Journal of Mind Theory

Posted on February 21, 2009


Ricardo Sanz, who is working on autonomous systems at the Universidad Politecnica de Madrid, is working on putting together a new journal to be called Journal of Mind Theory. The focus of this publication is to be on formal theories of mind. While in Madrid last year I had a chat with him about this project and was pleased when he suggested that I become a member of the editorial board. Given my own qualms about the limitations of formal models, I gather that I am to play something of a devil’s advocate role. Indeed, that seems to be the idea behind a debate between myself and Ricardo that is to be published in the first issue of JMT. To give a rough idea of what I will be saying there, I include below the very first question and my own answer to it:

First off, for the purpose of putting things in perspective, there seems to be some skepticism about the usefulness of formal approaches.  Is formal logic the best mode for thinking about mental processes?

Are the grounds of validity of the laws of logic to be found in language, in conceptual structures, in the nature of representation, in the world, or where?

KT: Given that I am supposed to present the limitations of formal methods I should probably begin by making it clear that I do not think that any serious inquiry into cognition can go far without the use of such tools. Formalization is a vital element of the approach used by scientists to understand the world, allowing us to obtain a precise grasp of natural phenomena, as well as revealing to us when we lack such understanding. Any attempt to think about the mind without recourse to formal tools would be unlikely to get far beyond a collection of insights. Having said that, I think it fair to say that for much of the previous century assessments of the value of formalization have been overly hubristic. The depressing longevity of GOFAI seems to me one aspect of this phenomenon. In short, my position is not so much skeptical as pluralist – formal tools are necessary but far from sufficient. And they carry with them numerous problems.

The utility of formal tools for investigating mental processes will look very different depending upon the context in which one places those processes. One can see them as the imperfect implementation of reasoning strategies, the structure of which is investigated by logic and other formal approaches. One can also see them as the evolved means certain organisms use to direct their interactions with their environment. While the two views – the logical and the biological – are not necessarily contradictory they do start with very different kinds of assumptions. The biological view forces a bottom-up perspective in which mental capabilities are the result of evolutionary and developmental processes that began with nothing more than simple chemical replication. On this view cognitive limitations are not an unfortunate detail to be abstracted away from but are at the heart of how we’ve managed to incrementally transcend those limitations by making efficient use of the limited resources we did have access to. Also, biological solutions are not optimal nor are they universal. Instead, they are at best adequate to current needs in the particular environment they are usually applied. If one uses logic as a model of what mental processes are meant to be like one ends up with highly unrealistic assumptions, such as a deductively closed set of beliefs, that must be continuously fought against if one is to arrive at something plausible (Brown 1990, Hooker 1995, Bickhard 2009). The point isn’t that all formal models must embody those assumptions. They do not. However, to avoid having to identify them as problematic one by one, it is far preferable to start from the point of view that mental processes are to be understood as biological phenomena.

It might be argued that while the logical view does a poor job of description, it is primarily meant to be normative – it talks not of what human mental processes are like but of what we should be trying to make them like. Yet this move does not buy the freedom from human foibles one might desire. On the one hand, it is generally accepted that ought does imply can. This means that while the details of current limitations may be no longer as relevant, the overall unavoidability of limitations is not. On the other hand, putting the view in normative terms raises the question of the relevance of the norms. Why should people try to have true beliefs, for example? The fall back claim that this is simply what it means to be rational solves nothing unless one is comfortable with the conclusion that the choice to be rational is arbitrary. Any substantive naturalist answer, however, will have to come back to the human predicament of needing to make our way in a world whose capacity to affect us exceeds our understanding of it.

The reservations I have raised may seem to be merely grounds for a view of reasoning that acknowledges the relevance of both the biological grounding and the logical structure. The two are not on a par, however. The problems with the logical view of human cognition can be ultimately traced back, I would argue, to Hume’s old problem: a problem that I see as necessitating the naturalist (rather than scepticist) response that there can be no universal solution to the problem of how to come to grips with our world. While the problem has come to carry the name of the problem of induction it might be much better titled the problem for induction, given that it was always clear, pace Popper (1959), that deduction had nothing useful to say on the topic. What is worse, as Couvalis (2004) points out, the problem also affects our justification for using deductive arguments since our use of these must presume our ability to use them is reliable, and evidence for this must be inductive. So long as induction was conceived of as a logic, no useful response could be given to the problem. In the end, the trick has turned out to be not to seek a solution by proposing ever more complex logics but to learn to live with the problem – “The Humean predicament is the human predicament”, to quote Quine (1969: 72). Without an overall framework to work within we are left muddling along in the best of biological fashion. In our efforts we are free to make use of whatever tools we can access and, undoubtedly, formal tools are among the most useful. However, they are only made use of within the broader biological context.

In talking about this way of seeing cognition one is obliged to bring up Herbert Simon. Through his collaboration with Alan Newell, Simon can be deemed to be one of the father’s of GOFAI (Newell, Simon 1976). Throughout his life he made brilliant use of numerous formal methods to model aspects of human mental processes. At the same time, however, his bounded rationality approach brings together the biologically-informed points I have been seeking to make (Simon 1983). Something of the significance of the view of rationality he provides us with can be seen in the ensuing disagreement between two groups of researchers who adopted his concept of heuristics: Kahneman and Tversky (Tversky, Kahneman 1974) on one hand and Gigerenzer and his colleagues on the other (Gigerenzer, Todd, ABC Research Group 1999). Not surprisingly, I stand with Gigerenzer in claiming that Kahneman and Tversky failed to appreciate Simon’s overall position when they used statistics as the standard they compared human cognitive heuristics to instead of examining how effective the two are when dealing with real-life data. In effect, I see Gigerenzer’s work as a very good example of just what can be done using formal tools within an overall biological view of human mental processes.

Where does this leave the laws of logic? Clearly, I can not hold that they are the laws of reasoning. Hume put that notion to rest, I think: even if many did not notice and insisted on exhuming it. Yet, I have no wish to claim laws of logic to be a human construct. Having said that I must own up to a certain degree to fascinated ignorance as to what their actual nature is. One insight I do find convincing in this context is Peirce’s (1905) definition of what is real as that which has properties that are independent of what we think of that thing. Thus, the Earth’s equator is real in this sense, as its length is roughly 40 thousand kilometers independently of what we think about it.