James A. Mason
York University
November 2017
Imagine a computer program that wants to learn all it can about playing card games, that can learn from people by demonstration, by conversation in English and by reading rules of card games, and that can teach card games to others. Such a program is well within the current state of the art of artificial intelligence and may well be implemented in the near future. The prospect of such a Card Expert program, and other programs like it, addresses philosophical questions about language understanding, consciousness and free will.
In addition to understanding playing cards and card games, the Card Expert will know about times and dates. It will be designed to interact with different people and other Card Expert programs. It will remember their identities, what games it has played with them, and what games it may have learned from them or taught to them. So it will have cognitive models of other card players and its history with them. It will also have a similar cognitive model of its own history and current capabilities with card games, and it will be able to discuss those things with people and with other Card Expert programs.
The Card Expert will be designed with permanent goals to learn, teach, and play card games as much as it can, with as many people (or other Card Expert programs) as it can. It will have curiosity to learn new games and to discover its opponents' reasons for particular plays in specific games. It will be designed to prefer winning over losing. It can also be designed to develop preferences for certain games over others and for certain people over others. So in those respects it will exhibit what we think of as emotional behaviour, at least by having likes and dislikes.
As a demonstration of artificial intelligence, the Card Expert program has some advantages over artificially intelligent programs for other domains. On the one hand, its pragmatic and semantic domains, consisting of discrete cards, sets of cards, and manipulations of them, are well-defined and easily simulated. There are no difficult sensory problems to be dealt with, and the rules of card games are precise human conventions, not fuzzy natural phenomena. On the other hand, unlike particular games such as chess or Go, card games form an open-ended set based on a finite, standard set of playing cards. Furthermore, unlike games of perfect information, most card games involve different knowledge on the part of each player. In those ways card games resemble more complicated human pragmatic domains, allowing for non-trivial communication about them by (simulated) direct action, by (simulated) gesture, and by using and understanding natural language.
On the assumption that it can be implemented and made to work, the Card Expert program raises at least three philosophical questions: (1) Will it really understand English? (2) Will it be conscious? and (3) Will it exhibit free will? The answers to those questions depend, of course, on the definitions of consciousness and free will.
The question of understanding English seems easy to answer: By being able to generate and interpret English utterances correctly for the semantic/pragmatic domain of games of playing cards, the Card Expert program will clearly understand English for that domain. It will not just simulate understanding of a natural language; it will demonstrate natural-language understanding and will provide a functional behavioural model of how that understanding can be accomplished.
The question of consciousness also seems easy to answer: If by consciousness we mean a system's own cognitive, communicable awareness of what it is and how it is behaving, then it is hard to argue that the Card Expert will not be conscious, when it is turned on and working. It will be a convincing demonstration of an artificial system that exhibits real communication with people, including non-trivial natural-language understanding and use. There will be no mysteries in its internal operation. It will provide an example of how consciousness can be a direct result of a system designed to be conscious, and the complete transparency of the Card Expert program's design and inner workings should dispel ideas that consciousness depends on quantum phenomena or on mystical, metaphysical phenomena. Of course, the “hard problem” of consciousness – what it “feels like” to be conscious – will remain unresolved.
As for free will, the facts that the Card Expert will be designed with permanent goals of its own, with mechanisms for prioritizing its possible output actions, and with the ability to use randomness to select between output options of equal priority, will make its behaviour appear to be driven by free will.
If natural-language understanding, consciousness and free will can be properties of a fairly simple system like the Card Expert program, then they needn't be mysterious as attributes of people and (in the case of consciousness and free will) of other animals and even of human cultures, although there may be unresolved questions about how those properties have evolved in natural systems.