- tion as opposed to another requires 2 the interpretation matist, going back to Peirce, to describe the situation act of a cognitive agent. Nature, in its nomic patterns, and to provide theories of the sign utilization process oers many opportunities for data systems that can be often described as semiosis, following Peirce (1940, given semantic signicance, it oers ubiquitous poten- based on work from 1890s) and later Morris (1938). 3 tial datums, but it does not oer any well-dened and For us the important point is the strategy and its con- complete data sets. The tree rings, as a data system, sequences. are not constructed by human cognition, but they are One contrast between the pragmatic approach and interpreted by the human cognition. the canonical/amendment views is what systems are re- It is reasonable to conjecture that the cases where garded as more general. For the pragmatic approach the data set can be specied independently of meaning the most general scenario is not one that has the most and use are cases where a cognitive system is involved abstract notion of data where the constraints from indispensably in the process of specication. This is semantics and pragmatics are weakest but one where not to say that the informational medium must be a all dimensions (or parameters) of the scenario are in- part of a cognitive system, or that investigating its in- cluded. The cases where the data system can be isolated formational properties requires bringing in cognition. and regarded independently of its role in the semiotic But, if the conjecture is correct, it would follow that no process are special cases where an abstraction of the informational medium appropriate for an amendment pragmatic and then semantic parameters of the situa- analysis could exist without a cognitive system in the tion is possible. Therefore, the paradigm examples of background. Now, this is only a conjecture I have semantic information media examples such as lan- oered only an inductive argument but it is su- guages or road maps are special, degenerate cases of cient to support a methodological prescription: If we the general pragmatic scenario. Being degenerate cases, want to provide a notion of semantic information that 4 they are atypical , and therefore lessons derived from is pre-cognitive, we should reconsider the priority of the them should not be regarded as prima facie generaliz- components of semantic information. able. Instead of starting from within and working outward The classical pragmatic tradition and its more tech- by regarding the data set as conceptually primitive nical ospring, classical semiotics, made similar assump- we can start from without and work inward. We tions about the availability of a mind (or cognition), as can start with pragmatism and let the components of the classical analytic tradition motivating the amend- semantic information be co-determined. I call this the ment approach. Pierce, for example, readily assumed decomposition approach, where the notion of semantic that the user of the signs is a thinking being, and he information is dened rst and then its components regarded ideas as a sui generis notion in the theory. are decomposed from it as aspects of the informational For him, the mind qua mental, was an irreducible and process. an essential element of semiosis. The theory was not intended to provide a basis for the study of mind itself, at least not directly. 3 The pragmatic approach to semantic In the second part of the 20th century, with the information emergence of system theory and one of its ospring, cybernetics (Wiener 1965; Turchin 1990), a more ab- Pragmatism in general, and the pragmatic approach stract, mind-independent version of the pragmatic pro- to semantics in particular, is fundamentally a theory gram was attempted. Within this tradition it became about what the general scenario is where meaning as- possible to generalize the semiotic notions of sign and signment occurs. The general scenario is one where a sign interpretation with the help of the notion of infor- user utilizes some meaning vehicle system, a medium, 3 There are important dierences between Peirce and Morris in its interaction with another system. In the scenario as to the notion of semiotic, and the structure and purpose for the meaning, which could be a classical representation a general theory signs. Most of what nowadays is called semiotics relation, is determined by the nature of the interaction is aected more by Morris' (and Saussure's). (Burch 2010) and the mode in which the user regards the medium 4 They are atypical, even though they are ubiquitous and the signicance of the medium for the user in the highly salient in human experience. Similarly, an atmosphere high in oxygen is highly atypical as far as planetary atmospheres go, 2 The requirement is not constitutive, of course, but causal. even thought for us it is the stereotypical atmosphere. 5 mation. It also became possible to ask whether and how Simon 1981) Then, to be a data set is to be a symbolic information can be used as a principle of cognition. expression. To have semantic content is to refer to an- The most systematic attempt to understand infor- other expression or a symbolic rule governed process. mation within the pragmatic/semiotic tradition was made For s to interact with its environment is for s to have by Deode Nauta, in his The Meaning of Information.(1970) some input/output behavior. This example shows that In it, Nauta argues that information is fundamentally a the pragmatic strategy is suciently general to capture semiotic notion, and that the most general phenomena certain familiar conceptions of information processing. where the notion of information is appropriate are prag- However, if this is all there is to being an information matic. That is, in the most general case, the notions of system, the pragmatic strategy is worse o because it information and information content make sense only brings unnecessary complications. One nice feature of when there exists a user of the information and the in- symbol systems is that they can be described formally, formational mechanisms are a part of the control mech- only at the level of data relations. There is no point anisms of the user. to bringing the user system because it gets abstracted The strategy of pragmatic analysis of information away in the analysis. is the following: The most basic notion is information The pragmatic strategy oers the hope of produc- system (or i-system in Nauta's terminology). An infor- ing a condition C that is pre-cognitive, naturalizable, mation system s is a physical/dynamical system that is and that claries how information systems may emerge. in active interaction with an external environment and If this is possible, one can use the strategy to provide that satises a set of conditions C. (We will discuss a naturalistic account of informational phenomena in what C could be below.) The important requirement general. The idea is that it is easier to provide a bot- is that the conditions C do not presuppose the notion tom up naturalistic theory of information systems than of information. Instead, C must be some set of system to provide a naturalistic theory of information directly. conditions on s related to the dynamical or physical or- This still does not imply that the amendment view is in- ganization of s and on the mode of interaction of s with appropriate in the general case of information systems. the environment. C must guarantee the existence in s The failure of the amendment view is exposed by the of a sub-system, M , that can be interpreted as an in- actual proposed solution that fullls the hope. To this formation medium. Moreover, the functional role of M I turn next. in s in relation to the interaction with the environment, must be sucient to dene the semantic content of the states of M. 4 Information systems According to this strategy, s is an information sys- tem not because it operates with meaningful (and truth- Nauta suggests a denition of information systems (i- ful) data, i.e. because it operates with information, but systems). The denition is inuenced by ideas form conversely, it operates with information because it is an Acko (Acko 1958) and MacKay (MacKay 1969a). A information system. The most important idea is that system s is an i-system when it satises the following what counts as data and what gives the data semantic conditions: content is determined by the role it plays in the infor- mation system. 1. s is an open system, i.e. it is a system that is distinct from its environment, but it is in constant interac- As stated, all we have is a strategy. There is noth- tion with the environment . ing to demand that the pragmatic approach has some advantages over the classical view. How the strategy 2. s is a partially isolated open system, i.e. some of performs depends on the conditions C, and on how the the interactions between s and the environment are structured through well-dened limited channels of conditions allow a denition of M and a denition of inuence. Two kinds of channels are signicant. One semantic content. If C is such that one can always de- ne the data set rst, and only then the meaning of the of physical symbol systems. In the early days of cognitive science data, then the strategy reduces to the amendment view. it was common to equate intelligence with cognition. Nowadays, If C includes some cognitive or mental requirements, it is mostly recognized that physical symbol systems are not nec- essary for simpler forms of cognition. Here I am suggesting some- then the notion of information cannot be pre-cognitive. thing stronger, however. I am suggesting that simpler forms of C , for example, could include the condition that s satis- cognition are required for physical symbol systems to exist. This 5 es the physical symbol system hypothesis. (Newell & is either because cognition is required for creating physical symbol systems, in the case of articial systems; or, because a physical 5 Newell and Simon oered the physical symbol system hy- symbol system may emerge naturally only within a system that pothesis as a hypothesis about the nature of intelligence: that contains simpler cognitive capacities already. I will not argue for the collection of intelligent systems is included in the collection this claim here. 6 kind is the channels of receptors, R, which trans- arbitrary sub-system a cohesive, isolatable system are mit inuences from the external environment to complex to provide and are not important for our pur- the system. Another kind is the channels of emit- pose. The important idea is that there are parameters ters (or eectors ), 6 E, which transmit inuences that apply exclusively to the system; they allow us to from the system to the environment. talk about the states of the system as distinct from (but 3. s is a purposeful system. That is, there is at least one not independent of ) states of the world. Moreover, the proper set of states, G, that the system attempts parameters allow us to describe aspects of the dynam- to be in (or near) by aecting its environment. The ics of the system with minimal inuence from external relation between the behavior of s and G provides parameters. In a cell, many things happen inside that the system with normative signicance. are barely noticeable from the outside, while simulta- 4. s contains a sub-system M that can correlate with neously, many things happen outside whose eects on an external system S (via R). M can aect the the inside are shielded. states of E. M can be interpreted as a model or The condition of partial isolation further constrains a map. the class of systems by insisting that some of the in- 5. s contains a second distinct