Randomness: the Ghost in the Machine? (2014) | Hacker News Reader

Randomness: the Ghost in the Machine? (2014)

3quarksdaily.com

56 pointsprobe32 comments

Just want to add that biological neurons are stochastic and that artificial neural nets rely on noise injection, random connection dropout, random initialisation and random mini-batch reshuffling to work well. GANs start from a random number in generating images or other types of data. Alternative optimisation strategies to gradient descent - such as evolutionary strategies, use randomness in combining genes for the next generation of agents.

Without randomness we would have no intelligent artificial agents. It's like intelligence can only happen at the demarcation line between chaos and order.

> Without randomness we would have no intelligent artificial agents. It's like intelligence can only happen at the demarcation line between chaos and order.

I doubt most NN's are trained using true randomness. Most likely a fairly simple deterministic generator works just fine.

> you are not your body, you are not your brain, you are not your genome, you are not your connectome, you are not what you eat, you are not your social network, you are not your taste in music, you are not your history

This totally reminded me of the Chuck Palahniuk quote from Fight Club: “You are not your job, you're not how much money you have in the bank. You are not the car you drive. You're not the contents of your wallet. You are not your fucking khakis. You are all singing, all dancing crap of the world.”

This article covered a lot of ground, I quite enjoyed it. It made me think there’s a big irony, or maybe it’s a paradox.

We need the existence of true randomness in physics in order to allow for free will. And yet willfulness is the exact opposite of randomness.

I never really understood why randomness would be needed for free will. If that's the only thing keeping the will "free" would it even matter if our decisions are made deterministically or by a dice roll?

Guess I have some reading to do.

I think the point is that if there's no randomness, people's actions must be completely predictable, and if they're predictable it's impossible for people to do other than what's been predicted and so there's no free will.

On the other hand, having randomness opens up the possibility to act in ways that are not completely predictable and that's where the potential for freedom of the will comes in. But randomness alone isn't enough for free will, for the reason you allude to (that one might be "forced" to follow some sort of cosmic "dice roll", as unpredictable as it might be). It just seems to be one pre-requisite, in a certain way of viewing the problem.

But the universe is turing complete, so the only way to predict what will happen is to run a perfect simulation of everything in the universe. This is clearly impossible within the universe, and so your actions remain unpredictable.

This line of thought applies equally well to every individual hydrogen atom in the universe as it does to a person, yet it does not follow that they also have free will. Where OP says 'predictable', I think he means 'determined' or 'pre-ordained'.

But since you cannot differentiate between predictable and preordained, what is the difference?

That something is not predictable does not mean that it is not pre-ordained, even if one cannot tell the difference. Classical mechanics is a model that is fully deterministic yet ultimately unpredictable, and similarly, syyrim's model of the universe as Turing-complete seems to assume it is pre-ordained (a model that goldenkey challenges), and yet ultimately unpredictable.

On re-reading the thread, I am not sure what point syyrim was making. The ultimate unpredictability of the universe doesn't seem to settle any issue with regard to free will. Free will is not a claim that we cannot predict what we will do; it is a claim that we can make a choice (within some constraints) about what we will do.

Not even. If you take Heisenbergs uncertainty principle to be true, then we can never truly know the full state of any matter. So even if we knew the rules, we wouldn't have the seed. This is also known as the knowledge balance principle. See link below:

https://en.m.wikipedia.org/wiki/Spekkens_toy_model

While I have seen explanations of how randomness might lead to the illusion of free will, I have not seen any line of reasoning that explains free will itself as being dependent on randomness (because I have not seen any explanation of free will, period.) My guess is that randomness is imagined to have a role in free will only because it is the alternative to strict determinism (which is, as you point out, completely antithetical to free will) rather than because it is clear that randomness does, in fact, enable free will.

It certainly feels to me that I have free will, but the older I get, the less sure I am that I actually do. Consequently, I have no idea how I came to write this post.

> I never really understood why randomness would be needed for free will

For some reason it had to be pointed out to me (awhile back) that being ruled by the rigorous iron fist of science's "yes technically non-deterministic but totally statistically modelable" quantum dice rolls isn't necessarily much better, if any, than being ruled by the rigorous iron fist of deterministic cause and effect.

Ruled by chance or by force of sheer causality alone - at this point, I feel that any useful definition of "free will" probably equally applies (or doesn't) to both scenarios, and won't hinge on quantum unpredictability.

There are two senses in which we mean "free will". The first is physical non-determinism - "My actions are not decided by deterministic processes", which is absurd. The second, is psychological - "I select my actions for internal reasons". That's trivial, no mystery there.

Maybe when we think about free will we actually refer to cooperation-betrayal strategies between people (prisoner's dilemma) which are essential in the functioning of societies. We want to see ourselves as free to choose our strategy in response to other agent's actions. "If they betray me, I have the choice between cooperation and retaliation" - that's what we feel as free will that counts.

