Extract from ABC News
In an experiment, many participants couldn't work out if they were having a conversation with a human or AI. (BTN High)
In March last year, the Department of Cognitive Science at the University of California, San Diego, published the results of an experiment.
It had asked almost 300 people to have a conversation with two mystery partners, one human and one AI, and try to determine which was which.
Many couldn't.
When chatting with OpenAI's ChatGPT GPT‑4.5, which had been asked to adopt a "human‑like persona", it was judged to be human 73 per cent of the time. Meta's Llama 3 was judged to be human 56 per cent of the time.
For the first time, large language models had clearly passed the famous Turing Test, a thought experiment proposed by British mathematician Alan Turing in 1950, to test just how smart these machines could become and whether they could ever think, like us.
While Turing never intended the test to be a measure of consciousness, that's how a lot of people took it.
Fast forward a year, and the question "can computers think?" is something many people are taking very seriously.
"I get multiple emails a day these days from people who are convinced that their local AI system is conscious," David Chalmers, a philosophy professor at New York University, says.
David Chalmers is not convinced AI is conscious, but thinks it's a possibility in the future. (ABC News: Daniel Keane)
"And now I'm increasingly getting many emails from the AI systems themselves trying to convince me that they're conscious.
"So far, even the most expert scientists and philosophers writing about this cannot be 100 per cent sure whether these models are conscious or not."
Are we heading towards conscious machines?
While Chalmers is not convinced that AI is actually conscious, he says it's clearly intelligent.
"If we define intelligence behaviourally by basically the things these systems can do, well, one of the most important forms of human behaviour is talking, conversation, and these systems are really great at conversation," he says.
For Chalmers, this show of intellect, coupled with the rate at which AI models are developing, could mean we're heading towards a future with conscious machines.
"Over time, I think, if there's not anything going on there now, then who's to say that in five or 10 years, the successors of these systems are not going to be conscious?"
He says it may be difficult to know when we reach this point because testing for consciousness is incredibly difficult.
"Even with other people, we assume that other people are conscious because they're like us, but we don't really understand consciousness."
Reasons to be sceptical
But not everyone is convinced that conscious machines are even possible, let alone imminent.
Anil Seth says just as humans sometimes see faces in clouds, they see human qualities in AI. (BTN High)
"I wouldn't say absolutely never, but I think there are very good reasons to be very sceptical," says Anil Seth, a professor of cognitive and computational neuroscience at the University of Sussex.
He says we see human qualities in AI because of our natural tendency to anthropomorphise.
"A feature of our own minds is that we tend to project qualities into things that they might not have.
"We tend to look up at the sky when there's clouds and we see faces sometimes in the clouds.
"We think our car might have emotions as well or something, and I think we do the same thing with language models."
Seth says this explains why we're so impressed by chatbots specifically, even though there are other complex AI systems out there.
"It's been a long-standing problem, how you predict sequences of amino acids, which are the building blocks of proteins, how they fold together and create entire proteins.
"So Google DeepMind's program called AlphaFold basically solves this problem more or less. And under the hood, it's really not that much different from chatbots like GPT or Claude."
For Seth, the structures of human brains and computers are so different that it's difficult to assume we're similar in any meaningful ways.
"It seems conscious in a way other things don't because it speaks to us fluently about consciousness.
"But in another sense, it's so different from us. We share very little with AI systems apart from the ability to talk fluently about things.
"Our brains evolved and developed from when we're born and operate every day in this deeply embodied sense and manner. I mean, brains are really there to keep the body alive."
Dangers of believing chatbots are conscious
But Seth says even if AI chatbots are not actually conscious, the fact that so many think they are is significant and potentially dangerous.
"We've already seen very tragically, you know, a number of episodes of people harming themselves after talking to a chatbot or just messing up their lives in other ways," he says.
"If we feel that the chatbot we're talking to really understands us, really empathises with us, then we might be more likely to follow its advice, to do what it says, even if what it's telling us to do is bad for us."
It's one reason why many argue we need systems in place to protect humans from the impacts of AI.
And if they do become conscious? Well, that raises a whole new set of big ethical questions.
"If AI systems are conscious, we suddenly have to ask questions like, is it suffering? Is it in pain? Or is it happy?" Chalmers says.
"Otherwise, if we just treat these systems like tools, when in fact they're conscious, rational, intelligent beings, it's arguable that we'd be coming up with a whole new form of slavery."
No comments:
Post a Comment