=Paper=
{{Paper
|id=Vol-1315/paper13
|storemode=property
|title=Information for Cognitive Agents
|pdfUrl=https://ceur-ws.org/Vol-1315/paper13.pdf
|volume=Vol-1315
|dblpUrl=https://dblp.org/rec/conf/aic/Fresco14
}}
==Information for Cognitive Agents==
https://ceur-ws.org/Vol-1315/paper13.pdf
Information for Cognitive Agents
Nir Fresco
The Edelstein Centre,
The Hebrew University of
Jerusalem, Israel
fresco.nir@gmail.com
Abstract. Humans use information in everyday activities, including learning, planning,
reasoning and decision-making. There is broad agreement that, in some sense, human
cognition involves the processing of information, and, indeed, many psychological and
neuroscientific theories explain cognitive phenomena in information-theoretic terms.
However, it is not always clear which of the many concepts of ‘information’ is the one
relevant to understanding the nature of human cognition. Here, I suggest that
information should be understood pragmatically. Whatever the criteria for information
are, what makes some x informational has to do with how an agent either processes or
can process x. Information is defined as meaningful structured representations of
perceptual data. Their meaningfulness is determined by their behavioural effect on the
agent.
Keywords: Cognition; Information; Behavioural effect; Data; Cognitive Science.
1 Introduction
There is broad agreement that, in some sense, human cognition involves the processing of
information. Humans regularly use information in learning, planning, reasoning and
decision-making. Many theories in cognitive science explain cognitive phenomena in
information-theoretic terms. Yet, ‘information’ means many things to many people. So, it is
not always clear which of the many concepts of ‘information' is the one relevant to
understanding the nature of human cognition. C. Shannon and W. Weaver defined
information-content as the probability of a message being selected from a finite set of
messages with any selection being equally probable [1]. R. V. L. Hartley before them had
developed measures for the capacities of different types of information systems to transmit
information [2]. More recently, Kolmogorov Complexity has defined the information-
content in a binary string s as the length of the shortest program that produces s on a
universal Turing machine [3, 4].
However, all these offer quantitative analyses of information for measuring the
information-content in a message, rather than a theory of information as the thing that is to
be measured. As noted by Hartley, Shannon and Weaver, their theories focused on physical
features of signal communication, rather than the psychological or semantic features of
information. Whilst quantitative aspects of information-content are clearly of importance to
an information-theoretic analysis of cognition, it seems crucial to fix the concept of
semantic ‘information’ that is used by information theories of cognition in artificial
intelligence and cognitive science broadly. In the next section, I survey a few of the well-
known theories of semantic information and point out their deficiencies as the basis for
informational theories of cognition.
In this paper, I suggest that information should be understood pragmatically first and
foremost, if we are to understand human cognition information-theoretically. Whatever the
criteria for information are, what makes data informational (for an agent) has to do with
how the agent either processes or can process these data. (Here, I adopt L. Floridi’s data-
oriented definition of information [5] with important modifications as is discussed below.)
Information should be best understood as meaningful structured representations of
perceptual data as is discussed in Section 3. The meaningfulness of perceived data is
determined by their behavioural effect on the agent as a triadic, rather than dyadic, relation
Page 148 of 171
involving a physical object (or event or property or state of affairs), the agent’s neural state
and the behavioural effect on the agent. The account sketched here resembles other neo-
Peircean analyses of representation [6, 7] as well as more recent accounts of information [8,
9]. The relationships between the present account and other neo-Peircean analyses are
discussed in Section 4. Section 5 concludes the paper with some general reflections.
2 A brief survey of accounts of semantic information
An important principle underlying many probabilistic accounts of semantic information had been
originally formulated by K. Popper. “[T]he amount of empirical information conveyed by a
[set of sentences...] increases with its degree of falsifiability” [10]. This principle was later
coined the Inverse Relationship Principle (IRP): the less likely a message is, the more
informative (or rather informational) it is [11]. The first systematic theory of semantic
information based on IRP was formulated by Y. Bar-Hillel and R. Carnap [12]. According to this
theory, the thing that carries information or has informational content is sentences.
The meaningfulness of information is relative to some logical probability space.
Information is assigned to messages about events and the selected information measure
depends on the logical probability of events or some properties of an object the message is
about. Logical probability is defined in this context as a function of the set of possible
worlds a sentence rules out.
