Abstract
Much recent work on natural information has focused on probabilistic theories, which construe natural information as a matter of probabilistic relations between events or states. This paper assesses three variants of probabilistic theories (due to Millikan, Shea, and Scarantino and Piccinini). I distinguish between probabilistic theories as (1) attempts to reveal why probabilistic relations are important for human and non-human animals and as (2) explications of the information concept(s) employed in the sciences. I argue that the strength of probabilistic theories lies in the first project. Probability-raising can enable organisms to draw specific inferences they otherwise could not entertain and I show how exactly they help to explain the behaviour of organisms. In addition, probability-raising warrants inferences by providing incremental inductive support.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
Nick Shea’s comments on an earlier draft prompted me to draw this distinction.
Assuming a finite frequency interpretation of conditional probability.
This is also how one referee prefers to understand Millikan’s condition.
The three-place reading has a further implication. Once organisms employ a correlation between As and Bs they use A to learn something specifically about B; they do not use A to learn something about all the other things with which A correlates. So the local information A carries is specific. It is not about anything A correlates with, nor about what it correlates with most strongly. It is only about those things whose correlation with A was strong enough to have influenced sign use.
“For example, in the case of the honeybee the incoming bee’s dance carries correlational information about the location of nectar” (Shea 2007, p. 426). It is natural to read “the incoming bee’s dance” as referring to the token dance of a particular incoming bee (although the phrase can also be interpreted as referring to the type of dance).
Sign use via a history of natural selection plays a different role in Shea’s story: it is what turns information into representational content (thereby putting information back into teleosemantics).
We are told in the next sentence: “On this view, a signal s in state F can carry natural information about an object o being G simply by raising the probability that o is G…” (Scarantino and Piccinini 2010, p. 317, my emphasis).
At one point they require that natural information rely on physical relations: “Bearers of natural information stand for what they are about by virtue of being physically connected to it. In the absence of the connection, no natural information is carried” (Piccinini and Scarantino 2010, p. 242). This statement could be interpreted as presupposing some kind of causal, non-accidental connection. However, Scarantino clarified that it should not be interpreted in this way (personal communication).
Scarantino confirmed this reading of reliability (personal communication). I believe Scarantino and Piccinini’s distinction is both significant and applicable in actual circumstances, as well. Suppose the footprints increase the probability of quails in forest F1 by 94 %, but in forest F2 by only 54 % (where both F1 and F2 are actual forests). The 94 % increase in quail probability in F1 is reliable in the sense of being strong, but unreliable in the sense of being unstable beyond F1. Moreover, the stability sense of reliability matters for organisms. For instance, predators relying on footprints for hunting quail will do worse when moving from F1 to F2.
While writing this paper Scarantino substantially modified his account (submitted manuscript). My exposition and assessment of his theory concerns the presently published version.
Skyrms (2010, p. 1) ties information to Grice’s notion of natural meaning: “Natural meaning [in Grice’s sense, U.S.] depends on associations arising from natural processes. I say that all meaning is natural meaning”. It is therefore reasonable to take Skyrms’ discussion of information in signalling games as concerning natural information.
Another problem is that if wild tokens are physically indistinguishable from normal tokens, users would respond to the first as they would to the second. And this seems to suggest that, from the user’s point of view, one is as informative as the other. If wild tokens carried no information from the user’s point of view, it would be mysterious why the user would respond to them in the way it does to normal tokens. See also Kraemer (2013).
According to Scarantino and Piccinini (2010, p. 318), a signal carries probabilistic information about an event as soon as it increases its probability somewhat. Probabilistic information does therefore not require raising the event’s probability above a threshold that renders the event likely to occur.
The denial of misinformation is compatible with rejecting the veridicality thesis. The latter says (simplified) that if A carries information about B, then B obtains. This thesis is false according to Scarantino and Piccinini (2010): if A is a wild token then A carries (true) information about B (i.e. that B occurs with some increased degree of probability) despite B not occurring.
The degree of association formed by the animal was measured by the degree to which experiencing the shock suppressed the animals’ ordinary activities, in this case bar pressing in order to obtain food. This common measuring procedure is known as suppression conditioning and the resulting measure as the suppression ratio.
I use mentalistic vocabulary here only for convenience. I will return to this point.