sub-system (or mecha- teractions between the system and the environment are nism) P that lters the states of M and their eect highly structured. This allows a further state decompo- on E to satisfy its purpose. In other words, P steers sition of the system. Signicant aspects of the behav- the system towards G by modulating the eect of ior of the system are determined by the states of the M on E .7 eectors. The set of possible states of the eectors is assumed to be much smaller than the possible states of Let us analyze the conditions of the denition. We will the system or the possible interaction relations between do this incrementally to see what each of the condi- the system and the environment. (Otherwise, there will tions adds. It is helpful to divide the conditions into be no point of isolating the eectors.)The reduction of two groups. Conditions 13 can be regarded as condi- states aecting the behavior makes it theoretically use- tions for an autonomous system. These are conditions ful to describe the system as modulating its behavior. of the system as a whole. The conditions identify the The system is seen as the source of the modulation. systems for which the architectural qualication, infor- mation system, becomes relevant. Conditions 4 and 5 When the receptors are included as another state target the internal organization of the autonomous sys- reduction of the interaction, this time for the inuence tem that make it an information system the actual from the environment to the system, the behavior can architecture of the system. be modulated by the control relations between the re- First I will focus on the conditions of autonomy. ceptors and the eectors. Signicant aspects of the be- The minimal condition here is the open system condi- havior of the system can be described by tracking how tion. A system must be isolated for theoretical focus. the state space of receptors is related to the state space It must be possible to identify a suciently cohesive, of the eectors. The relation need not be simple. The temporally extended sub-system, s, of the world such control relations are modulated by the internal states of that description of s in terms of variables restricted to the system, including by eects not channeled through s can be a basis for an eective description of aspects the receptors. However, it becomes possible to focus of the behavior of s. For example, s may be a rock. on the internal dynamics of control relations. A small The structural links of the molecules of the rock make causal/dynamical pathway between the receptors and it appropriate to describe the rock in terms of aggre- eectors acquires high relevance for the behavior of the gate variables such as mass, shape, temperature, hard- system. Systematic dierences in this pathway are ele- ness, etc. When the rock is rolled from a cli, most of vated to systematic dierences in behavior. We have a its behavior can be described with such variables and phenomenon that we can describe as high local rele- their interactions with external environmental parame- vance density. ters. The eect of the internal molecules is, in a sense, The last condition of autonomy is the condition of ltered by the system's variables. Similar approach can purposeful system. One necessary requirement to de- be adopted to other systems, such as biological cells, an- scribe a system as purposeful is the identication of a imals, robots, etc. General conditions for what makes an set of goal states G and some metric specifying how far the system is from G. A second necessary requirement is 6 Nauta uses the term emitter. I prefer the term eector. for the system to have some tendency to move towards Emitter has the connotation of something being emitted, while eector conveys the idea of a general causal eect. G or at least to put an eort to move towards G. The 7 We assume that both M and P are non-trivial systems. That dynamics of the eectors must be such that it creates a is, M and P play an active role in the dynamics of s. tendency to minimize the distance to G, even if exter- 7 nal factors completely overwhelm the eort. A sh can account of information is not about ends (goals) but swim towards food, even if the current is faster than about means. Even though information systems, ac- the sh, and it actually gets farther from the food. Be- cording to Nauta's denition, have functional teleology, cause the condition of purposefulness does not depend the teleological function is not the most important ele- only on the system but also on the specication of the ment of the story. As we shall see, it is not what facili- set G, it is really a relational condition. Every system tates the information semantics. may be regarded, trivially, as purposeful with respect The remaining, most important conditions of the to some set of states just make G to be the entire denition address the question of means. The body of state space. Or, if the system has some attractors in its the denition is the requirement of a sub-system M state space, the set of attractors can make the system that is correlated with some other system S. M is the into a purposeful system. intended system of information vehicles information Saying that a system is purposeful is interesting if medium. It is the system within s whose state space is there are independent reasons to identify the set G. In used for computing the various information measures some cases such reasons exist. In biological systems nat- according to (e.g.) Shannon's mathematical theory of ural goal sets are the set of states where the system is information. The pragmatic approach wants more out in good health, or the set of states where the system is of M however. M must play a control role in s. The likely to produce t ospring. More specic goals may states of M must also be correlated with the states be states where the system can have access to nutrients of the eectors, E. The correlation between S and M and energy, where it can be protected from predators, transfers via the correlation between M and to pro- E etc. What makes such states special may depend on the vide a correlation betweenS and E . Stated in a dierent system's organization. The organization of aerobic bac- way, the correlation between S and E decomposes into teria is such that they can extract energy from oxygen. two correlations via the mediating role of M . It lters Oxygen has a normative signicance for such bacteria, the interactions as to facilitate a channel of inuence it is a good. (Maturana & Varela 1980; Weber & Varela from S to E, and thus to modulate the behavior of the 2002; Di Paolo 2005) If the bacteria can modulate its system. This is the sense in which M has a control role eectors to move towards an oxygen rich environment, for s. Nauta's story is a bit more complex here, because we can regard the set of oxygen rich states as the goal there is also the intermediary system of receptors, R, states of the bacteria its G and we can regard which presumably mediate between M and S . While the bacteria as a purposeful system with respect to G. for modeling sensing systems R is important, for con- For articial systems G can be determined by the de- ceptual understanding of the role of M, R is not as signer. For the designer of an active seeking missile, the important it even need not be distinct from M . goal states are the states where the missile destroys an The denition of an information system could have enemy. stopped here. All the components for describing var- The idea of autonomy can be summarized as fol- ious binary information relations are already present. lows: the system can selectively control its behavior so The non-pragmatic approach may even be forced to that it can (attempt) to achieve a goal it can engage stop here because what happens after M, how M is in goal-directed behavior. The central idea is that the used in s, is not considered dening for a notion of se- locus of control is in the system itself, and the control mantic information. The canonical approach attempts is purposeful. For i-systems, the locus is determined by to explicate the relation between M and S and to base the higher local relevance density of the receptor to ef- the semantics on this relation. This is for example what fector pathway, which is part of the system. The goal Dretske does with the notion of indicating. (Dretske seeking behavior is determined by the functional role of 1981) This is also how the situation theory group at CSLI approaches the problem. 8 The immediate con- the eector states with respect to the G states. cern that follows is that very simple systems get to be This is a minimal notion of autonomy. It is not as- information systems. A thermostat-furnace system can sumed that the system can represent or internalize the easily be described as a partially isolated open purpose- goal. It however focuses the role of the remaining con- dition of the denition of information system. It sets ful system with a map M that mediates its behavior. a problem in a need of a solution. Namely, by what The bi-metal strip is M , it correlates with the tempera- ture of the environment (via the temperature of the air means does the system maintain the control relation- ship between inputs, outputs and goal? The remaining around the strip, which can be interpreted as R) and it turns on or o the switch of the furnace. We can call this conditions provide a specic design strategy for this problem: by the means of informational mechanisms. 8 See in particular (Barwise & Perry 1983) as well as the many This is important because it means that the pragmatic contributions in (Aczel et al. 1993; Cooper et al. 1990, 1991) 8 the thermostat problem : if a system as simple and silly Why should i-systems be information systems? We as a thermostat can count as an information system, should be careful how we manage our intuitions here. i- then there is something wrong with the notion of infor- systems are quite general, and we should not expect to mation system. One can, of course, bite the bullet and meet the familiar information media immediately. As I admit that the notion of information is that wide. This insisted above, i-systems are pre-cognitive, while the fa- is the strategy that Barwise, Perry, Israel, Seligman, miliar information media require cognitive machinery. and others take.(Barwise & Perry 1983; Israel & Perry Instead, we should focus on the theoretical merits of 1990; Seligmen 1991; Barwise & Seligman 1997) Or, one describing i-systems as (semantic) information systems. can demand some further quality of M that makes the i-systems should be the minimal systems that are use- states of M information. One can, for example, add that fully modeled as systems receiving information from the the states of M are intentional. The problem with this world, using the information, qua semantic information, is that it makes the project of naturalizing information to guide their behavior. It should be accepted that alter- systems that much harder. Even pragmatists are not native, non-informational models may be available immuned to relying on sui generis mentalist notions. such as dynamical system models or mechanism/causal Peirce was aware of the thermostat problem. He was models and for some simple i-systems such alterna- unhappy with the possibility that sunowers may be re- tive models may be equally eective in compressing the garded as using signs.(Peirce 1940) His solution was to description of the behavior of the systems. insist that the interpretant of the sign must be a mind. The question is, why should M be regarded as an From a naturalistic standpoint this is unacceptable. 