The hardest part of creating conscious AI may be convincing yourself it’s real

As far back as 1980, American philosopher John Searle distinguished between strong and weak AI. Weak AI is just a useful machine or program that helps you solve problems, whereas strong AI has real intelligence. A strong AI will be conscious.

Although Searle was skeptical about the very possibility of powerful AI, not everyone shares his pessimism. The most optimistic are those who support functionalism, a general theory of mind that holds that conscious mental states are determined solely by functions. For functionalists, the task of creating powerful AI is simply a technical challenge. If we can create a system that functions like us, we can be sure that it is conscious just like us.

Recently, we have reached a tipping point. Generative AI such as Chat-GPT is now so sophisticated that its responses are often indistinguishable from those of a real human. For example, see this exchange between Chat-GPT and Richard Dawkins.

This question of whether machines can trick us into thinking they are humans is the subject of a famous test devised by British computer scientist Alan Turing in 1950. Turing argued that if a machine can pass a test, we should conclude that it is truly intelligent.

Back in 1950, this was just speculation, but according to a pre-print study earlier this year, the Turing Test is now passed, although the study has not yet been peer-reviewed. Chat-GPT convinced 73% of participants that it was a human.

What’s interesting is that no one is buying it. Not only do experts deny that Chat-GPT is even a thing, they don’t seem to take the idea seriously. I have to admit, I’m with them. That doesn’t seem plausible.

The key question is: What does a machine actually have to do to persuade us?

Experts tend to focus on the technical aspects of this problem. That is, identifying what technical features a machine or program needs to satisfy our best theory of consciousness. For example, a 2023 article reported here on The Conversation compiles a list of 14 technical criteria or “awareness indicators,” such as learning from feedback (Chat-GPT was not evaluated).

But creating powerful AI is as much a psychological challenge as it is a technical one. It’s one thing to build a machine that meets the various technical criteria we set out in our theories, but it’s quite another to believe that it’s conscious when you’re finally faced with it.

The success of Chat-GPT has already demonstrated this problem. For many, the Turing Test was the benchmark for machine intelligence. But if it passes, as pre-print research suggests, the goalposts have moved. They are likely to continue to change as technology advances.

Mynah difficulties

Here we step into the murky territory of an age-old philosophical conundrum: the question of other minds. After all, we cannot know for sure whether anything other than ourselves is conscious. For humans, the problem is just idle skepticism. None of us can seriously consider the possibility that another human being is a mindless automaton, but the opposite seems to be the case with machines. It’s hard to accept that it could be otherwise.

A particular problem with AIs like Chat-GPT is that they can seem like mere copycat machines. They are like mynahs who learn to vocalize words without having any idea what they are doing or what the words mean.

Of course, this does not mean that we will never create conscious machines, but it does suggest that even if we do, we may have difficulty accepting them. And that may be the ultimate irony. Although we succeeded in our quest to create conscious machines, we refused to believe that we had achieved it. It may already be happening.

So what would machines need to do to convince us? One tentative suggestion is that they might need to exhibit the kind of autonomy observed in many living things.

Current AI like Chat-GPT is purely responsive. If you keep your fingers off the keyboard, it will be as silent as a tomb. Animals are not like this. At least not the ones we commonly think of, such as chimpanzees, dolphins, cats, and dogs. They have their own impulses and inclinations, and a desire to pursue them (or at least appear to do so). They initiate their own actions for their own reasons and on their own terms.

Perhaps if we could create a machine that exhibited this kind of autonomy, a kind of autonomy that goes beyond just a copycat machine, would we really accept that it is conscious?

It’s hard to know for sure. You may want to contact Chat-GPT.

This article is republished from The Conversation under a Creative Commons license. Read the original article.