William Ratoff

Cognitivism about practical reason is the doctrine that certain aspects of practical reason are really instances of theoretical reason. For example, that intentions are beliefs or that certain norms of practical rationality just are, or reduce to, certain norms of theoretical rationality. Radical cognitivism about practical reason, in contrast, is the more heady view that practical reason just is a species of theoretical reason. It entails that what it is to be a motivational state (of any kind) is to be a belief (of some particular kind), that all practical reasoning is just a variety of theoretical reasoning, and that all practical norms are really a kind of epistemic norm. In this paper, I introduce and motivate this view, before investigating the cognitivist theory of intention and means-end practical reasoning to which such a radical cognitivist is committed. Unlike other cognitivists about practical reason, the radical cognitivist reduces all practical reasoning to theoretical reasoning and all practical norms to epistemic norms. She therefore faces unique challenges in accounting for the basic desiderata on any adequate theory of intention and means-end practical reasoning: whereas other cognitivists can appeal to_ sui generis _practical states (desires) and norms, the radical cognitivist is restricted to the sparse resources – cognitive states and the epistemic norms that govern them – to which she has restricted herself. Here I develop and critically examine the options the radical cognitivist has for satisfying these desiderata. My conclusions are optimistic: there is a defensible formulation of the radical cognitivist’s theory of intention and means-end practical reasoning.

A new theory of the mind, the predictive processing model, is ascendant in recent work in cognitive science. According to this theory, all the mind ever fundamentally does is make hypotheses about the environment, generate prediction-errors by comparing its predictions with its sensory data, and use these prediction-errors to update its representation of the world. The theory of motivation and action to which the predictive processing model is committed has been the subject of lively debate in the literature. However, the upshots of the predictive processing model for the theory of practical reason have received relatively scant attention. Here I investigate how proponents of predictive processing should conceive of practical reason. The predictive processing model has it that the practical attitudes – desires and intentions etc. – reduce to certain theoretical attitudes – in particular, to beliefs and predictions. I argue that this psychological reduction precipitates a broader normative reduction, in which practical reasoning reduces to a kind of theoretical reasoning about what you will do, and in which practical normativity turns out to really be a species of epistemic normativity. In other words, that practical reason reduces to theoretical reason. If I am correct, the predictive processing model of the mind is committed to – what I am dubbing – a ‘radical cognitivism about practical reason’.

How do we ensure that future generally intelligent AI share our values? This is the value-alignment problem. It is a weighty matter. After all, if AI are neutral with respect to our wellbeing, or worse, actively hostile toward us, then they pose an existential threat to humanity. Some philosophers have argued that one important way in which we can mitigate this threat is to develop only AI that shares our values or that has values that ‘align with’ ours. However, there is nothing to guarantee that this policy will be universally implemented—in particular, ‘bad actors’ are likely to flout it. In this paper, I show how the predictive processing model of the mind, currently ascendant in cognitive science, may ameliorate the value-alignment problem. In essence, I argue that there are a plurality of reasons why any future generally intelligent AI will possess a predictive processing cognitive architecture (e.g. because we decide to build them that way; because it is the only possible cognitive architecture that can underpin general intelligence; because it is the easiest way to create AI.). I also argue that if future generally intelligent AI possess a predictive processing cognitive architecture, then they will come to share our pro-moral motivations (of valuing humanity as an end; avoiding maleficent actions; etc.), regardless of their initial motivation set. Consequently, these AI will pose a minimal threat to humanity. In this way then, I conclude, the value-alignment problem is significantly ameliorated under the assumption that future generally intelligent AI will possess a predictive processing cognitive architecture.