9 information medium? For some, all that is required for A pragmatist approach allows happenings after M semantic information is maintaining sucient correla- to determine its informational status. The proposal re- tion between the situations at M and the situations at quires a further systemP , a purposeful lter, that medi- S . Such correlations are widespread, provided the world ates the connection between M and E . What dierence contains sucient patterns and regularities. Why S and does P make? It has two important eects: (1) it de- why those correlations? To x S and the appropriate couples M from E , and (2) it gives M signicance for correlations, we need to see what distinctions make a s. The mechanism that implements P modulates the dierence, i.e. what correlations are signicant for con- interaction between M and E by controlling how the trolling the behavior. But, what are the dierences of states of M aect the states of E so that the states of the behavior that are signicant? Those are the dier- E contribute to (or work towards) moving the system ences that matter. Here the goal states and the goal- closer to G. P contributes additional variables to the directed modulation by P are important. It determines interaction so that the states of M only conditionally the relevant distinctions which propagate to M and the control the states of E . M is not, therefore, merely a correlation with S. This allows us to isolate the appro- causal link in the dynamical interaction, it can be in- priate relations that can be regard as informational. terpreted as a medium. In a description of the system s The decouplability of M from the control system is one can isolate the questions of what the state of M is also important. It makes it appropriate to analyze the and what the eect of the state of M is on the rest of system in two stages. First, we can ask: What informa- the system, particularly on E. M deserves independent tion does the system possess? Second, we can ask: In theoretical focus. the circumstances, how can the system use this infor- The fact that P is a purposeful lter i.e. it is sen- mation? Note that while one can force such a two step sitive to how close the system is to G and it modulates analysis to a thermostat, it is pointless because there the connection between M and E to minimize the dif- is no sense in which the thermostat can have the same ference allows the normative distinctions that arise information and do dierent things with it. It is more with respect to G to transfer to the states of M. The ecient to specify an appropriate equation between the states of M can be evaluated as to their relevance and temperature and the state of the switch than to describe signicance for moving towards the goal. 10 the system as utilizing information. 9 Peirce would have disagreed that the account is not natu- ralistic. This, however, is related to his somewhat obscure meta- 5 Dynamical Semantic Information physics, which had pan-physicist elements. (Burch 2010) 10 It is not assumed that P operates by representing the value Over-the-counter modules, such as bi-metal strips (in of the current state of M, and uses the value to select an action. a thermostat) , photo cells, CPUs, etc., make it ap- It, of course, may operate like that, but such an operation would likely involve cognitive machinery. In the same way, P should not specied my M. This would be to regard P as a desire. P is be assumed to represent the goal state and evaluate the action not supposed to be a straight forward generalization of a desire in light of which one would archive the goal under the condition architecture. 9 pear deceptively simple to decompose an articial (po- initial (or current) conditions. Especially interesting are tential) information system into functional units. Its the stability properties of the system special trajec- gadgets come with well-dened interface connections, tories, including xed points, where the system repeats with xed informationally relevant states, with deter- the same states. It turns out that much of the dynami- mined ways they can interact together. Note, for ex- cal behavior of the system depends on how the system ample, that when discussing whether a thermostat is moves around these special trajectories or points, and an informational system, it is never dicult to identify how they are distributed throughout the space. This is what M is the bi-metal switch what features of done in the branch of dynamical systems theory called the environment M tracks temperature what its stability theory. Important ideas here include: (dierent function is to turn on and o the furnace. Similarly kinds of ) attractors, including the strange attractors with digital computing systems the input-output seen in chaotic systems, basins of stability, bifurcation, states, processing states, and computational operations chaos, etc. (Hirsch et al. 2004; Jianbo Gao & Hu 2007; are all well-dened. Such systems are designed so func- Ivancevic & Ivancevic 2008) tional decomposition is easy in itself not an easy In a DSM, one often assumes a large all encom- task. Nature, however, does not come so transparently passing phase space, global space, but system properties well-structured. It is not easy to determine whether the are described with additional phase spaces related to biochemical network of processes in a bacterium imple- the global space. For example, one may consider only ments an information system, or how (and whether) the a subset of the independent parameters and describe chemical and electrical processes in a brain implement the behavior of the system in a projection to a sub- an information system. space spanned by the subset of the parameters (or to This is the problem of information from dynamics: a lower dimensional manifold in the space). It could be How and what organization of the underlying dynam- that the interesting behavior of the system is invari- ics of the world support the existence of information ant with respect to some of the parameters, say global systems? An account of dynamical semantic informa- position. In this case, by projecting away the invariant tion is an account of the conditions for the emergence parameters, one focuses on the signicant parameters. of information systems within dynamical systems, and The additional spaces need not be embedded in the what provides the information states with meaning. 11 large space. They could be related to the parameters in The word dynamical indicates the requirement that more complex ways. A common simple example is when the world is viewed as a dynamical system, described the system exhibits a periodic behavior with respect to with the machinery of dynamical system theory.(Katok some of the parameters, in which case it may be better & Hasselblatt 1996; Hinrichsen & Pritchard 2005) We to consider a space where only the phase of the period call this a dynamical system model (DSM) of a system: is represented. The world at a given time is described as a point in An important question in the dynamical description a phase space 12 , the space of all possible states of the of a system is: What is the smallest (or smaller) set of world. The phase space is usually decomposed into a independent parameters sucient to describe important set of (independent) parameters (a vector base for the aspects of the behavior of a system? We may interpret space). The temporal change of the world is described such a question metaphysically as asking what the es- as a trajectory through the phase space. sential qualities of a system are that determine its be- The dynamical system theoretic analysis of a system havior. A less metaphysically loaded interpretation is is essentially counterfactual. One is not only interested to regard the small set of independent parameters and in describing the trajectory of the current state of the the equations describing the system's dynamics in the world. One is interested in describing the dynamical new space as an eective compression of the descrip- ow of the entire phase space the trajectories of the tion of the system's behavior. The possibility of such world from all possible states. One pays special atten- an eective compression can be interpreted as a dis- tion on how the behavior of a system, depicted as a tra- covery of a real pattern in the system.(Dennett 1991; jectory in the state space, changes with variations of the Ladyman et al. 2007) When a system is described with 11 a smaller dimensional phase space, and when there is This is dierent from (though related to) the question of the information dynamics that some (Williams & Beer (Forth- a function mapping the states from the large space to coming)) investigate, which is how information propagates and the new space we say that there is a parameter reduc- changes within a computational system. tion of the system.(Haken 2000; Ivancevic & Ivancevic 12 Often the phase space is called also `state space'. For example, 2008) Each parameter reduction denes a partition on in quantum mechanics, the Hilbert space of quantum state is called state space. Or, in computer science, the space of possible the large space where whole sets of state of the large computation states is called a state space. system are regarded as equivalent with respect to the 10 reduced system. Such sets of states can be regarded as the order parameters, and the parameters that track macro-states of the system, while in this terminology, external inuences control parameters. 14 the original states in the global system are regarded as As a special case of an open system interacting with micro-states. an environment we can consider an open system inter- In many cases parameter reduction may be appro- acting with another open system. The interaction can priate only for a region of the phase space where some- be modeled by additional dynamical equations connect- thing of interest happens. Especially interesting cases ing the order parameters of the two systems. The two are when there is some abrupt change in the nature of interacting systems are viewed as a single coupled sys- the trajectory ow of the system. Such changes are de- tem, and the interaction equations are called coupled scribed as phase transitions. The classic examples are equations. Of course, the coupled system is itself just a the state of matter transitions that occur as tempera- sub-system of the global system, and the coupled dy- ture drops (or increases) that produce change from gas namics is just part of the global dynamics. to liquid to solid (or vice versa ). Such phase transitions Now that we have outlined some of the key ideas of are a more general phenomenon. An important classe the DSM model, we want to view information systems of phase transitions are when breaking of symmetry oc- as a special kind of sub-systems within a global dynam- curs and order increases in the system (as in the tran- ical system. The goal will not be to provide a general sition from a liquid to a solid crystal). Regions where theory of the emergence and dynamics of information such symmetry breaking happens are especially suscep- systems with DSM. This may still be too hard for the tible to parameter reductions.(Sethna 2009; Ivancevic current state of the art of the mathematics of dynamical & Ivancevic 2008) The new set of reduced parameters systems. 