Naive Causal Modeling, Vol. 1: Forward Causation, AI Applications, and the New Backward Causation

Citation

Andrew, A.M. (1998), "Naive Causal Modeling, Vol. 1: Forward Causation, AI Applications, and the New Backward Causation", Kybernetes, Vol. 27 No. 8, pp. 973-975. https://doi.org/10.1108/k.1998.27.8.973.1

The biographical note shows that this book is the fruit of a project the author has had in mind for a long time, and which he has completed, or at least taken as far as this first volume, following his retirement from employment as a computer engineer at the Johnson Space Center of NASA.

As is illustrated elsewhere in the AI literature, “causation” is a difficult topic. For a specific event, the causal explanation that first comes to mind is often one that can be used to assign credit or blame, and which indicates some deviation from an assumed “normal” situation. The cause of a car accident, for example, could be said to be that the driver was going too fast, but deeper analysis could take account of technical developments that made such speed possible, and to psychological and evolutionary factors that prompted the driver to want to go fast, and to prehistoric events that caused a ready supply of high‐energy fossil fuel to be available, and so on.

This book, however, is not concerned with these interminable ramifications that can be thought up for real‐life events; its concern with relatively‐direct causation seems to be the intended implication of the word “naive” in the title. Even so, there are plenty of other ramifications to be treated, and philosophical views of the nature of causation are comprehensively reviewed, with references to Hume, Schopenhauer, Russell, Wittgenstein and others including well‐known contributors to AI. Much is made of the distinction between, on the one hand, “regularity” assertions of the familiar type of “if A then B”, with their associated ideas of sufficiency and necessity, and on the other hand “counterfactual” assertions of the type “if A were so, then B would follow”. It is argued that the two forms of assertion are not equivalent, a view that is supported by a substantial number of items in the quoted literature which treat “counterfactuals”.

One of the conditions commonly associated with causality is that a cause must precede its effect in time, or may possibly be contemporaneous. Bean is, interestingly, ready to reject this, partly because of arguments stemming from quantum theory, and also because there are certain systems, notably those of Newtonian mechanics, in which time can be reversed. It does not seem right to have to relabel causes as effects, and vice versa, in the time‐reversed versions.

In the first nine chapters a formal treatment of causation is presented, starting from simple and uncontroversial expressions, but rapidly becoming difficult to follow. A degree of mystification is discernible in the two extracts from reviews quoted on the dust cover, in which a well‐known worker in the AI field comments “… an original approach …”, and the author of a previous treatment, quoted in the book, assures us that the present author is sound in his coverage of the relevant literature and refers to “… some thought‐provoking ideas …”.

In his final chapter the author positively rejects the need for temporal precedence of cause over effect, and argues for “backward causation”, claiming that it is implicit in modern physical theories. One piece of evidence is the set of hypothetical experiments often denoted by the letters EPR, indicating a paper by Einstein, Podolsky and Rosen as early as 1935. The experiments appear to imply instantaneous “spooky action at a distance” when the spin of one of two particles is measured.

There are, of course, many treatments which suggest that time is somehow different from the “ever‐rolling stream” that tends to be assumed in what is termed in AI “naive physics”. Although it is not exactly the argument used by Bean, a departure from the usually‐assumed temporal relationship of cause and effect is implicit in the “many‐worlds” view of quantum mechanics, in which a large number of possible worlds is believed to co‐exist following an event, until an observation forces a choice ‐ the view that is illustrated by the parable of Schroedinger’s unfortunate cat.

Some consequences of the many‐worlds viewpoint for computation are spelled out in a recent paper (Anonymous, 1998). In his Introduction, Bean relates his arguments to theories about time inside black holes and to the possibility of wormholes linking distinct times, with the intriguing possibility of time travel. He also introduces the difficult concept of a “variable past”.

The “new backward causation” mentioned in the title refers to this aspect of modern physics, and not to the kind of teleology that became respectable with the advent of cybernetics, namely the description of behaviour in terms of goals and feedback. The kind of teleology implicit in cybernetics is generally held to be warranted because it provides the most natural and satisfying explanation of behavioural phenomena, with a tacit assumption that a relatively complex non‐teleological account is also possible.

However, although the author goes well beyond the bounds of what is understood in the AI and robotics context by “naive physics”, his treatment, except for the final chapter, is intended to find applications in these areas. The dust cover emphasises this with a picture of a thoughtful‐looking android.

The reader interested only in applications is invited to read only a few small sections of the main text before turning to Appendix B, which describes three applications of robot devices forming internal causal models from experience. One of the examples refers to a confusing situation in which the learning robot is an assistant to an experimenter establishing a conditioned reflex in a dog, and the other two refer to rather simpler environments. There is a general principle of operation that certainly deserves consideration, depending on interaction between two databases maintained by the robot, one referred to as DBA, or Database of Actuals, and the other as DBH or Database of Hypotheticals.

The book thus has two distinct aims, since it has relevance to physics and philosophy of science as well as to practical AI and robotics. In both areas the treatment is innovative and controversial, and certainly deserving of attention.