Some have argued that this theory (and any other IRP-based theory) leads to a paradoxical
result [5, 13]. If all the consequences of known sentences are known, any logically true
sentence (that is, a tautology) does not increase knowledge and, hence, does not contain
information. A tautology excludes no possible worlds and its logical probability is 1. At the
same time, a self-contradictory sentence excludes all possible worlds and its logical
probability is 0. Counter intuitively it contains maximal information. I return to this so-
called paradox below, but for now, it should be noted that the Bar-Hillel/Carnap theory
cannot serve as a basis for human cognition broadly. For it is defined in terms of sentences,
and the domain of cognition is broader than language processing alone.
A more recent theory of information was offered by F. Dretske [14]. His theory is premised
on the idea that information can be used as part of a reductive analysis of knowledge and
cognition. On his view, the information carried by a message is relative to the epistemic state of
the agent receiving that message. He was motivated by the central observation in the
Shannon/Weaver theory that the receipt of information should reduce the agent’s uncertainty.
By applying the underlying communication model in the Shannon/Weaver theory to knowledge,
the source of messages is the physical world and the receiver is a would-be knower. For Dretske,
perceptual knowledge can (and should) be understood in terms of information. “K knows
that s is F = K’s belief that s is F is caused (or causally sustained) by the information that s
is F” [14]. The information that s is F affects K’s belief in such a way that the information
suffices for the formation of the belief absent other contributing (or conflicting) factors. K
must discern physical events in the world that carry the particular information, and those
events have to cause (or causally sustain) K’s belief that s is F. Moreover, the informational
content of a message is also conditional on what K already knows when receiving the
message. Importantly, Dretske maintained that information must be truthful. “Information
is what is capable of yielding knowledge, and since knowledge requires truth, information
requires it also” [14]. Other supporters of the idea that information must be truthful include
P. Grice [15], J. Barwise [11] and P. Allo [13].
Floridi has adopted some of Dretske’s main ideas (including the idea that information cannot
be false), whilst rejecting IRP and insisting on a stronger constraint on semantic content. His two
main motivations for adopting the Veridicality Thesis (i.e., that information must be truthful)
are (a) to provide a link between information and knowledge, and (b) to avoid the Bar-
Hillel/Carnap paradox concerning the alleged informativeness of contradictions [5]. The first
motivation is similar in spirit to Dretske’s in establishing a close link between knowledge
and information. The second motivation – being that tautologies contain no information,
whereas contradictions contain maximum information (an underlying principle of classical
Page 149 of 171
logic) – has led him to deny IRP and suggest a stronger constraint that is based on closeness
to truth. According to Floridi, “the amount of informativeness of each [message] can
be evaluated absolutely, as a function of (a) […] the alethic value possessed by [the
message] and (b) the degree of discrepancy […] between [the message] and a given state of
the world” [5]. (Note the difference from Dretske’s approach where information is
conditional on the epistemic state of the receiver.)
Yet, besides the veridicality constraint, he proposes to understand information as
meaningful and structured data. Unlike the Bar-Hillel/Carnap theory, information carriers
are understood as data rather than sentences only. What is a datum? In its simplest form, it
is the lack of uniformity in the real world. Examples of a datum include a black dot on a
white page, the presence of some noise, a light in the dark or a logical 0 as opposed to a 1.
A datum is defined as two distinct uninterpreted variables in a domain that is left open to
further interpretation [5]. Data are structured when they are “rightly put together, according
to the rules (syntax) that govern the chosen system, code or language being used. Syntax
here must be understood broadly, not just linguistically” [16]. That they are meaningful
means that the data “must comply with the meanings (semantics) of the chosen system,
code or language in question. […] The data constituting information can be meaningful
independently of an informee [and need not be] necessarily linguistic” [16].
There are clearly other important theories of information that are worth exploring, but
this exceeds the scope of this paper. For example, D. MacKay offered a quantitative theory
of semantic information based on the receiver’s increase in knowledge. “[W]e have gained
information when we know something now that we didn't know before; when ‘what we
know’ has changed” [17]. Another example is B. Skyrms’ analysis of information –
grounded in signalling games – where senders of signals observe states of the world and
communicate with receivers that in turn choose an act in response to receiving signals [18].
For him, information is correlated with states of the world as well as with actions.