15 The goal is to conceptualize information sys- is often described as order parameters. (Haken 2000; tems as dynamical systems to be able to view semantic Ivancevic & Ivancevic 2008) information as a dynamical system phenomenon. We Identifying regions of phase transition where param- want to focus on what conditions must be met to an- eter reduction is possible can be used to identify (or de- alyze a system as an information system, described in ne) cohesive sub-systems in the global system. If the terms of informational states, semantic relations, and global system is a chemical reaction system in a xed utilization of information. spatio-temporal region, the formation of a membrane I will take some shortcuts. I will assume that there wall, the emergence of an auto-catalytic cycle or of an exists a dynamical description of the sub-system s. s autopoietic system, etc., can be identied (in principle) has a well dened phase space in terms of order param- by identifying a region on the phase space bounded by eters. The region of viability V within the global state phase transitions, with order parameters tracking the space is dened so that it is clear under what conditions organizational dynamics of the sub-system. The phase s exists as a system. A lot of the internal structure of s transition boundary is the life/dead boundary for the may also be represented in a DSM. An important goal system, the internal region is the viability zone, and the of the dynamical analysis of s is understanding what order parameters can be the parameters describing the subset of V has the property that, for a signicant pe- internal workings 13 of the system. Sub-systems can also be nested. In the general case 14 It is important to clear a potential terminological confusion one uses multiple interconnected phase spaces, dened here. The term `control parameter' is internal to Haken's the- for dierent overlapping regions of the global space. In ory of synergetics.(Haken 1993b,a) Haken calls external inuence parameters `control parameters' because they are often used to this way one can model hierarchically organized sys- control the behavior of self-organized systems. I also talk about tems, and dierent levels of functional organization. control in a less technical sense: I say that a system controls its The global system is regarded as a closed system; behavior, or that the locus of control lies in a system. In this more general notion of control, the control relations may depend when dealing with sub-systems, however, it is impor- solely on the order parameters (in Haken's sense) of the system. tant to model the interactions between the sub-systems Also, some control parameters (in Haken's sense) may not have and their environment. From now on I will drop the any control signicance in my sense. The terminology is unfortu- `sub-' prex and I will assume that we are modeling an nate, but I stick to it to be consistent with the literature. Thus, when the expression `control parameter' is used, it is always in open sub-system, consistent with the discussion in the the technical sense of synergetics. Any other expression that has previous sections. Thus, for a system s one separates the word `control' is used in my (or a control system theoretic (Levine 1996)) sense. 15 There is a large recent literature attempting to analyze infor- 13 Often the economical order parameters do not have obvious mational and cognitive system with the machinery of dynamical interpretation as degrees of freedom of internal mechanisms of systems theory. (Thelen & Smith 1994; Kelso 1995; van Gelder. the sub-system. 1998; Beer 2000; Chemero 2009) 11 riod of time, the system will remain within V .16 In other When examining the loci of control one examines words, we are interested in the behavior of s from the the stability of the dynamical ow of the system in the point of view of its short term survival. 17 The decom- various phase spaces used to describe it, and the way the position of the dynamics of s in terms of order and con- stable ow changes with variation of particular parame- trol parameters allows us to analyze the problem as an ters. The ow may change continuously, or it may jump interaction between the states of the system (order pa- discretely from one attractor to another. For example, rameters) and the states of the environment that have an automobile's movement is very sensitive to the po- relevance for the behavior of the system (control param- sition of the steering wheel. If one examines how the eters). (Some parameters of the global system may be trajectory of the vehicle changes as the initial position irrelevant for the dynamics of s. This may be expressed is varied (while the direction is kept xed 18 ) one would by saying that there are projections to subspaces (or observe a parallel ow. As the position of the wheel is manifolds) of the global space that preserve the inter- changed, the curvature of the ow changes systemati- esting behavior of s.) We can examine questions such cally. Note however that the position of the heater dial as: (1) given a xed state of the order parameters, how has no detectable eect on the trajectory. Note also will changing the control parameters aect the future that an explosion of a roadside bomb has a dramatic behavior of the system, both (1a) as described within eect on the trajectory, but variation on the nature of the global phase space (or a reduced parameter space the explosion does not preserve the stability of the ow that includes the coupling relation with other systems) the system is not controlled by the explosion even and (1b) as described in the phase space of the order though it is aected by it. 19 In a digital computer the parameters? In other words, how do the external inu- organization of the physical matter is such that xed ences on s aect the behavior of s in relation to other sub-systems (memory cells, registers, etc.) exhibit bi- systems and how they aects its internal dynamics? (2) stable behavior (driven by positive feedback) i.e. Given a xed state of the control parameters, how will the sub-system orbits one of two stable attractors. In changes on the order parameters aect the behavior of virtue of the xed organization of the CPU (a kind of the system? In other words, how do the peculiarities of xed constraint on the system order that does not the internal organization of the system aect what the change) the state of some bi-stable cells can shift (or system does under the same conditions? Investigating not) the state of other bi-stable states at a later time (1) and (2) together allows us to determine where con- (the next clock cycle). If the system has N such cells, it N trol structures lie. We can isolate to what extent the has 2 possible attractor states, and at each clock cycle system's behavior is regulated by the order parameters the machine moves from one attractor state to another. and their dynamics and to what extent by control pa- At each move some cells have control signicance, oth- rameters, and in what circumstances. We can observe ers do not. that the behavior is more sensitive to control parame- While it is possible to localize control signicance ters, in which case the system is more heteronomous, in many systems, and therefore to place the system in i.e. externally controlled. We can observe that its be- the spectrum between heteronomy and autonomy, such havior is more sensitive to order parameters, in which localizations are context dependent and vague. When a case the system is more autonomous or, if there is car is airborne the control signicance of the steering a complex, context sensitive interaction of inuences. wheel disappears. When the car is on ice the control (3) Given a xed state of the control parameters, how signicance of the steering wheel diminishes and other does changes in the order parameters aect the evolu- parameters, such as the position of the gas peddle in- tions of the control parameters? In other words, how do crease in signicance. 20 This is not a failure of the con- the internal operation of the system aects the external 18 Keeping the initial direction is needed for illustrative reasons environment, including the external environment that only. The phase space of the problem is not physical space, but may have an immediate eect back on the system? (4) an abstract space including the direction. There is no problem What are the feedback relations between the control characterizing the stability and variability of the ow with respect and order parameters? to the position of the wheel. 19 Of course, this is relative to the phase space of interest. If the system is parametrized with two states operative and inoper- ative then a bomb can be regarded as a locus of control. It 16 Here I am assuming a complex, non-linear dynamics without is a binary switch that moves a vehicle from both operative and simple stable regions. There is no trajectory that remains forever inoperative states to an inoperative state. in V. 20 On a low friction surface, such as ice, often the only possible 17 Essentially, here I am assuming that the goal of the system is way of steering corners fast is using the so called drift method, survival. Or, stated dierently, that the G states in the denition where the car slides sideways in the direction of the turn and of a purposeful system compose the region V. one controls the attitude by adjusting the throttle (gas peddle). 12 cept of control; it comes from the nature and diversity ertheless, in many natural systems, such as bio-chemical of dynamical systems. reaction systems, or neural networks of brains, the in- To capture the notion of information system within crementalist approach has not been very eective. The a DSM, one must localize control signicance to a sub- dynamical approach to information allows us (in prin- system M and second order control signicance, control ciple) to identify information systems by analyzing the over the control role of M, to a sub-system P. P, fur- dynamics of the systems the emerging control rela- thermore, must be evaluated in light of its ability to tions leading to purposeful behavior. We are still inter- maintain the system in the viability zone of the global ested in the informational decomposition of the system phase space. M must be correlated with a particular in identifying the media and the information rela- set of control parameters the source system S tions in which they enter, the data, etc. The language and have a conditional eect, modulated by P, on the of information systems is not the language of dynami- general trajectory of s (via E ). The structure and topol- cal systems. It is a language a conceptual framework ogy of the ow of s admits of such decomposition. There that compresses the patterns of interaction in a spe- must be an appropriate real pattern of the global dy- cic class of dynamical systems in a dierent way that namics of the system in the viability zone such that DSMs. one can determine appropriate macro-states (really, a hierarchy of macro-states) and a dynamics of the sys- 6 Interface Theory of Meaning tem respecting the macro-states so that a collection of order parameters exist that track the patterns. Saying I oer a theory of what semantic information is, that that a system is an information system is saying some- unlike the classical view, does not start from a notion thing about what patterns exist within the dynamics of meaning. To be semantic information is to be the that there is a highly structured localized, goal-directed currency of an information system. It is still legitimate control modulated by a medium M. to ask for a given information state (of M) m for a In the most general case of an information system, given datum What is the meaning of m? The de- the pattern of the dynamics spans the entire system composition approach demands that an answer to such and environment. Even if it is possible to identify the a question be provided, although it may not look like a system M independently as a dynamical sub-system of familiar answer. In this section I will answer this ques- s, it cannot be guaranteed that the states of M, let us tion in terms of what I call the interface theory of