Hacker News Reader

Comment by frankzander | Hacker News Reader

This is not yet reachable (I doubt it would ever be). So why talk about such a topic?

Certainly there are arguments against the possibility of Artificial Intelligence, but they tend to boil down to one of three assertions: one, that there is some vital field or other presently intangible influence exclusive to biological life - perhaps even carbon-based biological life - which may eventually fall within the remit of scientific understanding but which cannot be emulated in any other form (all of which is neither impossible nor likely); two, that self-awareness resides in a supernatural soul - presumably linked to a broad-based occult system involving gods or a god, reincarnation or whatever - and which one assumes can never be understood scientifically (equally improbable, though I do write as an atheist); and, three, that matter cannot become self-aware (or more precisely that it cannot support any informational formulation which might be said to be self-aware or taken together with its material substrate exhibit the signs of self-awareness). ...I leave all the more than nominally self-aware readers to spot the logical problem with that argument.

Iain M Banks - A Few Notes on the Culture

rec.arts.sf.written 1994

The most standard objection of all is that we're trying to emulate an extremely complex long running chemical process in 1s and 0s on very different hardware, and whilst it may be logically possible to do this, the human brain isn't nearly smart enough to understand its own architecture well enough to design an even more complex emulation architecture (either a brain emulation program or an accurate evolution-of-consciousness simulator).

I don't think that requires a "soul": I think it works quite well with the assumption that human self awareness is linked to complex organism-survival-biased chemical processes we call emotions which are pretty orthogonal to the sort of probability-estimation optimizations we select for with ML models. And also pretty orthogonal to those models yielding useful results...

So the first assertion basically

It's not an assertion that it cannot be emulated in any other form even if understood, it's an assertion that it's highly improbable that humans are capable of understanding in sufficient detail to design an emulator

Wouldn't that also imply that we don't understand it well enough to say that we can't?

All these complicated biological processes could be extraneous or very round-about ways of accomplishing with carbon what could be accomplished much more simply in silicon.

Well futurists love to talk about stuff like this all the time. Besides, it seems progress is an exponential process.

So why not ask the questions now? Hell, why not answer them now. The questions and answers are theoretical, not practical after all. Optimizers, mesa-optimizers, alignment, uncertainty, morality, ethics, all of it. Why not ask the questions when we have time to ask them instead of when we're short on time?

Because artificial consciousness isn't reachable. Neurologists haven't discovert the secret of consciousness and how it works. And thinking that machines could ever be such universal computers like our brains are more or less human hybris (IMHO).

But that's exactly the deal with it isn't it? Neurologists have not discovered how consciousness works. However, we do know that our brains are probably where consciousness forms, that is, consciousness is a material phenomenon.

We can already cram far more transistors onto a chip than our brains can put neurons in a similar space, however, because neurons are represented in the architecture on a higher level, that does not automatically mean we're past transistors. We have far lower neuron density in our current generic computing designs than our brains have.

However, note that our current computing is done mostly on generic CPUs and PPUs (most often GPUs), which means that we aren't done yet. Recently, many companies have started pushing massively parallel architectures, more like our brains. If we figure out a way of getting to that brain neuron density, training aside, we'll be close to the material conditions which give rise to consciousness.

I see this sentiment a lot, that attempts at general AI are improving at an exponential pace, but it’s just not true. Hell, we can’t even agree what consciousness is, let alone how to measure it. The truth is we don’t have a clue how far away we are from true AI, we have literally zero reliable indicators to gauge our progress. It’s not like we have a Moore’s law for consciousness or something like that.

I agree that we should think about these questions ahead of time and not wait around until general AI is upon us though.

It's not a "sentiment", really, it's a provable fact from history. Human technology improves exponentially, that much is true. AI has continued to improve at a similar pace, if only because of transistor counts. We have no way of knowing however, where we are on the curve right now. So it's better to just play it safe and start thinking about it now.

You're allowed to not join in on discussions you deem unworthy of your time

Reply on news.ycombinator.com