Information in Explaining Cognition: How to Evaluate It?

Round 1

Reviewer 1 Report

The paper outlines six desiderata on theories of information that purport to elucidate the concept of 'information' as employed in cognitive science.

The paper is a clearly written and useful overview of issues in the application of the information concept to explanations in cognitive science. I have a few comments that I think would improve the paper.

1. pp. 3-4: Grice offered his natural/non-natural distinction as an analysis of two different senses of the English word 'means'. The first sense is employed in sentences like 'Smoke means fire'. The second sense is employed in sentences like 'Three rings on the bell means the bus is full'. Over the years, his analysis has been taken as distinguishing between two different types of information-carrying vehicle; to a first approximation, his distinction is taken to be the same as the cue/signal distinction. I think you should either (1) Point this out to the reader, or (2) Emphasise that you too are adopting the natural/non-natural distinction as a distinction between two different senses of the word 'information'. (I believe you really want to do (1), but it's your choice.)

2. l.169: I think "...a more restrictive or liberal view of representation" should read "...a more restrictive or liberal definition of representation". That makes clear that the game being played by philosophers (and some cognitive scientists) is conceptual engineering: we are trying to design a concept 'representation' that can do work in cognitive science, not necessarily trying to discover some pre-existing category that is more salient or important than other concepts related to information.

3. paragraph beginning l.176: I'm a little confused by the citations brought in to support the difference between non-natural information and representation. If I remember rightly (and I might be wrong, in which case ignore this comment), those theorists don't take a position on non-natural information. So as well as denying that the states in question are representations, they might also deny that they carry non-natural information. 

4. l.200-204: In general the discussion in this section, and especially these concluding lines, make it seem as though non-natural information is a more well-understood category than representation. I don't think that's true. As a result, some of these moves here don't sit too well with me.

5. footnote 7 seems unnecessary to me. I think you should remove it.

6. Bottom of p.5/top of p.6: the stated differences between natural and sensory information all seem to be a consequence of what strikes me as the fundamental difference: vehicles of sensory information are designed to carry sensory information. Vehicles of natural information (i.e. cues) are not designed to carry information (or at least, qua cue they are not designed).

7. l.277: Earlier you distinguished non-natural information from representation. Here you say non-natural information "refers to representations that can be true or false." This is confusing. Perhaps just delete this line.

8. l.296-300: The first and second reasons strike me as identical. Either combine them or make them more distinct.

9. l.304 typo "neuroscienfic"

10. l.353-354 This sentence seems to conflate natural and sensory information. Presumably it is natural information that is conveyed by light waves and air vibrations; sensory information is what those things are transduced into.

11. l.364ff: The Deacon quote and the following sentences don't quite seem to fit together. You say: "What matters is whether the substrate concerned has sufficient degrees of freedom to support the specific configuration of distinguishable states that constitute the relevant information." To me this sounds like the definition of channel capacity. But the Deacon quote is describing something "irrespective of the amount of Shannon information". (Of course, this isn't helped by the fact that everyone uses the term "Shannon information" to refer to different measures!)

12. l.425 typo: "takes the edge of information"

13. l463 typo "causes"

14. D5: We really need a definition of symbolic and non-symbolic information. The examples were good and quite comprehensive; I just want a definition to hang them on. This becomes important below.

15. Paragraph beginning 505: Because we have no definition of symbolic information to hand, it's difficult for me to understand why the waggle dance is non-symbolic. The waggle dance is iconic, and I can certainly imagine a definition of symbolic information that excludes iconic signals. But among your examples of symbolic information is "iconic pictures", so these examples aren't quite consistent until you say more.

16. l.511 "[not] spatially or temporally distal -- in contrast to symbolic information." Here you are conflating distinctions. Information that is about the here-and-now may be symbolic or non-symbolic. Information that is about a spatially and/or temporally distal state of affairs may also be symbolic or non-symbolic. Unless you have something in mind I haven't considered, in which case, say it.

17. Paragraph beginning 542: it's really not clear to me how the example in this paragraph contributes to its concluding sentences. I would rewrite or remove the whole thing.

18. l.576-8: This is another sentence that's difficult for the reader to evaluate without a definition of symbolic information to hand.

19. l.593: regarding Scarantino's account, a succinct way to say what's going on would be something like: "The probability of an event can truly be 90% even if the event does not occur." 

20. Paragraph beginning 600: I didn't really understand what was going on in this paragraph. I think it's superfluous to the paper, and I think you should remove it. (With apologies to Baker!)

21. l.633: "information theory" should be "theories of information" to avoid confusion with the mathematical toolkit.

22. The conclusion could be refined to more closely mirror the sections of the paper. In particular, there's no mention of GNWT in the conclusion. I think it should be there because I think you did a good job in section 5.