mean- call them the local micro-states 21 of M, are the states ing. First, however, I will examine two opposing ways relevant for the informational system. We have to fur- the question is addressed in a friendly crowd. I will use ther identify a collection of macro-states of M that cap- the discussion as a motivation to my proposal. ture the correct distinctions relevant for the information Traditionally, (foundational) theories of meaning/content, system. It is these macro-states that are interpreted as both for language and for mental states, have been di- informational states as data. More on this below. vided into two categories: externalist and internalist. Viewing information systems as a special kind of There are other interesting divisions, but most are rel- organized dynamical systems allow us to be liberated evant only for suciently complex media that require from some of the proto-mechanistic, incrementalist in- cognitive underpinning. The externalist/internalist di- tuitions that to understand the operation of a system vision, however is completely general, and can be made we must understand its parts independently, and then for any information medium. Roughly, the distinction is we must recover the system by specifying how the parts about where the primary constraint of the determina- t together. Such intuitions are, I think, at the root of tion of meaning for an information state derives from. the amendment approach to information. Undoubtedly, An externalist theory focuses on constraints outside of in many cases, especially for articial systems, such an the user of the informational state. Particularly, it fo- incrementalist approach is the correct one to take. Nev- cuses on the relation between the informational state and the sources or object of the information. The meat This is a very dicult and dangerous technique. Leave it for the of the semantic connection derives from some nomic (or professionals! Besides, most modern cars with front wheel drives teleonomic) connection between the source system and and electronic stability control cannot drift steer. 21 M , of course, is already dened by order parameters, which the information medium (receiver) system. The focus dene macro-state in the global phase space. Indeed, there may of semantics for an externalist theory is the determina- be several levels of such macro-states until the right invariance tion of the way the world is. Examples of such theories are identied that isolate the system M. Still, we can think of are Dretske's and those of the situated semantics group. the states of M as local micro-states in the immediate reduced phase space of M . Yet, further reductions are possible, and fur- Also, Millikan's and those of the teleosemantic commu- ther macro-states can be dened. nity.(Millikan 1987, 1995, 2006) I will focus below on an 13 account due to Bogdan (Bogdan 1988a, 1994), that is of the aspects (or features) of the source are relevant quite similar in spirit to mine. Bogdan's account diers for goal-directed behavior. (2) It determines internal, from other externalist in that it takes goal-directedness architectural functions for the system using the mate- to be a fundamental requirement for semantic informa- rial information to achieve the goal. It is not sucient tion. for the system to have material information from the An internalist theory, on the other hand, considers source of the aspects relevant for the goal. The system as the primary constraint of meaning what the informa- must be organized in a way material information tion state does for the user. The model of the internalist which is nothing more than a form of nomic correlation account is not reference xation and fact determination, can aect the system's goal-directed behavior. (3) but message interpretation. The question that an inter- It solves the proximal stimulus problem, i.e. it allows nalist asks is not whatm means, but what m means to a distinction between the true source of the information given user. Of course, for m to be informative about the and any proximal systems in the information pathway world, it better be suciently correlated with a source, from the source to the receptor that co-vary with the but this is not a constitutive condition of the meaning source. For example, a state of a visual system may con- of m. It is a condition of a good interpretation system. tain material information from a chair, but also it may In an internalist account, m can have a meaning for a contain material information from the retina. Only the user even if the user is, so to say, completely out of it. chair is relevant to the goal-directed activity of nding Below I will consider the internalist account of MacKay a place to rest. The retinal state has no rest-inducing (1969b; 1969c), who, as I indicated above, was a strong properties. inuence to Nauta. (1) and (2) provide a basis for a system to utilize Bogdan, like me, wants a notion of semantic in- semantic information to select (and convert) ma- formation to serve as a basis of understanding cogni- terial information to meaningful information. There is tion. He makes a distinction between material informa- similarity between Bogdan's conditions (1) and (2) and tion and semantic information. Material information in Nauta's conditions of i-system. It is condition (3), how- one receiver system from another source system results ever, that separates from from about. It xes the con- from the systematic, nomic relation between structures tent of the information. of the source and receiver. Such a notion of material The semantics of Bogdan's notion of semantic infor- information is fairly uncontroversial among people that mation, the answer to the question What is the mean- take the notion of information seriously. It is what some ing of an information state m?, is of the form: m is describe as environmental information (Floridi 2003; about a system s and it is in so-and-so state (or has so-and-so probability distribution). 22 The back-end of Barwise & Seligman 1997; Dretske 1981), or physical information, or potential-information (Nauta 1970). Se- the information process the functional business mantic information is a kind of material information focuses (or constrains) the front end the correlation where the from is converted to about it is when part but it is the correlation that determines the the receiver can be said to have information about the meaning. source. One way of understanding Bogdan's eort is as Now let us consider an internalist theory. MacKay's explicating in a naturalistic setup the notion of about- internalist account of meaning is aimed for the follow- ness. To this end, semantic information is characterized ing general situation. There is a system S that is capa- as follows: Semantic information is material informa- ble of goal-directed activity. S receives a message m. tion with a functional business determined by teleol- m could be a message sent by another system with a ogy. (Bogdan 1988a: 89) specic intention, or it could be a signal from the en- The key task is explicating the notion of functional vironment. What does m mean for S? He proposes the business determined by teleology. Bogdan's theory is following answer: complicated and it is not my goal to develop it here, nor is it to compare it to my use of goal-directedness [T]he meaning of a message can be dened very which is based on Nauta. There are many similarities, simply as its selective function on the range of and some apparent dierences in the two approaches, the recipient's states of conditional readiness for but, I must admit, it is still not completely clear to goal-directed activity. (MacKay 1969b: 24) me how deep the dierences go. Here are some ideas 22 Bogdan actually does not address this question in (Bogdan that cast light on how teleology converts material in- 1988a, 1994). He resists applying the term `meaning' to such formation into semantic information. It does this in at simple systems to avoid undesirable connotations. (Private cor- least three ways: (1) the goal acts as a lter of rele- responds) But it seems to follow from the discussion that this vance for aspects of the information source. Only some would be the form of the answer, if one must be given. 14 Consider the following example, adapted from MacKay: thought experiment (Putnam 1975) can be interpreted Steve is seating on an armchair reading a book. Some- as suggesting that a purely solipsistic internalist the- body enters and says It is raining. (This is m) Steve ory of mental content would not be able to distinguish does not make any dierential response to the state- between the dierent contents of water (twater) of be- ment. What has m done? It has not changed the be- lievers among Earth and Twin-Earth inhabitants or havior of Steve in any way (assuming that he con- so the intuition goes. Such hybrid theories of mean- tinues to attend to the book with full engagement). ing have targeted cognitive information media lan- However, when Steve stops reading the book, as he guages, mental states (beliefs), etc. This analysis of leaves the room he grabs the umbrella. The message meaning cannot easily transfer to the domain of dy- has induced in Steve the readiness to take the um- namical semantic information. brella, or to imagine the streets wet, or to be con- In the case of dynamical semantic information, the cerned about whether he brought the lawn mower to externalist and internalist conceptions of meaning col- the garage, etc. Steve (read his cognitive system ) is in a lapse into a single notion. The reason for this is the co- conditional state of readiness to respond dierentially determination of macro-state structure of informational to various states of his environment. The message it systems. Let us examine this in a bit more detail. As is raining changes these conditional states. Metaphor- indicated above, to claim that a given open dynamical ically speaking (MacKay's metaphor) the message ad- system is an information system is to identify a col- justs the switch-boxes of Steve's response function. It lection of sub-systems, a collection of macro-states for is this tendency of a message to select the switches of each sub-system, and a set of dynamical relations that the response function that determines what the mes- respect the macro states (plus various other things). Let sage means for Steve. The same message, qua physical us focus on the macro-states. For each sub-system, e.g. (syntactic) form, would adjust the switch-box of Peter S or M , we must determine what micro-states should be in slightly dierent ways. It, therefore, would have a dif- regarded as functionally equivalent with respect to the ferent meaning for Peter. The message would normally dynamics of the information system, qua information change the conditional state of readiness of both Steve system. In the case of M, this amounts to specifying and Peter in similar ways, allowing them to coordinate what the data states of the system are. In the case of their actions. If Steve calls Peter for a ride from work S, this amounts to specifying what the structure of the Peter would not be surprised, and Steve would expect source looks like form the prospective of the informa- Peter not to be surprised, etc. tion system? 