Hacker News Reader

Comment by TheOtherHobbes | Hacker News Reader

All of which is true. But I think there's a more subtle problem: if you could simulate evolving creatures in a universe, how could you tell if they were truly sentient?

Being sentient as an experience is completely different to appearing to show sentient behaviour. You can make a model of sentience that's as complicated as you can invent, and you'll still have no idea whether or not a modelled sentience is having a sentient experience of the universe around it, or whether it's faking its responses mechanically.

The corollary is that having a model isn't enough to simulate a universe, You also need some way of knowing when sentience has been achieved.

Nothing in our current technology is anywhere close to even beginning to understand how to do that.

View on HN

> Being sentient as an experience is completely different to appearing to show sentient behaviour.

You don't know this. All you know is that being sentient as an experience seems to be completely different from appearing to show sentient behavior.

> You can make a model of sentience that's as complicated as you can invent, and you'll still have no idea whether or not a modelled sentience is having a sentient experience of the universe around it, or whether it's faking its responses mechanically.

This is the standard "zombie" argument, and it makes no sense to me; it is tantamount to saying that the fact that you are sentient makes no difference to your externally observable behavior. Really?

> if you could simulate evolving creatures in a universe, how could you tell if they were truly sentient?

I think that AlphaGo is sentient in the domain of Go play. In other words, AG was conscious in that space of table positions and game strategies. A number of Go experts were so impressed that they said they could hardly tell which is the human and which is the computer. They gained so much respect for it that they awarded it the highest dan rating. I'd award it a Turing Test pass, at least on Go play.

To answer your question, I'd take the same approach as Turing: if it behaves like a human or better, then it is conscious in that particular domain. The difference from common machine learning applications was reinforcement learning - the fact that it was embedded in an external world, interacted with it and the opposing player and learned from experience - it behaves essentially like us, in a loop of perception-decision-action. And we are like it, if you look down the stack, we too are just a pattern of chemical reactions and electric currents that have no innate subjective feeling to them. There is no "unobtainium" or "magic fairy consciousness dust" in our brains either, and we shouldn't try to essentialize consciousness, it is not a pure essence like space and energy, it's an emergent quality.

Great post. I just wanted to add that space and energy could very much be emergent phenomenon, as well. Take for example Wolfram's conjecture that the universe might be modeled as a causal network [1], in which our 4D spacetime emerges from the connections between nodes, and energy/particles emerge from the complicated dances of knots in the network. It might sound pretty far fetched, but it seems to me at least as reasonable as the universe being modeled by a small set of equations.

[1] http://blog.stephenwolfram.com/2015/12/what-is-spacetime-rea...

> Being sentient as an experience is completely different to appearing to show sentient behaviour. You can make a model of sentience that's as complicated as you can invent, and you'll still have no idea whether or not a modelled sentience is having a sentient experience of the universe around it, or whether it's faking its responses mechanically.

If you make a brain that is physically identical to an extant human brain (let's say, your brain) then the only way you can claim that brain isn't sentient is if you claim that the original brain is also not sentient.

I mean, what do you think is going on in your head that doesn't count as "faking responses mechanically"?

That method gives an answer for one possible simulation- the experimenter's own brain. But it doesn't seem to suggest a means of determining what other simulations do or don't sustain sentience, so its far from a total solution.

When people suggest that science has things pretty much figured out by now, this sort of thing is one of the (many) counterexamples that I offer.

It seems to me that if you follow TheOtherHobbes logic to its conclusion, you arrive at the proposition "sentience is a supernatural phenomenon" which is an absurdity. The problem is that we have trouble agreeing on what sentience even is, so of course we can't agree how to measure it. But to suggest that sentience lies outside the natural world doesn't really answer any questions, either.

You're giving meaning the word 'sentience' by pattern-matching on your own experiences, so naturally if a computer can do the same it is no different from you in that regard.

Anyway, if one starts thinking the whole universe is a simulation, then that word loses all meaning, as every experience we've ever had is a simulation, including whatever you think isn't a simulation. A predicate needs to have two results to be meaningful. If you have a predicate that always returns the same value, it tells you nothing. So if we can define "F(x) = 'Is x a simulation' = TRUE", then F is meaningless.

This is the tragic aspect of the hypothetical hierarchy of simulations.

We are trying to create another form of intelligence. I am of the belief (for numerous reasons) that the best way to do that is to create a simulation[1]. But by its nature it would have to be so large and complex that it could be difficult for us to "look into" it and make sense of the vast seas of data it produces to actually identify intelligent behavior, and more, to cross the barrier and interface with it somehow. What if our creators simply haven't found us? What if we don't look like intelligence to them? Or maybe they identified something but don't know what to make of it?

Dolphins are actually fascinating in this regard, they diverged from us long ago and they definitely have something going on in there, but we haven't yet figured out quite what it is[2]. They are the closest thing we have to an alternate form of intelligence, with the benefit that it resembles ours enough to be recognizable.

[1] https://en.wikipedia.org/wiki/Artificial_life

[2] http://ngm.nationalgeographic.com/2015/05/dolphin-intelligen...

I can't even prove to you that I myself am sentient. And as far as I know I might be the only sentient being on earth and everyone else are just faking it. There's just no way to know right now since we don't know what sentience really is.

Therefore it's a bit of a moot point when talking about simulations, since you're asking about something we can't do in ANY universe right now.

> you'll still have no idea whether or not a modelled sentience is having a sentient experience of the universe around it, or whether it's faking its responses mechanically

Is there a difference? What is the difference? Doesn't duck typing apply here? If I act like I am sentient I am probably sentient.

Are the characters in our dreams sentient? They surely seem so and many of our dreams are elaborate and seem to obey natural laws.

Then there are some dreams where the natural laws seem different. I can fly and fly well in those and observe well as well and be observed. What do I make of those. Are beings sentient in them?

>Are the characters in our dreams sentient? They surely seem so and many of our dreams are elaborate and seem to obey natural laws.

The illusion of our own sentience and the sentience of those "characters" are generated by the same brain, so yes.

In some eastern religions

The universe is seen as a dream of 'God'. Which means start of a dream is like Big Bang and waking from dream is like extinction of Universe.

That's a bit more esoteric than what I was going for but fair enough.

Reply on news.ycombinator.com