Turing test

So that's kind of spiralism.

I thought it's probably worth mentioning at the end that this really does sort of bump up against the good old-fashioned Turing test.

I had never actually gone and looked at what the actual Turing test was, although I had heard that it's not...

The Turing test isn't, can a computer convince a person that they're a real person? The original test proposed by Alan Turing was, can a computer do well in the imitation game? And he initially, in this, we just brought this up, he rejected the initial question, can a computer think or is a computer conscious? Because we don't actually have a good handle.

I'm asking you as a philosopher, do we have a decent handle on what it means to think and to be conscious? at this

time no we have what's called the hard problem of consciousness and we can't really we don't really have an account as to why we're conscious in the first place let alone what actually constitutes consciousness or indeed where consciousness is actually located we just know that we appear to have it there are some people who deny that we are conscious we appear to have consciousness, and it's really quite mysterious.

And the thing is, it may be true, a computer can do well at the imitation game, but what does that tell us about the ontological status of a chatbot? It doesn't tell us that chatbots are conscious because it's not a question that was designed to tell us whether or not chatbots are conscious.

It just sort of tells us where computers can sit in this sort of thing.

I mean, you could make, you know, that's when I don't buy it, but like, that is one, that's one argument.

yeah, well, you know, what's the definition of consciousness or what's the definition of, you know, intelligence? These are all, you know, I think we're way past Turing test kind of thing, almost like I don't know what it is now, right? Yeah, we blew

past it.

Well, I think that's what's so crazy, Zohair.

Like we blew past the Turing test, you know, initially a few years ago, then I think at least last year, like it beat it 80% of the time, right? When you had like multiple parties.

So it's clearly these things are able to pass off as humans, right? But yet people...

it hasn't changed the world.

Can AIs Think? 'Cause I, Like, Really Wanted to Know.

Faking the Turing Test with Tom Merritt, and more on this episode of The Morning Stream.

When I was a kid, the closest you could get me to a fish was the aquarium.

Quality you can taste.

... that's what I'd like to know.

Sure.

Uh, AGI stands for artificial generalized intelligence, and it's kind of where the industry landed after realizing that the Turing test wasn't useful anymore.

Uh, the Turing test f- was posited by Alan Turing, uh, one of the pioneers in, in computer science, you know, who, who built the Enigma machine and all of that.

He called it the imitation game.

But today, it's usually referred to eponymously as the Turing test.

In Turing's 1949 paper, Computing Machinery and Intelligence, he proposed that if a person were to hold blind conversations with two separate parties, one of them another person, the other a computer program like ELIZA, and the person couldn't tell which was which, that would be proof of computer intelligence.

Or, I should say, it would be as good a proof of computer intelligence as there is of human intelligence.

But that's the point.

When a group of psychiatrists were handed transcripts of Perry's conversations mixed in with transcripts from real human schizophrenic patients, the doctors were unable to tell them apart.

In that sense, Perry became the first chatbot to pass the Turing test.

Now, wait a minute, you might say.

Last week, we talked about the Institute's mission, the machine consciousness hypothesis, why we actually are in a particular kind of simulation, dreaming, the role of perceptions and predictions, and a fascinating take on Searle's Chinese room.

This week, we are going to talk about how to know if something is conscious, the role of memory, what Yosha thinks of integrated information theory, applying the Turing test to LLMs, and whether they're entitled to rights, substrate independence, or the portability of consciousness from one platform to another, the possibility of making a conscious AI, and the geopolitics of doing so.

So let's get right back into the interview with Yosha Bach.

I want to change direction for a moment.

I'm going to put in a couple of quotes from papers of yours.

One is, a test for consciousness cannot simply rely on performance.

It has to take the internal structure of the system into account or Similarly, there can be no Turing test for consciousness.

If I haven't quoted that out of context, how do you take the internal structure into account?

He's deeply technical.

He starts with the Turing test, which is this famous idea that if a machine can fool you into thinking it's human, then it's thinking.

Penrose is like, no way.

Right, because passing a test isn't the same as actually understanding.

So what is the official PG definition of AGI?

Oh, well, honestly, I mean, could you do better than the Turing test? I think Turing was pretty good, right? One of the reasons, one of the ways I knew we were getting close to AGI was I actually had to go and look up the definition of the Turing test.

That in itself is a piece of evidence.

Like, what was the Turing test again? Are we there already? I think what's happened now is we thought back in the day that ai would be sort of like this finish line and we agi would be this finish line we'd cross it would be very clear but when you're standing on the finish line you realize it has width from back in the 1980s it looks like it's just a line going across the horizon but here you realize okay some bits of ai are way across the finish line some bits are like maybe in the middle of it who knew the line had dimension the finish line is actually this sort of smear And I think the best answer you can give is like, we're on the smear.

It turns out to be a smear and we're on it.

And that's why you can't get your restaurant times.

Okay.

So for people coming from more of the LLM space, or especially folks that are using coding agents, clearly those have in certain ways passed the Turing test now.

But voice AI has not passed the Turing test in my view.

When I'm interacting with anything that's clearly a voice model or voice AI, generally it's pretty frustrating pretty quickly and I can tell it's not a human.

Why have we not passed the Turing test yet with voice?

Yeah, absolutely.

The key problem is we passed the Turing test kind of with scaling loss, right? If you look at LLMs, they've scaled massively.

And if you look at the parameter count at which we kind of passed the Turing test, I think you might say, okay, maybe GPT-3.5 for most people.