Comment by hacker42 | Hacker News Reader

Comment by hacker42 | Hacker News Reader

Parent Full thread

I liked this interview with Judea Pearl on free will: https://youtube.com/watch?v=sg7Oq4suH_E

The summary is basically that to our best knowledge there is no free will because decisions are caused by neural activities which in turn are caused by sensory input and noise (but unlikely by quantum noise). However, humans have evolved a strong sense of agency because that's simply an efficient way to reason about machines that produce actions in response to the entirety of their sensory input (especially regarding parent child relationships and mutual behavior correction). This neuroarchitectural bias is essentially an illusion of free will that is so firmly wired into our brains that we cannot escape it. It is also the reason why the idea of a God comes so intuitively to many of us: an invisible actor which can be used to explain inexplicable chains of causation and can serve as a very effective metaphor for behavioral error correction (as a proxy for actual social repercussions, and hence relieved from all the complicated and hence fallible power relations to actual social error correction instances).

How does seeing a human as "neural activities which in turn are caused by sensory input and noise" contradict an idea of free will? The neural activity is obviously so complicated that it can make "sense" of the sensory input and the noise, and make intelligent decisions based on past experience, on it's own generatlizations of experience, on external ideas, etc. etc.

There are obviously some mechanisms for creating new information and action, based on past experience as well (thus also creating new unforeseen behavior). These mechanisms can clearly be implemented on the neural machine of the brain, since it's evident that it can even already be approximated in Google Deep Dream to create new unforeseen images based on previous inputs.

Whether the human can always verbally describe the decision tree (or whatever other decision mechanism is used), is another question. But even if it cannot describe, so what? The decision is done somewhere deep in the net, and the verbal processor does not have access to it. It's still the network making a decision...

So what makes you say that free will is an "illusion"? Our brains obviously soak up the information and then make future decisions based on that information (subject to effectiveness of learning, etc...).

That quote makes brains sound like deterministic machines. Which would contradict free will.

The notion that determinism leads to seeing the world in terms of simple predictable clockwork universe is obsolete. Look up "chaos theory". It describes how even very few deterministic rules applied to a vast number of elements can very quickly produce chaotic, unpredictable and non-deterministic results. Especially if those rules include feedback loops. BBC has a very good documentary on this called "The secret life of chaos".

To summarize, given all this new informatio: no, deterministic machines do not contradict free will. Because those machines are intelligent and have feedback loops and can (deterministically) make intelligent decisions based on the information.

For some frankly ridiculous definition of free will, yes. I think definitions of free will that this assumes - impossible to predict from pre-existing conditions - is synonymous with random choices. But random choices don't seem to be very free. Where is the agency?

I am the pre-existing condition that (largely) determines my choices. That's what makes my decisions mine and not just 'free' decisions devoid of context, responsibility or attribution.

So if my decisions are made by me and are not coerced or biased by limited access to information, then they are mine and they are free and I will happily accept responsibility for them. But magically occurring decisions free of conditions and not influenced in any way by my actual mental state or faculties are simply not my decisions.

A construct that can react in only one way to any given set of inputs (including its internal state, etc.) intuitively doesn't have 'free will'.

A construct that can react in multiple ways to a single given set of inputs, but does so by combining them with some internal inputs which are non-deterministic but generally random in nature, also intuitively doesn't have 'free will'.

What you need to really satisfy the intuitive concept of 'free will' is some analytical agency, external to our physical reality, which affects the outcome in some purposeful way. So, basically, a 'soul'.

Of course, to move past the 'intuitive' sense we're gonna need to actually rigorously define 'free will', which is something that is curiously lacking in virtually all discussions of this kind of stuff.

> A construct that can react in only one way to any given set of inputs (including its internal state, etc.) intuitively doesn't have 'free will'.

Then intuition is wrong, as is often the case. Either the decision is random, or it is mine, or it is somebody or something else's. There can be combinations of those factors, and of course that is actually the usual case. In fact arguably in practice it's always the case.

> What you need to really satisfy the intuitive concept of 'free will' is some analytical agency, external to our physical reality, which affects the outcome in some purposeful way. So, basically, a 'soul'.