23 It is possible that both S and M have According to MacKay, it does not make sense to ask independent macro-structure and the dynamical corre- about the meaning of m in isolation the notion of lation relevant for the information system matches this meaning makes sense only relative to a user. Meaning independent structure. S is what it is intrinsically, M is does not dependent only on the form of the message. what it is intrinsically, and they are simply connected Of course, for some classes of users, messages of partic- by some causal process that matches the properties of S ular forms change their conditional state of readiness with the properties of M and thus an information con- systematically. This may be so because of xed conven- nection is formed. This possibility is, in fact, the stan- tions or because the form of the message has a useful dard conception of the information process as physical correlation with the environment in which the goal di- phenomena. While such a dynamical scenario is possi- rected activity takes place. Communication in a com- ble, it is not necessary. mon language would be impossible otherwise. In the general case we cannot assume that the macro- structure relevant for the information system descrip- It is easy to take the externalist and internalist ap- tion is determined locally. We may need to look at the proaches to the question of the meaning of m as in- entire dynamical system the entire process of interac- compatible. They, after all, have very dierent form. tion between the system and the environment as the The externalist approach excludes the user from hav- basis of determination of the macroscopic structure. 24 ing a constitutive role for meaning. The internalist ap- proach makes meaning primarily user dependent. The 23 This may be the informational equivalent to (and the dy- approaches need not be incompatible however. One strat- namical basis of ) the phenomenological notion of umwelt (von egy for reconciling externalism and internalism is to Uexküll 1909, 1932, 1982). Nauta explicitly utilizes the notion in take a hybrid account of meaning/content. Such hybrid his analysis of information systems. 24 It is very temping to describe the problem with the language theories are motivated by an observation that exter- of supervenience. I recommend caution in using supervenience nal or internal considerations are not suciently ne here because the notion of supervenience has its roots in the clas- grained. Such hybrid views are especially important sical object/property metaphysics, while my discussion is based in discussion of mental content. Putnam's Twin-Earth on the dynamical system theory approach to system analysis. The 15 To determine whether a particular macro-state of S is evant for the semantic evaluation of a datum depend informationally relevant, i.e. whether it is dierentially on S alone. They depend on the dynamics of the entire signicant for the purposeful behavior of the system, we system. From the prospective of the information sys- must trace the dynamical trajectories of the system and tem, there is no independent objective reality that its determine (at least) two things: (1) whether the micro- informational media track. This does not preclude the state variation within the macro-states is insignicant possibility that an external observer can identify both for the purposeful behavior, i.e. the dynamical trajecto- independent (from the information system) real pat- ries in the appropriate reduced phase spaces that track terns (or properties) of S and real patterns emerging the viability parameters (or in general the parameters from the informational interaction. related to the goals states) are stable. Let us call such Nor can we assume that the informational states of macro-states informationally stable. (2) Whether other M be specied independently of S and the rest of the informationally stable macro-states produce a dieren- system. The appropriate macro-state structure of M tial dynamical response of the system in the same re- is, in general, under-specied by the correlation to S duced phase space. alone, or by the relation to the eectors E of the sys- Not all informationally relevant states of S need tem alone, or by the eective distinctions that P can make a dierence for the system qua information sys- make alone. Indeed, there can be structural relations tem. Only in the distinctions that make a dierence between S and M that could be useful for the goal di- (MacKay 1969a) for the internal control mechanisms rected behavior of the system but that never make it to mediated through M and modulated by P are impor- control service. Similarly, there can be many interesting tant. In other words, only some of the informationally connections between M and E that could be relevant relevant states are actually signicant for the system. 25 for behavior but the corresponding macro-states of M What states of S are signicant depends on the internal may not track anything interesting in the environment, organization of the M - P system. The metabolism of a etc. Again, in the general case, the entire dynamical bacterium may respond dierentially to many types of system/environment complex must be used to identify nutrients. This is signicant for the goal state of the sys- what macro-states of M count as the data states of the tem. Some nutrients may be better at maintaining the system. system deep into its viability zone; others may merely Of course, it does not follow that one always must slow down the exit. 26 Still, the control mechanisms that consider the entire complex. There can be sucient in- modulate the bacterial purpose-modulated response to ternal structure to S or M or P or E so that the signif- the nutrients (if they exist) may make only a small num- icant macro-states stand out on their own. Many arti- ber of distinctions. Even more the internal dynamics cial information systems may be designed with highly may force and utilize distinctions that, considering S as structured components that x the relevant macro-states an independent system, are not derived from the struc- from within. Evolutionary processes may also gener- ture of S .27 It follows then that in the general case, we ate highly modular systems where the relevant macro- cannot assume that the distinctions in S that are rel- states are xed from within. Still, a general theory of semantic information must admit information systems two approaches are not incompatible, although I think that the dynamical systems approach is more general. The notion of super- where all the relevant patterns of organization derive venience, as used by Kim for example, is not readily convertible from the entire complex no o the shelf parts. to the dynamical systems approach. But, if I must describe the It follows that neither an external relation between problem in terms of supervenience, I can describe it as follows: we cannot assume that the macro-properties of the sub-systems M and S, nor an internal function of selecting condi- supervene on the micro-properties of the same subsystem. They tional readiness states is sucient to provide a general may supervene on the entire environment/information system en- notion of meaning, for they don't even x the syntax of semble. the information system independently. To specify the 25 In fact, only a small subset of the informationally relevant meaning of a state m we must do something dierent. states of S would, in general, be signicant for the internal control pathway. This dierence a kind of informational deciency of What does M really do in the information system? the system is central for understating the role of cognition for It acts as an interface between the (external) world an organism. (Vakarelov forthcoming) and the control system. It structures inuences to allow 26 Eating spoiled food may help an organism not die of hunger focused purposeful control. If any sense of signicance now, but it may cause food poisoning that may harm the organ- ism later. can be given to a particular state m of M, it must 27 Some (Maturana & Varela 1980; Varela et al. 1992) have sug- be related to this interface function. The signicance gested that the system of color discrimination and categorization of m is neither that it tracks something external nor of many organisms only partially depict physical reectance (or other) properties of external objects. It depends to a large de- the organism imposes structure on the world that is not there gree on the internal dynamics of the visual system. In a sense, independently, but that is utilized by the system. 16 that it can aect the control mechanisms of the system, Most of the work in sections 4 and 5 was done to but that it can connect one to the other. The idea of assure us that nothing irreducibly semantic lurks in interface allows us to specify a notion of meaning for a the notion of information system viewed as a dynami- datum in an information system: cal phenomenon. 30 Therefore, (1) can be achieved. (2) is much more dicult. It requires the specication of Meaning: The meaning of an informational state m of an information system that utilizes language semanti- an information system is given by the dierential cally. This is a daunting task. Current states of psy- interface function it serves in the whole process of chology or computational linguistics are very far from purposeful interaction between the information sys- understanding how languages can emerge in natural bi- tem and the environment. ological agents or how to design articial agents that In what sense is this a denition of meaning? This is use semantic language. The kind of informational sys- clearly not a denition produced by conceptual analy- tems that may get anywhere close to being language sis of meaning. It is not intended to explicate or cap- users are vastly more complicated than the simple pre- ture typical meaning ascriptions. Almost exclusively, cognitive information systems discussed here. It follows meaning (or content) ascriptions involve languages or that providing a full justication of the interface func- language-like entities like beliefs. As noted above, within tion conception of meaning as a proper generalization the pragmatic approach of information such media should of stereotypical conceptions of meaning is beyond my not be seen as stereotypical. In fact, they should be seen current abilities. What I oer is a conjecture that the as atypical. I would regard a general account of mean- interface theory can contribute towards understanding ing that looks too much like an account of meaning for of semantic processes for language-using or other cog- language as suspicious. nitive information systems. I can provide however some My denition is foremost technical. Still, there has hints about how the notion may lead us towards more to be some connection between it and more common familiar notions of meaning. notions of meaning. Otherwise it makes a deceptive use Let us go back to the observation that the denition of terminology. The proper connection is that of gen- collapses the external and internal conception of mean- eralization. The way to evaluate whether the dened ing. Specifying the dierential interface function of a notion deserves to be called meaning is to satisfy the state requires looking at the entire system/environment following two requirements: (1) It must be a notion that can be dened within appropriate general framework. complex. We can think of the datum state m as partici- pating in a process 31 of interaction where causal eects (2) When the general framework is instantiated to the from the environment are channeled through the inter- stereotypical case(s), the notion must reduce to the nal M-P control pathway to produce actions, which ac- traditional notion. Thus, if the medium of an informa- tions modify the system's behavior, and which in turn tion system is a language, then the dierential interface changes the state of the environment (including the re- function related to the linguistic expressions must boil lations between the system and other external systems). down to something like a stereotypical notion of mean- This aects future causal