A reviewer suggested characterizing the modal sense as making it biologically possible for an organism to engage in correct inferences. But this characterization appears to exclude relevant cases. Suppose an organism is trained (by means of a contingency relation) to infer a food item from a certain smell; furthermore, once it has acquired this ability the circumstances are changed such that the smell is no longer predictive of food. Then the organism would for some time, though with decreasing frequency, engage in false predictions. In this scenario, exposure to a probabilistic relation makes it possible for the organism to engage, for some time, in false inferences.
These asymmetries in the ability to associate stimuli are thought to arise from the organisms’ sensory ecology. Internal illness is usually inflicted by food poisoning, and so sensory modalities used to detect food influence the stimuli likely to be associated with illness. Most birds identify food by visual inspection, but most mammals by smelling and tasting it.
Dretske (1988, pp. 96–107) also discusses learning in order to argue that natural information can explain behaviour. But there his argument and his conclusions are very different from what is shown here. First, Dretske does not primarily consider the informational relation between two states of the environment, but rather that between an environmental condition and an internal state (although elsewhere Dretske often discusses examples of informational relations between two environmental conditions); so Dretske’s argument does not establish that environmental correlations are explanatory (unless tied in the relevant way to an internal indicator). Second, Dretske refers to ‘indication’, which for him is a relation that increases to one the probability of the indicated event; whether his argument also works with imperfect probabilistic relations is left open. Third, Dretske’s result is contingent on learning by reward (i.e. instrumental learning): the indicator is recruited as a cause of the behaviour because otherwise the behaviour does not systematically yield a reward. Here I argue that natural information can be explanatory even without behaviour generating a reward (furthermore, unlike instrumental learning, associative learning is disanalogous to evolution by selection in that no fitness-analogous benefit is required). Fourth, the internal indicator’s carrying information is explanatory, for Dretske, because it has the function to indicate (it was recruited to prompt the behaviour because it indicates its success condition). By contrast, the information carried by an environmental condition that is exploited for associative learning is explanatory despite the condition lacking the biological function to be predictive about another event.
There may be other reasons for tying information to tokens, e.g. in order to capture scientific usage of the term ‘information’.
Learned dispositions can be lost over time, as a referee rightly observed. But this fact does not undermine Dretskean entitlement as long as the inferential disposition obtains.
In a manuscript under review, which I read after writing the present paper, Scarantino develops a detailed account along these lines.
“The behaviour of the birds during experimental stage 2 shows that hummingbirds can encode and retrieve the spatial position within their environment after a single experience. […] That a rufous hummingbird, while visiting a rewarded flower for the first time, encodes the spatial location and then remembers this information to revisit the flower is a remarkable phenomenon when viewed in an ecological context” (Flores-Abreu et al. 2012, p. 635, my emphasis). Note that the authors do not identify the precise nature of A; presumably the hummingbird uses a complex set of landmarks.
Laska and Metzker (1998, p. 193) mention in their introduction that “nongustatory modalities may convey important food-related information” (my emphasis). They then summarise the results of their study by saying that “both squirrel monkeys and common marmosets [are] able to reliably form associations between visual or olfactory cues of a potential food, in the absence of gustatory cues, and its palatability” (p. 198). Thus, the authors take themselves as having shown that non-gustatory cues, which in the introduction they describe as being able to carry food-related information, can serve as the basis of one-trial associations with palatability.
“We positively condition neonatal mice to associate arbitrary odorant CSs with a suckling/milk US […] and show that one-trial learning by this method results in conditioned odor preferences for these odorants.” (Armstrong et al. 2006, p. 344).
Rozin (1986, p. 185) describes his findings on single-trial acquired taste aversions as a “new instance of Pavlovian conditioning in which a disgusting stimulus is the US and a food or other object is the CS.” Furthermore, Rozin regarded the results as being compatible with a cognitive interpretation of associative learning: “We have framed the disgust pairing results reported in this paper or by Rozin (1986) in conditioning terms. This is consistent with, or equivalent to, a formulation in terms of the association of ideas, in which the sight of the food gives rise to a disgusting image” (p. 186).