3 Towards a theory of semantic information as meaningful structured
representations of data
Space only permits a few, brief remarks regarding the adequacy of the theories of
information outlined in Section 2. (This is discussed elsewhere [19].) The Bar-Hillel/Carnap
theory of information is defined in terms of sentences, and, thus, is unable to account for
many non-linguistic informational aspects of cognition. Dretske and Floridi’s accounts of
information aim specifically at explaining knowledge. Yet, that objective has led them to
adopt the Veridicality thesis that restricts the applicability of information to other cognitive
phenomena. Cognitive agents cannot always ascertain the veracity of the information they
process and one of the most important methods of learning is by trial and error that clearly
involves making mistakes (or false information). The processing of information in cognitive
agents is insensitive to the veridicality of the information. Belief change models, for
example, explain rationality is terms of justified doxastic commitments that are consistent.
These models are underpinned by the principle that all information, even veridical
information, is defeasible and subject to revision under the right conditions. Besides, on
standard frameworks of belief change, false perceptual information can actually lead to
truth approximation via belief revision and increase the agent’s overall knowledge base.
To underscore the pragmatic value of information for the receiving agent consider a
simple example. Suppose that the same message is sent twice by the same information
source. The two messages clearly carry the same information-content. Nevertheless, only
the message that is successfully received by the receiver first is informative. Of course,
receiving the second message – with the very same information content – can still be useful,
for example, in the presence of noise: the first message could have been distorted during
transmission. Moreover, in some contexts, each of the messages, arguably, carries
additional meta-information that is its temporal indexing: message one was sent (or rather
Page 150 of 171
received) at Tx and the second at Ty. This temporal indexing might also be pragmatically
significant: it may tell the receiver that some state of the information source has remained
unchanged. Nevertheless, all this is meta-information in addition to the information-content
of each of the individual messages (e.g., if each message includes a timestamp as part of its
content, the information-content of the two messages is different).
Crucial to the new theory sketched herein is the triadic basis of information. Rather than
taking information to be a dyadic relation that obtains between signs and objects (or states
of affairs) in the world, information requires a third element: its receiver. On Floridi’s
theory, for example, some information (i.e., environmental information) supposedly exists
in the world independently of any receivers (e.g., concentric rings in the trunk of a tree that
can be used to calculate the tree’s age qualify as information even in the absence of any
perceiver) [5]. But as argued by Dretske, the informativeness of a message is relative to the
epistemic state of the receiving agent. Smoke in the forest (usually reliably) signifies there
being fire to receivers of information that interpret the signals (smoke particles or
combustion aerosols) as a potential imminent danger nearby. This triadic relation can
already be found in the works of C. S. Peirce: something is a sign (also “representamen”)
only if it signifies an object with respect to an “interpretant” (i.e., a mediating
representation in the mind of some agent) [20]. Whilst there is a causal correlation between
smoke and fire based on natural regularities, the receiver of the signals (smoke particles)
plays a key role in the formation of the information (there being fire in the forest). The
receiver may know that smoke machines are used in the forest (for some bizarre reason)
and, consequently, may not interpret the signals received as there being fire in the forest.
The theory proposed here uses Floridi’s data-oriented definition of information with
some important modifications. Objects, events or states of affairs in the world are sources of
physical signals or data with which they are causally correlated. Physical data as
discontinuities in the world exist “out there” unstructured. Their structuring is an ongoing
dynamic interaction between the receiving agent and her environment. But data need not
always originate externally to the receiver. An organism, for example, can receive pain
signals from one of its limbs. Further, the structure of the data in the wild is determined by
an agent-environment function. If either of these two contributing factors is missing, there
is no information just data. In that sense, the physical data “out there” constrain the
information that can be formed by the receiver on their basis. Unless the agent is
hallucinating in a void or dreaming, her perceptions are formed on the basis of stimuli
(understood as data) from the world to which she is sensitive. Our cognitive apparatus only
allows us to discriminate some, but not all, physical discontinuities and nomic regularities
in the world. (Whilst elephants, for example, are sensitive to infrasound, humans are not
readily sensitive to infrasound signals.) Only those data to which we are cognitively
sensitive can give rise to the formation of information. Any perceived physical data “out
there” are encoded, or represented, as some form of neural patterns (e.g., as action
potentials or activation patterns). The precise form of representation is a further empirical
question.
The meaningfulness of the perceptually structured data is determined by their
behavioural effect (either positive or negative) on the receiver. Such behavioural effect is
broadly construed to encompass more than just observable behaviour. It amounts to,
roughly, the change produced in the receiver’s action(s), belief(s) or goal(s) resulting from
the data perceived (e.g., leaving the forest immediately when smelling or seeing smoke on a
very hot day). In that sense, the state of the world – as signified by the perceived data – and
the receiver are connected. This change implies, as argued in [21], that there exists a
requisite flexibility of behaviour in the receiver, such that the perceived data can yield some
change in the receiver. It makes little or no sense to describe a rigid system S as being
informed by something if S cannot somehow behave differently upon receiving these data.