But any analytical agency is going to encapsulate state. Moving that outside our physical bodies is just kicking the can down the metaphysical road but doesn't actually solve anything.

If we want to invoke pure randomness, there's always Quantum Mechanics. In fact there's a post on the intersection of quantum mechanics and biology on the front page right now [0]. But of course random input doesn't seem very 'free' either.

My point is that 'free will' in the abstract isn't an agency. To have agency there must be an agent and agents have state. When we are talking about free will, we should be talking about the free will of the agent. To the extent that the agent made the decision without undue external influence then they have free will. Can we ever be free of ourselves?

Edit: byt +1, good post and I definitely concur with your last point.

[0] https://news.ycombinator.com/item?id=12314600

The illusion is actually measurable. For example when you cause activity in cortical motor neurons by direct stimulation, then the subject will move e.g. their arm involuntarily, but they will later convincingly explain that they've moved their arm on purpose because of an itch or because they just felt like moving their arm.

Think about it. Isn't there something curious about this (probably innate) way of representing actions that were initiated inside our brains?

There are more examples like these, e.g. in split-brain patients where the speaking side simply rationalizes whatever the non-speaking side does. All of that points toward the same idea: Our feeling of being an entity that can choose in each moment what to do next misrepresents what actually happens inside our brains. Under the hood, every thought and every movement can be tracked back to either noise or to a recurrent control program that in turn was shaped by a reward maximizing mechanism and by learned statistics of real-world dynamics. On top of that there is an episodic self-representation which basically tells a story about itself and which has a bias to represent autonomously moving objects such as the structure it is part of, by a free, self-initiated intentionality. That story it tells itself about itself can of course affect future actions, but again the way it does is fully contingent on the laws of nature, on genetics, on the individual life experience etc.

If someone stopped caring about their life after reading this because they are not in control of their thoughts and movements anyway, then even that is fully contingent on the experience of reading whatever I am writing right now and society might decide to contain said subject for everyone's safety (i.e. to ensure their continued flow of reward signals). This exposes the role of this innate feeling of free agency as a mechanism of behavioral control. We need this representation to be efficient at attributing certain outcomes to certain individuals so that we can correct their behavior.

But we don't need to give up on anything knowing this. We still can do blame attribution. Actually we can probably gain something by improving the representation of ourselves and agents in general: Everybody is deeply shaped by their individual experiences and often it is actually quite insightful to go on 'auto-pilot' and ignore our representational urge to be our own initiator for a while and just see what the recurrent circuitry in ours brain can come up with.

Intuition is subjective. A long time ago it was very un-intuitive that the earth was spherical.

For many people the physical description of the brain does actually have free will.

The point is to change people's intuition about free will, based on all this new information we've discovered in the past century.

"external to our physical reality" There is no need for that.

Is the world inside of "No mans sky" game "external" or "internal" to our physical reality? Soul is something that is run on the hardware of the brain, and thus can have different properties than the physical properties of neurons. It's virtual. No less real for that though.

> Soul is something that is run on the hardware of the brain

Well no, that's kind of the point. If it's run on the brain's hardware (wetware?) then it's just a physical machine following physical laws, and it's no more or less conscious and has no more or less free will than a computer.

As we understand the 'natural' world, there's no room for free will. Your three sources of data for choosing your state n+1 are your state n, your perceptions of the world around you, and maybe some truly random factor.

> has no more or less free will than a computer

Exactly. The only difference in our modern times is that a brain is orders of magnitude more complex than a computer. None of the scientists who are working on replicating the consciousness think that nowadays computers have comparable consciousness to a human. They all understand that the complexity needs to go up MANY times before we can talk about it. But it's still very clear that a consciousness (with free will and all other aspects of it) is definitely possible to have in a (future, much more advanced) computer.

> Your three sources of data for choosing your state n+1 are your state n, your perceptions of the world around you, and maybe some truly random factor.

That IS free will!

Perhaps you define it in some other way? If you say that "free will" is not possible if the decisions of such entity are based on some past experience (partly)? Well in that case there is no living entity that we know that have your definition of "free will", so what's the point of trying to recreate it? How would it even look?

Reply on news.ycombinator.com