What Can Artificial Intelligence Learn from Wittgenstein’s On Certainty?

Abstract

Meta-philosophically speaking, the philosophy of artificial intelligence (AI) is intended not only to explore the theoretical possibility of building “thinking machines,” but also to reveal philosophical implications of specific AI approaches. Wittgenstein’s comments on the analytic/empirical dichotomy may offer inspirations for AI in the second sense. According to his “river metaphor” in On Certainty, the analytic/empirical boundary should be delimited in a way sensitive to specific contexts of practical reasoning. His proposal seems to suggest that any cognitive modeling project needs to render the system context-sensitive by avoiding representing large amounts of truisms in its cognitive processes, otherwise neither representational compactness nor computational efficiency can be achieved. In this article, different AI approaches (like the Common Sense Law of Inertia approach, the Bayesian approach and the connectionist approach) will be critically evaluated under the afore-mentioned Wittgensteinian criteria, followed by the author’s own constructive suggestion on what AI needs to try to do in the near future.