Further, any consequence of the perceived data is the result of how the receiver interprets
the data and behaves in the world accordingly [22]. However, for the perceived data to
Page 151 of 171
be meaningful there need not be any necessary dependence on a kind of coordination
system amongst senders and receivers. Data need not be communicated amongst agents in
order to be meaningful, and can flow directly from the world to the receiver [21]. Indeed,
the world does not communicate with agents. It is rather the sensitivity of the receiver to
particular regularities or physical discontinuities in the world that “flow” to the receiver.
Moreover, the effect concerned need not be necessarily positive (e.g., the receiver being
informed about a nearby reservoir of water); it can often be negative (e.g., drawing a false
conclusion regarding the distance of the reservoir). The distinction between negative and
positive effects is what determines the relevance of information, as argued by D. Wilson
and D. Sperber [23], not whether the meaningful structured data qualify as information. On
their view, information is relevant to the agent when it (1) relates to her background
information to derive conclusions that matter for her beliefs or actions, and (2) requires less
processing effort by the agent. Others define the relevance of information relative to goals.
A piece of information is relevant (for a goal) iff “it is a candidate for a belief that supports
the processing of that goal” [24]. But either way, the relevance of information can only be
determined once we have established what qualifies as information. The meaningfulness of
the perceptual data is a prerequisite for the information being relevant. Understood this way,
there is clearly room for mistakes (as a negative effect) in the agent forming information. An
agent may mistake smoke particles for indicators of fire nearby, where, as a matter of fact,
that smoke may be produced by smoke machines. Her escape from the forest would be
rationally justified absent other overriding factors, despite there being no fire or imminent
danger.
The theory proposed herein postulates that there is an important distinction to be made
between information-that and information-how on the basis of the role information plays in
cognitive processing. Information-how (e.g., ‘In case of fire, break the glass and press the
button’) is prescriptive and informs an agent about which action has to be performed to
achieve a particular result. As such, for cognitive agents it expresses an expectation for
some goal-directed action on the part of the receiver in a given context. Information-that
(e.g., ‘Not all birds can fly’) is descriptive and is about events, objects and states of affairs
in the world. Cognitive agents use information-that to represent and form beliefs about,
rather than merely externally react to, their environment. Both types of information play an
important role in the way cognitive agents negotiate with their environment in terms of
acting and believing. Neither information-how nor information-that need be restricted to
sentences.
Lastly, why is this particular view of information considered apt to capture the kind of
information processing often invoked in cognitive science? First, understanding information
as being carried by data allows a broader applicability of the theory beyond linguistic
aspects of cognition alone. To understand cognitive agency, what we want is a theory that
focuses on physical information, and in that regard data-centred theories fare better.
Sentences convey information, but so do sunlight and smoke, for example. Yet, unlike the
Floridian data-centred theory of information, the present theory does not insist on the
Veridicality thesis. Cognitive agents all too often make mistakes in interpreting perceptual
data. Such mistakes should also be accounted for in explaining cognition. Second,
information in cognitive science provides a naturalistic foundation for the explanation of
cognition and behaviour. Humans and other organisms survive and reproduce by tuning
themselves to reliable but imperfect cues that represent correlations between internal
variables and environmental stimuli as well as between environmental stimuli and
opportunities and threats [25]. The meaningfulness of perceived data described above is
determined precisely by such “reliable but imperfect cues” the agent is sensitive to.
Third, the theory is neither too narrow nor too broad for our purposes. It is not too
narrow in either imposing strict conditions that only few cognitive processes satisfy (e.g.,
the veridicality of the data for knowledge) or being limited to a subset of cognitive
phenomena (e.g., language processing). It is compatible with the contemporary cognitive
Page 152 of 171
scientific view that “the brain reveals itself proactive in its interface with external reality”
being an interpreter rather than a mirror of that reality [26]. “[R]esearch […] has shown
how signals coding predictions about […] simple features of relevant events can influence
several stages of neural processing” [26]. The proposed theory is equally compatible, for
example, with a recent, and contentious, view of the brain as an hypothesis-testing
mechanism that attempts to minimise the error of its predictions about perceptual data from
the world [27]. Both “bottom-up” signals (perceptual input data) and “top-down” signals
embodying predictions about the probable causes of the perceptual input data can qualify as
information according to our theory. At the same time, the theory is not too broad so as to
make information vacuous. Information can come at degrees. Some data do not give rise to
information, since the receiver is not sensitive to them. Other data are simply not
meaningful to their receiver. And although both a tautology and a contradiction, for
example, can be informational, they are less or more useful and/or relevant in a given
context.
