Free Will Belief as a Consequence of Model-Based Reinforcement Learning | Hacker News Reader

Free Will Belief as a Consequence of Model-Based Reinforcement Learning

arxiv.org

2 pointsraindeer24 comments

>Put simply, we suggest that by applying the RL framework as a model for human learning it becomes evident that in order for us to learn efficiently and be intelligent we need to view ourselves as if we have free will.

Granting the authors' assumption that we don't in fact have free will, it seems very difficult to argue that we need a false belief in order to learn efficiently and be intelligent. If it's possible to give a coherent explanation of why the false belief is helpful, then that's sufficient to indicate how we could do without the false belief.

Indeed, this is exactly the point.

The argument is not that we need a false belief, but rather that we need to view ourselves as free on one level to learn efficiently, even if we know we are not free on another level of abstraction. Most ppl mix up these two level of abstraction though.

Hmm, that’s not how I read the paper. It says:

> summary, we argue that, in order to efficiently learn from our successes or mistakes, we need to imagine that we are free. This imagined freedom is physical freedom in the simulated reality of our world model.

I don’t understand why we would need to imagine that we are free in order to engage in certain patterns of behavior. Surely we could just engage in these patterns of behavior for the reason given: that they enable us to learn efficiently.

The paper seems to conflate reasoning counterfactually with imagined freedom, but I don't think the two are so tightly linked. For example, I can reason counterfactually about what would happen under alternate laws of physics, but this doesn't mean that I (necessarily) believe that these altered laws correspond to possible states of affairs.

Sure, but when you reason counterfactually about what would happen under alternate laws of physics, you somehow imagine that these laws where true. You run a simulation of an alternative reality (although you know it is just a simulation). But you need to run the mental simulation, i.e., you need to imagine.

Imagining something obviously does not mean that you necessary believe it is true, but the argument is that since ppl reason counterfactual about doing things they didn't do so naturally, they confuse this imagination with their ability to actually have done differently. If you don't believe you have free will in the physical sense, then you maybe don't do this mistake, but most ppl believe they have free will.

And even if you don't believe you have free will, it arguably feels like you are free to have done differently in a lot of situations (although this is ofc very subjective).

Reply on news.ycombinator.com