eects and the way the sys- ing. It must be noted that an instantiation operation, as an inverse to a generalization operation, is one-to- 28 Generalization always looses complexity, so a 30 There is one important contention here. Isn't the notion of many. goal, and thus purposeful system already semantic? Such an ob- language-using information system may involve com- jection has been raised by Dretske (1988) in response to Bogdan, plexities that do not appear in the general case and and more generally by Floridi (2010). Careless use of goals can that may produce dierent notions of meaning depend- indeed sneak in semantics. The important thing is not to assume ing on how the complexities are xed. 29 that goals are explicit (like desires). Goals should not be regarded as kinds of propositional attitudes. My, and I believe Bogdan's, 28 Of course, the opposite is true too. There may be many ways notion of goal is not content determining. For Bogdan's reply to one can generalize a specic situation. Here I follow one specic Dretske see (Bogdan 1988b). In my case, the notion of purposeful generalization supported by the pragmatic approach. system is purely dynamical. It captures a particular patterns of 29 The theoretical method of generalization and re-instantiation interaction between a system and its environment. Such a pattern is a great tool for resolving disagreement between competing the- may be selected by an external designer, in which case Floridi's ories of something (e.g. of meaning). By obtaining a general the- zero semantic condition (Floridi & Taddeo 2005, 2007) is not sat- ory and then showing how specic but competing scenarios are ised, but it could result from (or be) a natural pattern in the instances of the generalization, one can demonstrate that the dis- global dynamics. As it has been argued by some (Maturana & agreement is not conceptual but results from a dierent xation Varela 1980; Varela 2000; Weber & Varela 2002), convincingly at of some theoretical parameters. It may turn out that both specic least to me, the phenomenon of life may be related to the natu- theories are correct but they are theories for dierent domains, ral emergence of purposeful systems. This, however is a separate and moreover, both are justied in using the same concept be- issue that I do not wish to discuss here. cause the concept turns out to be a specic instance of the general 31 Here I use the notion of process informally. It is assumed that concept. the system is ultimately describable with a DSM. 17 tem responds to them. The complex feedback process other information media. The properties of the interface can be decomposed into segments/sub-processes. It is relations look a lot like the properties that a correspon- possible to focus on the regularities that exist between dence semantics may have, but these interface relations the system M and the source S, or on the regularities do not capture the true interface roles of the language that exist in the way a datum aects the state of con- datums for the information system. To determine the ditional readiness of the system. In other words, it may true interface role, we need to link all local interfaces be possible to extract two interface sub-functions, one and see how the entire complex participates in the pur- related to the external correlation between the medium poseful behavior. In other words, the correspondence and the source, and another related to the selective con- relation underlying correspondence semantics, accord- trol function of the medium. It is not clear that the ing to my rough hypothetical pragmatic analysis, is a interface function is always completely determined by relation that exists only between carefully orchestrated the two sub-functions, but it is important that we can media, not a relation that exists between the datum recover the external and internal notions of meaning as and the world. However, such an inter-media informa- aspects of the process. tion relation can be elevated to important cognitive sig- nicance by the other media and their interfaces with The story gets more interesting when the structure perception and action. of M and P gets more complicated particularly, In this picture there is no need for some mysteri- when the system utilizes dierent sub-systems that act ous sui generis mentalist notion of intentionality (or as information media. The system may have media M1 , M2 . . . Mn Pierce's notion of interpretant) to support the seman- (and a collection of dierent purposeful lters), each tics; it is all a story of organized dynamical sub-systems with dierent roles and interface connections. Some me- of s channeling and controlling eect of external and in- dia may be connected to dierent external systems or ternal inuences on behavior. It is all complex patterns dierent aspects of the same systems, others may in- in the dynamics of the world, patterns of the ow of the terface with other media, yet others may be connected global phase space. with eectors or control the states of other media, etc. When the system is organized as a complex network of information media, complex interface (sub-)functions 7 SSI vs. WSI can emerge. Some can depend almost exclusively on external connections to outside sources, others can be I will end with a short discussion of how the notion analyzed entirely in terms of their control role or ef- of dynamical semantic information relates to the de- fects on other media. I conjecture that the canonical bate about whether truthfulness must be included as a examples of information media that shape many of our condition of semantic information. One position states intuition about semantics are media that exist (within that information is simply meaningful data. This is the an information system) as only one of a large network so called weakly semantic theory of information, (WSI). of other information media that jointly control the sys- Proponents of WSI are (Carnap & Bar-Hillel 1952; Fet- tem's behavior. Thus, to take correspondence theories 32 Another position insists that truthfulness zer 2004). of meaning as an example, it is tempting to say that must be included in the denition of information. This the word `chair' means a property of external objects. is the so called strongly semantic theory of informa- Thus, in the expression, This is a chair, the meaning tion, (SSI). Proponents include Dretske (1981); Barwise is given by some fact in the world that the object de- & Seligman (1997); Floridi (2010). The most system- picted by the indexical has the property of chairhood. atic defense of SSI can be found in Floridi (2004, 2007, In an information system using language we can ana- 2010). The gist of the debate is the following: imagine lyze this idea in a dierent way. The language medium, Steve receives a message It is raining outside. When whose datum may be some structural equivalent to the has Steve received a piece of information? A proponent expression This is a chair, interacts with other non- of WSI claims that Steve has received a piece of in- linguistic media connected to perception, allowing the formation regardless of whether it actually is raining system to identify and interact with patterns in the outside. A proponent of SSI claims that Steve receives world that can be clustered through some data state information only if the statement is true, i.e. only if it of some internal media. To make Fodor happy, we can is actually raining outside. Another way of formulat- assume that there is a single medium that gets in an in- ing the debate is whether mis-information (when it is formation state uniquely correlated with chairhood not raining) is a kind of information or a kind of non- a kind of a concept of chair. The language system, in this picture, is not interfaced with the world (or some 32 ... and most everyone working in the eld of IST. See (Floridi abstract realm of propositions). It is interfaced with 2010: 4.2) for many examples and references. 18 information (a kind of pseudo-information). For WSI in the system (its meaning), the informationally rele- mis-information is a kind of information. For SSI mis- vant distinctions in the source, and the available eector information is something dierent all together. (To use states (the possible informationally controlled actions), Dretske's metaphor, mis-information is no more a kind are determined by the global pattern of interaction. The of information that a rubber duck is a kind of duck.) structure of the data and its meaning cannot always My goal here is not to enter the debate head on, but to exist as distinguishable macro-states without the inter- ask a dierent question: To which notion of information action with S and the purpose modulated control rela- is DSI a more appropriate generalization? tion between the information medium and the eectors. Floridi is clear that the question about whether trutht- Using the notion of truthfulness in the situation is per- fulness is necessary for information is specically tar- haps an undesirable stretch of terminology it should geted to declarative, factual semantic information.(Floridi be reserved for declarative information but the in- 2010) This is the kind of informational presentation formational states in the system are signicant because where one can separate the question of meaning from of the way they are actually correlated to the states of the question of truthfulness. Let d be a datum. One the world and because of the way they control the be- asks two separate questions: (1) what is the meaning of havior of the system in light of those correlations. This d? (2) Is the fact depicted by d the case ? It should be is, I think, the true motivation behind the SSI. The clear immediately from the discussion in the previous insistence for the condition of truthfulness was never sections that in DSI we cannot assume that a datum motivated by a conviction that a formal semantic valu- can have meaning independent of the way it partakes ation must be added to the concept of information. It is in the dynamical process of a user interacting with an- not about merely xing an alethic parameter to `true'. other system S. In fact, in the general case, what the The motivation behind the theory has always been the datum is is determined by the entire process (including idea that when we say that someone has information counterfactual conditions). From the point of view of about something we are interested in how the state of DSI, the case of declarative information for which the the world is internalized in the person (or organism, questions (1) and (2) can be separated is a restricted robot, etc.) and how the person can act accordingly. case of semantic information. I would claim (though I When one asks for information about the weather one would not argue for this here) that deliberative infor- does not ask for a random meaningful statement with mation is a cognitive phenomenon. the weather as the topic; one asks for a link to the weather so that she can change her actions accordingly. Which theory, WSI or SSI, needs the separation In a sense, the motivation behind SSI has always been more? WSI cannot even be formulated with media for (a hidden form of ) pragmatism. which questions (1) and (2) cannot be asked separately. To say that meaning, but not truthfulness, is necessary Couldn't one object, however, that my analysis of for a datum to count as information demands specifying WSI is unfair? It could be claimed that DSI generalizes the meaning of the datum independent of the circum- WSI just as much as SSI. The two theories simply col- stances under which it is truthful. SSI does not require lapse when one cannot separate between questions (1) the separation, even