4 A comparison with other neo-Peircean theories
In this section, the relationships between the proposed theory and other neo-Peircean
analyses of representation and information are discussed. To begin with, consider B. von
Eckardt’s analysis of non-mental representation. In [6] she adapts Peirce’s triadic relation
that obtains amongst the represented object, the representing vehicle (representamen) and
the mental effect in the mind of the interpreter of the sign (interpretant). The represented
object could be a physical object, a relation, a state of affairs or a property. The representing
vehicle – what she calls the representation bearer – such as a map, a photo or a spoken
word, can be individuated in terms of its nonrepresentational (or material) properties. Both
the represented object and the representation bearer are, at least in principle, objectively
verifiable. von Eckardt claims that in order for R to be an actual – rather than merely a
possible – representation there must currently exist an actual interpreter bearing the right
relation to R. The resemblance to the proposed theory of information should be clear.
Information is understood pragmatically and in a manner that requires an actual consumer
of physical data (that can be upgraded to information under the right conditions). On the
other hand, data need not be communicated by senders. Physical data “out there” can at best
be classified as potential information in the absence of consumers.
G. O’Brien and J. Opie build on von Eckardt analysis of non-mental representation and
add that the vehicles of mental representation should be understood as some kind of neural
states [7]. Given their commitment to a naturalistic account of cognition, they seek to
explain the act of interpretation in naturalistic terms in order to avoid a vicious circle. They
claim that the only viable alternative is treating interpretation in terms of some modification
of the cognitive agent’s behavioural dispositions towards the represented object. Here, too,
the similarity is clear. The proposed theory of information suggests that the meaningfulness
of perceived data (and, therefore, their being informational) is determined by their
behavioural effect on the agent. It is suggested that on receiving new information some effect
in the agent triggers an action or a response (e.g., forming/changing a belief-state).
On E. Jablonka’s functional-evolutionary analysis of semantic information, the
distinction suggested above between information-that and information-how becomes very
blurry. That is the case, for example, when ‘functional’ means that signals received by
either a human- or natural-selection designed system play a causal role that “usually
contributes to the goal-oriented behavior of this system” [9]. An apple pie recipe and a
piece of software are instances of functional-evolutionary information for a cook and a
computer, respectively, in a manner akin to the appearance of black cloudy sky leading to
the shelter-seeking action of an observing ape. Nevertheless, insofar as we seek to
understand the role information processing plays in cognitive tasks in the lifetime of an
agent, rather than over evolutionary time, the information-how/information-that distinction
Page 153 of 171
seems worth preserving.
Lastly, on J. Queiroz, et al. neo-Peircean theory, information has the nature of a process
of communicating a “form” to the interpretant [8]. That process constrains the possible
patterns of behaviour of the interpreter. Information is taken typically as an interpreter-
dependent “objective” process. Accordingly, it cannot be dissociated from a situated agent.
On their view, it is only as a result of the interpretation process that information triadically
connects the sign, object(s), and an effect on the interpreter. A sign (somehow) effectively
communicates a form from the (represented) object to the interpretant, whilst changing the
state of the interpreter. This account raises some interesting questions, which are not tackled
here, about the objectivity of this process when it is dependent on a particular agent and
about the communication of the form of an object to the interpreter (the world does not talk
to us…). Nevertheless, it can be seen again that information is not simply “out there” in the
world independently of a perceiver. Information is a dynamic construct that results from an
ongoing interaction between the agent and its environment.
5 Concluding remarks
This short paper contributes to a long-standing and much-debated question of what concept
of ‘information’ is suitable for understanding human cognition in terms of information
processing. It is often argued, in cognitive science, that cognition is an information
processing system. The literature contains many diverse theories of information (of which I
have surveyed but a few here) pulling in different directions, thereby leading to disparate
definitions of ‘information’. Information, so I have suggested whilst adapting a neo-Peircean
approach, should be understood pragmatically. Whatever the criteria for information are,
what makes x a piece of information has to do with the way the agent either processes or can
process x in actively engaging with her environment. Of course, it does not follow that a
unified theory of information is either forthcoming or even possible. In different contexts,
such as game theory or economics, information may be defined differently. The theory
proposed herein is motivated by doing justice to the cognitive sciences. However, much
more work is required to fully develop it.