Determinism removes internal/external separation. They commute when both are deterministic. Consider deterministic system A and deterministic system B in a deterministic environment C. Then the entire system is deterministic and though the situation might not have played out yet, it essentially was decided for the participants. Non determinism is the best free will can be -- neither nature knows the future, nor the participants. Everyone is blind. So the will is "free" with respect to control.

It's simple: true randomness is a back door for God. If there's a true source of randomness in the universe, and our brains make use of it, you can postulate that the soul is what magically makes this "random source" not random, hence enabling "free will" as religions see it.

I think for free will to exist, something neither totally random nor totally deterministic would be required. That’s why free will isn’t really a scientific concept.

The idea that there something non-deterministic give some hope that though it “look random” there actually order being directed by “something” there. Essentially if you’re a mind-body dualist, this is the point where mind meets body (would be my understanding).

> If that's the only thing keeping the will "free" would it even matter if our decisions are made deterministically or by a dice roll?

This is a good point, and relates directly to the paradox I see.

It's not that randomness is required for free will, or even desired, it's that all other known processes are deterministic. It's an age-old discussion that if the state of the universe and all the particles in it at any given moment is determined entirely by physical laws like conservation of energy, then, it is theorized, free will cannot exist because the future state of the universe can be predicted.

We may have the appearance of free will, but if our thoughts emerge from predictable phenomena, then autonomy and consciousness as we know it can only be a facade.

Quantum randomness is the only process we know of that can't be predicted by a sufficiently powerful simulation. Therefore, quantum randomness is what some people rely on to posit the existence of free will, only because it's the only thing left by process of elimination.

Your point is similar to my question, namely that even if there is randomness, we still have a problem. What's willful about an unpredictably chance event?

Interesting how the concept of randomness can only exist in contrast to the recognition of order, but who notices order other than a mind that defines it as such? But this mind could not have been created in the first place were it not for some intrinsic order, some consistency in reality. Makes my head spin...

> But this mind could not have been created in the first place were it not for some intrinsic order

The mind was created by self replication and competition over multiple generations (evolution). Self replication requires ability to adapt to environment, to collect energy and materials, to protect from predators. This was the playground where minds evolved. We're using similar techniques in AlphaGo, for example, where a sequence of players are pitted against each other.

Randomness is essential for exploration, which is a step without which we couldn't be intelligent. In reinforcement learning, the so called epsilon-greedy strategy of exploration says to select a random action with probability epsilon. It is a simple strategy that leads to intelligence. Of course, humans have smarter exploration strategies which nonetheless rely on randomness.

"Free Will" as a concept just makes no sense. I think it's obvious that any system can be conceptually described as a function of all-inputs-that-affect-it, plus randomness. No room for free will.

The only thing free will has going for it is the unmistakable, unshakeable feeling that I really have it.

And that feeling probably just means: "I can't reliably predict myself." Which is more about one's ability to predict, than about some intrinsic unpredictability.

I'm not familiar with the inner workings of random number generators but is true randomness even possible or is the randomness just so complex that we can't reverse engineer it?

My instinct says that if you run the same random number generator in the same hardware with the same Unix timestamp and same inputs wouldn't the outputs be the same?

Yes, true randomness is possible through quantum phenomenon. No, it isn’t possible to generate true randomness using deterministic processes (eg, almost all computational ones).

Determinists would ask, is the quantum phenomenon just ignorance by current lack of understanding to what humans can possibly know with the tools we have now?

True randomness is like believing in god if your life is good or praying to a god if your life is bad.

According to the Copenhagen interpretation of quantum mechanics, there are no local hidden variables and the world is truly random. Even knowing the complete state of a particle will not tell you how its wave function will collapse.

I am convinced God does not play dice.

If your random number generator is pure software, it’s repeatable.

People sell hardware random number generators that are based on Geiger counters, since radiation is (as far as we know) truly random.

I believe some (all?) new Intel processors have built-in hardware components which are a source of truly random numbers.

I just read about it, and it looks like a very strong and fast RNG.

https://spectrum.ieee.org/computing/hardware/behind-intels-n...

But, fwiw, this isn’t “truly” random in the quantum sense. It’s only difficult to predict as far as anyone knows, but is still built on deterministic processes. Radiation, on the other hand, is (as far as anyone knows) physically “truly” non-deterministic.

A similar thought experiment:

If you could somehow pause the universe, create an exact duplicate, then press "play" on both at the same time, would they remain in sync? Would their futures be the same?

If so, it implies that any randomness we think exists is actually deterministic but too difficult for us to currently understand.

If they diverge then randomness or free will do exist. But this doesn't explain what mechanism makes it possible.

Right. I suppose that would reveal if quantum fluctuations are truly random.

In a maximally efficient interactive system, user input appears as random noise. This is because such a system will contain a model of the distribution of its input in order to be able to predict it and react as soon as possible. The system only needs to react to the component of input which it was not able to predict using this model. Thus if there is an intelligent entity manipulating our universe, and if our universe is efficient in this sense, then the outside entity's input will appear as perfectly random maximum entropy noise when viewed from within the system.

Reply on news.ycombinator.com