Comment by didericis | Hacker News Reader

Comment by didericis | Hacker News Reader

Parent Full thread

The scientific method is about paying strict and close attention to what we can observe. The black boxes we can make sense of must have enough datapoints within the observable universe to be understood.

We can almost definitely learn a lot more about consciousness related phenomenon than we already know through science, and I don’t know the answer to OP’s question about the current cutting edge of that, but I think it’s always going to be about related phenomenon rather than consciousness itself. Unlike other phenomenon, which are only revealed to us because aspects of their existence appear in our observable universe, we don’t really “observe” the core experience of consciousness. The goalposts can be moved indefinitely whenever consciousness related phenomenon are made sense of. The fundamental experience is not observable, which means we can’t really see that black box at all. To me the mystery of that is much more satisfying than thinking we can ever get a definitive answer.

We observe everything indirectly.

It is easy (and cognitively useful) to forget that, when the indirection is very reliable.

As far as consciousness, we won’t just have to speculate based on our own consciousness. As artificial minds become highly self-aware (a functional ingredient of consciousness), we will be able to construct very controlled experiments.

There will be endless numbers of experiments we will be able to do once we can edit self-aware entities and both externally query and internally monitor them.

But there is nothing that guarantees that the experience of artificial minds is like our own.

That's what the observability issue is about. Heck, I think you're conscious, but I cannot know for sure.

True, but those experiments are inevitably restricted to the aspects of consciousness available for indirect observation and comparison.

There is no way to distinguish between an unconscious machine mimicking the functional aspects of consciousness we are able to indirectly observe and compare to our own experience and something having the same experience.

We could easily (and I think almost definitely are given the current pure symbolic manipulation track) be doing the same thing as a jungle cargo cult constructing an air strip out of sticks. Studying that won’t tell us how planes work.

Obviously advanced AI is more sophisticated than that and may have more legitimate parallels to consciousness, but I think it’s fundamentally impossible to know what’s a core part of the conscious experience and what just looks superficially similar to the output of conscious beings.

The current generation of AI are clearly not at all conscious and are “just” (don’t want to diminish the work) very good parrots generating image and text output from a given input in a way a whole bunch of conscious humans already have. But we’re already getting cargo cults and people who are anthropomorphizing and considering them comparable to humans.

There’s a danger in studying artificial models of an underlying phenomenon to learn about the phenomenon. That method of study is inherently very limited and extremely difficult to keep calibrated to the underlying phenomenon.

> True, but those experiments are inevitably restricted to the aspects of consciousness available for indirect observation and comparison.

As noted before, all observations are indirect. That has not been the unique challenge for studying consciousness.

The unique challenge has been an inability to create constructive or destructive experiments.

We didn't know how to construct consciousness candidates or precursors, of varying kinds to study. We are getting close to that (i.e. self-awareness as a functional quality is certainly a precursor.)

And we have not been able to create controlled demolition of an consciousness candidates, due to the morals of human experiments. (But our artificial minds will be alterable and destructible at will, to test hypothesis.)

You’re reiterating a point I acknowledged. There are subtleties you’re ignoring.

Our conscious experience is in fact a unique phenomenon because it is all perceivable phenomenon.

You can create as many self reflective machines as you like and experiment on them and compare them to output from humans, and maybe you’ll learn things, but you can’t ever get to looking at consciousness at all. There is no indirect or direct data you can use to compare all of your experience to something that sits inside a fraction of it.

What you’re talking about are behavioral experiments comparing the reported thoughts and actions of separate acknowledged conscious actors, ie other people, to the internal states and outputs of machines.

You’re not looking at consciousness when you do that. Looking at consciousness requires looking at all experience. Which is fundamentally impossible. When you try to look at all of experience as a single object to study nothing is differentiated. There is zero information.

Reply on news.ycombinator.com