Acknowledgements. This paper has benefited from the insightful comments and suggestions of two
anonymous referees for the 2014 International Workshop on Artificial Intelligence and Cognition.
Thanks are due to Aditya Ghose, Patrick McGivern, Michaelis Michael and Joel Pearson for useful
discussions on topics related to the paper. This research has been supported by a fellowship from the
Edelstein Centre for the History and Philosophy of Science, Technology and Medicine and a research
grant from the Israeli Ministry of Aliyah and Immigrant Absorption. My attendance at the workshop
was funded in part by the Rosselli Foundation.
References
1. Shannon, C.E., Weaver, W.: The mathematical theory of communication. University
of Illinois Press, Urbana (1949).
2. Hartley, R.V.L.: Transmission of Information. Bell Syst. Tech. J. 7, 535–563 (1928).
3. Kolmogorov, A.N.: Three approaches to the quantitative definition of information.
Probl. Inf. Transm. 1, 1–7 (1965).
4. Chaitin, G.J.: Algorithmic information theory. Cambridge University Press,
Cambridge, UK (2004).
5. Floridi, L.: The philosophy of information. Oxford University Press, Oxford (2011).
6. Von Eckardt, B.: What is cognitive science? MIT Press, Cambridge, Mass. (1993).
7. O’Brien, G., Opie, J.: Notes toward a structuralist theory of mental representation. In:
Staines, P.J., Clapin, H., and Slezak, P.P. (eds.) Representation in mind!: new approaches to
mental representation. pp. 1–20. Elsevier: Morgan Kaufmann, Amsterdam (2004).
8. Queiroz, J., Emmeche, C., El-Hani, C.N.: A Peircean Approach to “Information” and
Page 154 of 171
its Relationship with Bateson’s and Jablonka’s Ideas. Am. J. Semiot. 24, 75 (2008).
9. Jablonka, E.: Information: Its Interpretation, Its Inheritance, and Its Sharing. Philos.
Sci. 69, 578–605 (2002).
10. Popper, K.R.: The Logic Of Scientific Discovery. Routledge, London (2002).
11. Barwise, J.: Information flow: the logic of distributed systems. Cambridge University
Press, Cambridge (1997).
12. Bar-Hillel, Y., Carnap, R.: Semantic Information. Br. J. Philos. Sci. 4, 147–157
(1953).
13. Allo, P.: A Classical Prejudice? Knowl. Technol. Policy. 23, 25–40 (2010).
14. Dretske, F.I.: Knowledge & the flow of information. MIT Press, Cambridge, Mass.
(1981).
15. Grice, H.P.: Studies in the way of words. Harvard University Press, Cambridge, Mass.
(1989).
16. Floridi, L.: Information: a very short introduction. Oxford University Press, Oxford!;
New York (2010).
17. MacKay, D.M.: Information, mechanism and meaning. MIT Press, Cambridge, MA
(1969).
18. Skyrms, B.: Signals: evolution, learning, & information. Oxford University Press,
Oxford (2010).
19. Fresco, N., Pearson, J.: How theories of cognition define information. (Unpublished).
20. Peirce, C.S.: On a New List of Categories. Proc. Am. Acad. Arts Sci. 7, 287–298
(1868).
21. Cao, R.: A teleosemantic approach to information in the brain. Biol. Philos. 27, 49–71
(2012).
22. Millikan, R.G.: Biosemantics. J. Philos. 86, 281–297 (1989).
23. Wilson, D., Sperber, D.: Relevance Theory. In: Horn, L.R. and Ward, G.L. (eds.) The
handbook of pragmatics. pp. 607–632. Blackwell, Malden, MA (2005).
24. Paglieri, F., Cristiano Castelfranchi: Trust in Relevance. In: Ossowski, S., Toni, F.,
and Vouros, G. (eds.) Proceedings of AT 2012 - First International Conference on
Agreement Technologies. pp. 332 – 346. CEUR-WS.org, Dubrovnik (2012).
25. Scarantino, A., Piccinini, G.: Information without truth. Metaphilosophy. 41, 313–330
(2010).
26. Nobre, A.C., Correa, A., Coull, J.: The hazards of time. Curr. Opin. Neurobiol. 17,
465–470 (2007).
27. Hohwy, J.: The predictive mind. Oxford University Press, Oxford, United Kingdom
(2013).
Page 155 of 171