Author Response

Dear Reviewer,

Detailed replies to your comment can be found in the attached PDF document.

All the best,

The Author

Author Response File: Author Response.pdf

Reviewer 2 Report

The article is well-written and the author seems to be widely read and informed about the theoretical stories as well as the relevant recent empirical material. But I think that the article could profit from a change of emphasis and/or re-contextualization.

The article starts by referring to the (age-old) problem of evaluating the notion of “information” in cognitive science and beyond. The author mentions various frameworks that have been conceived over the years that try to explicate the notion of “information” to some satisfaction. He rightly suggests that one should focus on the multi-faceted nature of information.

For example, one specific aspect pertains to the notion of “semantic information”. Semantic information, other than the “classical” concept of information due to Shannon, deals with the meaning of a message and the way this meaning can be an important factor for the behavior of an organism. The way how semantics is sometimes spelled out is via the notion of “reference” or: correlations between (signals and) world states. Importantly, such “meaningful” information is assumed to play an important role in the models of, e.g. animal behavior.

More generally, the author proposes that to evaluate the talk of “information” in specific theories or models in cognitive science, one should determine whether or not “information” does some explanatory work or merely serves as an “explanatory gloss” (l.59). Only in the first case should such theories be called informational and the notion of “information” can be evaluated.

The contribution should not be misunderstood as trying to advertise for any specific notion of “information” or even invent a new one. Rather, it pertains to evaluating theories in cognitive science, concerning their use of “information”. On the one hand, this is good, since there have been enough books written about supposedly good or useful notions of information (though without any definite conclusion!). On the other hand, this leaves me a bit puzzled as to what should actually be achieved here. Is it just the question of whether or not theory X actually uses the notion Y and does so justifiably? Why is this important (for theory X)?

It seems to me that the strongest, most specific claim of the present paper revolves around the “global workspace theory” of consciousness (section 4). I would thus suggest highlighting this in the title (e.g. “evaluating information in the global workspace” or similar) and perhaps (re-)building the narrative around that. It might appear that the author does a lot to “tailor” his general discussion toward making it pertinent to the claims associated with the global workspace hypothesis (e.g. emphasis on unity and integration). Also, the strategy of implying that “information” is highly depending on context (wrt theory and use) is at odds with the approach to give a list of desiderata and find that they are all instantiated by one specific theory. This does not mean that I don’t find the ideas interesting, I do. But perhaps the message would be that some theories pretend to work with a single notion of information, whereas in fact they do not.

Coming to the present narrative of the paper (Secs 2-5):

The first job is to find out whether “information” is important for an explanation. Hence, one should look at what “explanation” actually means. From a philosophy-of-science-perspective, this is a non-trivial task, the author devotes a full section to the question of explanation. One potential job that the notion of “information” could play in cognitive science is to unify or coordinate various research programs (of course, the is also the option of radical pluralism that says: cognitive science should not be unified at all). The question then becomes how the notion of information must be in order to enable this. One desideratum is, then, that information is “consistent across these programs” (l. 102).

Other potential roles of information are whether it directly contributes anything to a theory’s “predictive accuracy” (l.106), its “fruitfulness“ (l. 107), or its “logical consistency” (l. 110). So even if unification is a misleading dream, information might play a role within the individual explanatory enterprise(s). To connect to the final example (GWT): what role does “information” play within that theory?

Then comes a lengthy third section that spells out, first, a somewhat failed attempt, and then a (more successful) “trichotomy” for classifying information into “natural” (exogenous), “sensory”, and “endogenous” information and ask whether they satisfy six “desiderata for cognition-friendly” (l. 280) notions of “information”. This section is probably the most substantial, even though I found it a bit longish compared to the (to me: more) interesting take on global workspace theory. The six desiderata have to do with quantity, asymmetry (substrate-neutrality; receiver > sender), the (a)symbolic nature of information, and the possibility of information to be “mistaken”. (The last point is somewhat contested and different “information theories” accept this last desideratum or not).

In the ensuing section on GWT, the author argues that GWT nicely illustrates this way of thinking about information. This is interesting, but the immediate question for me is: does this in any way help the GWT? An alternative strategy would be: the GWT (wrongly) talks of “information” but means quite different things by this -- according to the taxonomy. Keeping the different aspects of “information” separate would thus profit GWT because … [that’s the interesting question] Perhaps I did not get the scope of this section, but then it should be made clear (and possibly expanded. The author’s discussion on the failed Gricean attempt is almost as long as the section on GWT…)

MINOR:

The numbering of the sections is slightly off. After section 3 (desiderata) comes section 5 (Global Workspace).

Author Response

Dear Reviewer,

Detailed replies to your comments can be found in the attached PDF document.

All the best,

The Author

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thanks for extensively replying to my previous comments. I found the expanded section on GNWT helpful.