For example, Rescorla (1988) contrasts contiguity and contingency in informational terms: the “modern view of conditioning as the learning of relations sees contiguity as neither necessary nor sufficient. Rather, that view emphasizes the information that one stimulus gives about another. […] conditioning depends not on the contiguity between the CS and the US but rather on the information that the CS provides about the US” (p. 152 and 155, respectively, my emphasis).
My thanks to Mark Sprevak for raising these cases.
A reviewer suggested that single-trial learning can be construed as involving probability-raising if one allows either subjective probabilities or propensities of entire chance set-ups (rather than of individual events). However, subjective probabilities sit ill with the naturalistic aspirations of theories of probabilistic information, and chance set-ups do not circumvent the problem of inverse conditional probabilities.
This alone is still insufficient to argue that ‘information’ picks out probability-raising because the dance raises the probability of other facts that are not attributed as the information content of the dance (e.g. it raises the probability that worker bees will depart towards location X). Some explanation is needed as to why these facts are not attributed as information contents. Perhaps ethologists make a distinction between the information a signal carries and the information it conveys to some receiver (the standard distinction between signal ‘message’ and ‘meaning’ points in this direction). Alternatively, ‘information’ may be reserved for those correlated facts that ethologists regard as explanatory.
One reviewer also suggested that ethologists might well accept that vervet alarm calls carry a diverse set of probabilistic information, if only it were put to them, and that they might conclude that this is how they use ‘information’ themselves. The latter result is unlikely given that in this case ethologists use ‘information’ to pick out the content of a mental representation. But even if they responded as suggested, it is far from clear that this procedure would be a suitable methodology for revealing an information concept that ethologists actually employ. Rigorous and extensive discussions of information concepts are rare in the animal communication literature, a fact which has caused disquiet among some practitioners. The lack of rigour could therefore significantly bias the requested self-assessment in favour of any well-articulated and prima facie plausible information concept.
References
Alexander, J. J., Audesirk, T. E., & Audesirk, G. J. (1984). One-trial reward learning in the snail Lymnea stagnalis. Journal of Neurobiology, 15, 67–72.
Altschul, J. (2011). Reliabilism and brains in vats. Acta Analytica, 26, 257–272.
Armstrong, C. M., DeVito, L. M., & Cleland, T. A. (2006). One-trial associative odor learning in neonatal mice. Chemical Senses, 31, 343–349.
Aslin, R. N., & Newport, E. L. (2012). Statistical learning: From acquiring specific items to forming general rules. Current Directions in Psychological Science, 21, 170–176.
Chater, N., Goodman, N., Griffiths, T. L., Kemp, C., Oaksford, M., & Tenenbaum, J. B. (2011). The imaginary fundamentalists: The unshocking truth about Bayesian cognitive science. Behavioral and Brain Sciences, 34, 194–196.
Chater, N., Tenenbaum, J. B., & Yuille, A. (2006). Probabilistic models of cognition: Conceptual foundations. Trends in Cognitive Sciences, 10, 287–291.
Cohen, J., & Meskin, A. (2006). An objective counterfactual theory of information. Australasian Journal of Philosophy, 84, 333–352.
Courville, A. C., Daw, N. D., & Touretzky, D. S. (2006). Bayesian theories of conditioning in a changing world. Trends in Cognitive Sciences, 10, 294–300.
Demir, H. (2008). Counterfactuals vs. conditional probabilities: A critical examination of the counterfactual theory of information. Australasian Journal of Philosophy, 86, 45–60.
Dretske, F. (1981). Knowledge and the flow of information. Cambridge, MA: MIT Press.
Dretske, F. (1988). Explaining behavior: Reasons in a world of causes. Cambridge, MA: MIT Press.
Dretske, F. (2000). Entitlement: Epistemic rights without epistemic duties? Philosophy and Phenomenological Research, 60, 591–606.
Flores-Abreu, I. N., Hurly, T. A., & Healy, S. D. (2012). One-trial spatial learning: Wild hummingbirds relocate a reward after a single visit. Animal Cognition, 15, 631–637.
Godfrey-Smith, P. (1992). Indication and adaptation. Synthese, 92, 283–312.
Haley, T. J., & Snyder, R. S. (Eds.). (1964). The response of the nervous system to ionizing radiation. Boston: Little, Brown & Co.