though for declarative information and (2). Indeed, if we can talk about data, as DSI does, it is formulated with the separation at hand. If one takes and talk about meaning, as DSI does, then WSI can call the pragmatic approach to information, SSI appears to it information. This is a valid response and a possible be a more general theory. Now, this may sound counter theoretical choice, but I think that is a wrong choice. intuitive. WSI appears to be more admissive than SSI The reason is that it violates the spirit of WSI. The it allows for more things to count as information. SSI, reason WSI proponents want to call more things, false after all, places a further constraint on the condition of things, information is because they think that for many problems related to manipulation of information 33 the information it adds a condition of truthfulness. WSI says: information = data + meaning. SSI says: informa- truthfulness value of the data states is irrelevant. One tion = data + meaning + truthfulness. The intuition may want to study how a person, a robot, or an ex- that WSI is a more general notion of information is pert system manipulates informational states to, say, a consequence of the amendment view of information. develop a theory of reasoning. For many theoretical The pragmatic approach however measures generality problems related to information systems (in the IST in a dierent way. SSI is more general because it sup- sense) the truthfulness value of the informational states ports a conception of information that is applicable to may be ignored. For such problems it is important to a wider set of systems. Let us see why! 33 ...and not merely manipulation of data. In information sys- According to DSI, in the most general case, the tems theory (IST) one is often interested in manipulating infor- structure of the data, its dierential interface function mation in a semantically sensitive way. 19 have a theoretical notion that maintains alethic neutral- Beer, R. 2000. Dynamical approaches to cognitive sci- ity.(Floridi 2003, 2010) Such problems can be important ence. Trends in Cognitive Sciences, 4, 9199. for pre-cognitive information systems (in my sense) as Bogdan, R. J. 1988a. Information and Semantic Cogni- well. The pragmatic strategy approaches the problem of tion: An Ontological Account. Mind and Language, information by starting with the notion of information 3(2), 81122. system and then dening information as a currency of Bogdan, R. J. 1988b. Replies to Commentators. Mind the system. This means that a theory of information and Language, 3(2), 145151. systems (in my sense) must be able to investigate the Bogdan, R. J. 1994. Grounds for Cognition: How Goal- internal informational dynamics of the system any Guided Behavior Shapes the Mind. Lawrence Erl- invariance and patterns that exist. It would be required baum Associates. to have a theoretical notion that abstracts away from Burch, R. 2010. Charles Sanders Peirce. In: Zalta, some of the constitutive elements, such as the source, E. N. (ed), The Stanford Encyclopedia of Philosophy which determine the structure of the information sys- (Spring 2010 Edition). tem. The approach of WSI provides exactly this. It pro- Carnap, R., & Bar-Hillel, Y. 1952. An Outline of a vides an abstract theoretic notion that can be used for Theory of Semantic Information. Tech. rept. 247. investigation of the currency information, qua inter- MIT. nal system operation. It should be distinguished from Chemero, A. 2009. Radical Embodied Cognitive Science. the more general notion of information related to the MIT Press. highly structured dynamics of a system interacting with Cooper, R., Mukai, K., & Perry, J. (eds). 1990. Situa- an environment. tion Theory and its Applications. Vol. 1. CSLI. The debate between WSI and SSI is, therefore, a bit Cooper, R., Mukai, K., Barwise, J., & Perry, J. (eds). misguided, in the sense that both ideas are theoretically 1991. Situation theory and its applications. Vol. 2. important and a mature general theory of information CSLI. should have use for both. I would prefer if the term Dennett, D. C. 1991. Real Patterns. Journal of Philos- `information' be preserved for the more general notion ophy, LXXXVIII, 751. connected with SSI, and another term is used for the ab- Di Paolo, E. A. 2005. Autopoiesis, Adaptivity, Teleol- stract notion of WSI, but I would live with the abusive ogy, Agency. Phenomenology and the Cognitive Sci- notation of calling both `information', as long as the ences, 4, 429452. context makes it clear whether the alethic/pragmatic Dretske, F. 1981. Knowledge and the Flow of Informa- dimension has been abstracted away. tion. MIT Press. Dretske, F. 1988. Commentary: Bogdan on Informa- tion. Mind and Language, 3(2), 141144. Fetzer, J. H. 2004. Information: Does it Have To Be References True? Minds and Machines, 14, 223229. Floridi, L. 2003. Information. In: Floridi, L. (ed), The Acko, R. L. 1958. Towards a Behavioral Theory of Blackwell Guide to the Philosophy of Computing and Communication. Management Science, 4(3), 218 Information. Oxford - New York: Blackwell. 234. Floridi, L., & Sanders, J. 2004. Levellism and the Aczel, P., Israel, D. J., Katagiri, Y., & Peters, S. (eds). Method of Abstraction. IEG Research Report IEG- 1993. Situation theory and its applications. Vol. 3. RR-4. Oxford University. CSLI. Floridi, L. 2004. Outline of a Theory of Strongly Seman- Adriaans, P. 2008. Philosophy of Information: Concepts tic Information. Minds and Machines, 14(2), 197 and History. In: Adriaans, P., & van Benthem, J. 222. (eds), Handbook on the Philosophy of Information. Floridi, L. 2005. Is Information Meaningful Data? Phi- North Holland. losophy and Phenomenological Research, 70(2), 351 Bar-Hillel, Y. 1964. Language and Information: Selected 370. Essays on Their Theory and Application. Addison Floridi, L. 2007. In defence of the veridical nature of Wesley. semantic information. European Journal of Analytic Barwise, J., & Seligman, J. 1997. Information Flow: Philosophy, 3(1), 3141. The Logic of Distributed Systems. Cambridge Tracks Floridi, L. 2008a. The Method of Levels of Abstraction. in Theoretical Computer Science, no. 44. Cambridge Minds and Machines, 18(3), 303329. University Press. Floridi, L. 2008b. Trends in the Philosophy of Infor- Barwise, J., & Perry, J. 1983. Situations and Attitudes. mation. Pages 113132 of: Adriaans, P., & van Ben- Bradford Book. 20 them, J. (eds), Handbook of Philosophy of Informa- MacKay, D. M. 1969a. Information, Mechanisms and tion. North Holland. Meaning. MIT Press. Floridi, L. 2010. The Philosophy of Information. Ox- MacKay, D. M. 1969b. Meaning and Mechanism. In: ford. Information, Mechanism and Meaning. MIT Press. Floridi, L., & Taddeo, M. 2005. The Symbol Ground- MacKay, D. M. 1969c. The Place of `Meaning' in the ing Problem: a Critical Review of Fifteen Years of Theory of Information. In: Information, Mecham- Research. Journal of Experimental and Theoretical nism and Meaning. MIT Press. Articial Intelligence, 17(4), 419 445. Maturana, H. R., & Varela, F. J. 1980. Autopoiesis and Floridi, L., & Taddeo, M. 2007. A Praxical Solution Cognition: The Realization of the Living. Springer. of the Symbol Grounding Problem. Minds and Ma- Millikan, R. 1987. Language, Thought, and Other Bi- chines, 17(4), 369389. ological Categories: New Foundations for Realism. Grunwald, P. D., & Vitanyi, P. M. 2008. Algorithmic In- MIT Press. formation Theory. In: Adriaans, P., & van Benthem, Millikan, R. 1995. White Queen Psychology and Other J. (eds), Handbook on the Philosophy of Information. Essays for Alice. MIT Press. North Holland. Millikan, R. 2006. Varieties of Meaning: The 2002 Jean Haken, H. 1993a. Advanced Synergetics: Instability Nicod Lectures. MIT Press. Hierarchies of Self-Organizing Systems and Devices. Morris, C. W. 1938. Foundations of the Theory of Signs. 3rd. edn. Springer, Berlin. University of Chicago Press. Haken, H. 1993b. Synergetics: An Introduction. 3rd edn. Nauta, D. 1970. The Meaning of Information. Mouton. Springer, Berlin. Newell, A., & Simon, H. 1981. Computer science as em- Haken, H. 2000. Information and Self-Organization: pirical enquiry. In: Hougeland, J. (ed), Mind design. A Macroscopic Approach to Complex Systems. 2nd. MIT Press. edn. Springer. Peirce, C. 1940. Logic as Semiotic: The Theory of Hinrichsen, D., & Pritchard, A. J. 2005. Mathematical Signs. In: The Philosophy of Peirce: Selected Writ- Systems Theory I - Modelling, State Space Analysis, ings. Routledge and Kegan Paul. Stability and Robustness. Springer. Putnam, H. 1975. The Meaning of `Meaing'. In: Gun- Hirsch, M. W., Smale, S., & Devaney, R. L. 2004. Dif- derson, K. (ed), Language, Mind and Knowlege. Min- ferential Equations, Dynamical Systems, and an In- nesota Studies in the Philosophy of Scince, no. VII. troduction to Chaos. Academic Press. Unversity of Minnesota Press. Israel, D. J., & Perry, J. R. 1990. What is Information? Seligmen, J. 1991. Physical Situations and Information In: Hanson, P. (ed), Information, Language and Cog- Flow. In: Robin Cooper, Kuniaki Mukai, J. B. J. P. nition: Vancouver Studies in Cognitive Science, vol. (ed), Situation Theory and its Applications, vol. 2. I. University of British Columbia Press. CSLI. Ivancevic, V. G., & Ivancevic, T. T. 2008. Com- Sethna, J. P. 2009. Statistical Mechanics: Entropy, Or- plex Nonlinearity: Chaos, Phase Transitions, Topol- der Parameters, And Complexity. Clarendon Press. ogy Change and Path Integrals. Understanding Com- Shannon, C. 1948. The Mathematical Theory of Com- plex Systems. Springer. munication. Bell Systems Technical Journal. Jianbo Gao, Yinhe Cao, W.-w. T., & Hu, J. 2007. Mul- Thelen, E., & Smith, L. 1994. Dynamics system ap- tiscale Analysis of Complex Time Series: Integration proach to the development of cognition and action. Of Chaos And Random Fractal Theory, And Beyond. MIT Press. Wiley-Interscience. Turchin, V. 1990. Cybernetics and Philosophy. Pages Katok, A., & Hasselblatt, B. 1996. Introduction to the 6174 of: Geyer, F. (ed), The Cybernetics of Complex modern theory of dynamical systems. Cambridge. Systems. Intersystems, Salinas, California. Kelso, J. A. S. 1995. Dynamic Patterns: The Self- Vakarelov, O. forthcoming. The Cognitive Agent: Over- Organization of Brain and Behavior. MIT Press. coming informational limits. Philosophical Psychol- Ladyman, J., Ross, D., Spurrett, D., & Collier, J. 2007. ogy. Every Thing Must Go: Metaphysics Naturalized. Ox- van Gelder., T. 1998. The dynamical hypothesis in cog- ford. nitive science. Behavioral and Brain Sciences, 21, Levine, W. S. (ed). 1996. The Control Handbook. New 615628. York: CRC Press. Varela, F. 2000. El fenomeno de la vida. Santiago de Li, M., & Vitanyi, P. 1997. An Introduction to Kol- Chile: Dolmen Ediciones. mogorov Complexity and Its Applications. Springer. Varela, F. J., Thompson, E. T., & Rosch, E. 1992. The Embodied Mind. MIT Press. 21 von Uexküll, J. 1909. Unwelt und Innenwelt der Tierre. Springer, Berlin. von Uexküll, J. 1932. Streifziige durch die Umwelten von Tieren und Menschen. Springer, Berlin. von Uexküll, J. 1982. The Theory of Meaning. Semiot- ica, 42(1), 2582. Weaver, W., & Shannon, C. E. 1963. The Mathematical Theory of Communication. Univ. of Illinois Press. Weber, A., & Varela, F. 2002. Life after Kant: Natural purposes and the autopoietic foundations of biologi- cal individuality. Phenomenology and the Cognitive Sciences, 1(2), 97125. Wiener, N. 1965. Cybernetics: or the Control and Com- munication in the Animal and the Machine. MIT Press. Williams, P., & Beer, R. Forthcoming. Information dy- namics of evolved agents. In: From Animals to Ani- mats 11: Proceedings of the 11th International Con- ference on Simulation of Adaptive Behavior.