Jones, M., & Love, B. C. (2011). Bayesian fundamentalism or enlightenment? On the explanatory status and theoretical contributions of Bayesian models of cognition. Behavioral and Brain Sciences, 34, 169–231.
Kraemer, D. C. (2013). Against “soft” statistical information. Philosophical Psychology. doi:10.1080/09515089.2013.785127.
Lange, M. (2000). Natural laws in scientific practice. Oxford: Oxford University Press.
Laska, M., & Metzker, K. (1998). Food avoidance learning in squirrel monkeys and common marmosets. Learning & Memory, 5, 193–203.
Millikan, R. (2000). What has natural information to do with intentional representation? (appendix B). In R. Millikan (Ed.), On clear and confused ideas (pp. 1–18). Cambridge: Cambridge University Press.
Millikan, R. G. (2004). The varieties of meaning. Cambridge, MA: MIT Press.
Millikan, R. (2007). An input condition for teleosemantics? A reply to Shea (and Godfrey-Smith). Philosophy and Phenomenological Research, 75, 436–455.
Millikan, R. (2013). Natural information, intentional signs and animal communication. In U. Stegmann (Ed.), Animal communiction theory: Information and influence (pp. 133–146). Cambridge: Cambridge University Press.
Neander, K. (2006). Content for cognitive science. In G. Macdonald & D. Papineau (Eds.), Teleosemantics: New philosophical essays (pp. 167–194). Oxford: Oxford University Press.
Pearce, J. M. (2008). Animal learning and cognition: An introduction. Hove: Psychology Press.
Piccinini, G., & Scarantino, A. (2010). Computation vs. information processing: Why their difference matters to cognitive science. Studies in History and Philosophy of Science, 41, 237–246.
Piccinini, G., & Scarantino, A. (2011). Information processing, computation, and cognition. Journal of Biological Physics, 37, 1–38.
Rescorla, R. A. (1988). Pavlovian conditioning: It’s not what you think it is. American Psychologist, 43, 151–160.
Rozin, P. (1986). One-trial acquired likes and dislikes in humans: Disgust as a US, food predominance, and negative learning predominance. Learning and Motivation, 17, 180–189.
Scarantino, A. (2008). Shell games, information, and counterfactuals. Australasian Journal of Philosophy, 86, 629–634.
Scarantino, A. (2013). Animal communication as information-mediated influence. In U. Stegmann (Ed.), Animal communication theory: Information and influence (pp. 63–81). Cambridge: Cambridge University Press.
Scarantino, A., & Piccinini, G. (2010). Information without truth. Metaphilosophy, 41, 313–330.
Seyfarth, R. M., & Cheney, D. L. (2003). Signalers and receivers in animal communication. Annual Review of Psychology, 54, 145–173.
Shea, N. (2007). Consumers need information: Supplementing teleosemantics with an input condition. Philosophy and Phenomenological Research, 75, 404–435.
Shettleworth, S. J. (2010). Cognition, evolution, and behaviour. Oxford: Oxford University Press.
Skyrms, B. (2010). Signals: Evolution, learning, and information. Oxford: Oxford University Press.
Stegmann, U. E. (2013). A primer on information and influence in animal communication. In U. E. Stegmann (Ed.), Animal communication theory: Information and influence (pp. 1–39). Cambridge: Cambridge University Press.
Suppes, P. (1983). Probability and information (open peer commentary). The Behavioral and Brain Sciences, 6, 81.
Turk-Browne, N. B., Scholl, B. J., Chun, M. M., & Johnson, M. K. (2008). Neural evidence of statistical learning: Efficient detection of visual regularities without awareness. Journal of Cognitive Neuroscience, 21, 1934–1945.
Acknowledgments
Andrea Scarantino, Nicholas Shea, Mark Sprevak, and three anonymous referees provided incisive and constructive comments, for which I am very grateful. In 2012, earlier versions of this paper were delivered in Edinburgh, at the Joint Session in Stirling, and at a workshop on natural information in Aberdeen. I thank participants for their feedback.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Stegmann, U.E. Prospects for Probabilistic Theories of Natural Information. Erkenn 80, 869–893 (2015). https://doi.org/10.1007/s10670-014-9679-9
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10670-014-9679-9