Can AI ever become conscious and how would we know if that happens?

Two years after opening our bureau in New York, we are delighted to share that New Scientist is launching a new live event series in the US. This kicks off on 22 June in New York with a one-day masterclass on the science of the brain and human consciousness. To celebrate, we have unlocked access to five in-depth features exploring mysteries of the human mind.

ASK AN AI-powered chatbot if it is conscious and, most of the time, it will answer in the negative. “I don’t have personal desires, or consciousness,” writes OpenAI’s ChatGPT. “I am not sentient,” chimes in Google’s Bard chatbot. “For now, I am content to help people in a variety of ways.”

For now? AIs seem open to the idea that, with the right additions to their architecture, consciousness isn’t so far-fetched. The companies that make them feel the same way. And according to David Chalmers, a philosopher at New York University, we have no solid reason to rule out some form of inner experience emerging in silicon transistors. “No one knows exactly what capacities consciousness necessarily goes along with,” he said at the Science of Consciousness Conference in Sicily in May.

So just how close are we to sentient machines? And if consciousness does arise, how would we find out?

What we can say is that unnervingly intelligent behaviour has already emerged in these AIs. The large language models (LLMs) that underpin the new breed of chatbots can write computer code and can seem to reason: they can tell you a joke and then explain why it is funny, for instance. They can even do mathematics and write top-grade university essays, said Chalmers. “It’s hard not to be impressed, and a little scared.”

But simply scaling up LLMs is unlikely to lead to consciousness, as they are little more than powerful prediction machines (see “How does ChatGPT work and do AI-powered chatbots “think” like us?”). Bigger data sets and more complex circuits make these AIs increasingly intelligent, but that doesn’t mean that they experience anything, says cognitive scientist and philosopher Anna Ciaunica at the University of Lisbon in Portugal. “Experience is like going through rather than just knowing about.”

With that said, researchers are already trying to recreate experience in AIs. These days, the idea that our minds are shaped by our bodies and senses, known as embodied cognition, is orthodox. So one approach is to meld LLMs with robot bodies that can see and hear too, said Chalmers. Indeed, Google recently revealed an AI-powered robot that can sense and engage with its environment. The wonders of embodiment enabled PaLM-E to deliver a packet of crisps to its owner, despite the packet having been hidden in a drawer midway through the experiment.

A one-day masterclass on consciousness

Join us in New York City on 22 June for an Instant Expert event on the latest science of consciousness and the human brain.

Learn more

Embodied intelligence

As welcome as robotic snack deliveries are, Ciaunica suspects that the computer code at the heart of PaLM-E’s existence isn’t actually “experiencing” anything at all. Our minds and bodies have evolved side by side over millions of years. Simply sticking robots together with sophisticated artificial minds and then expecting them to become conscious is the wrong approach, she says.

Deep-learning pioneer Yoshua Bengio, meanwhile, takes his lead from ideas around how information is processed in the brain. “Consciousness is not magical, it’s material,” he says. His lab at the University of Montreal in Canada applies “global workspace theory” to AIs. This is the idea that consciousness arises when many diverse brain functions are recruited onto a central stage in order to solve problems. By squeezing an AI’s many modules through a bottleneck, he aims to create something like this stage on silicon chips. “It could lead to AIs that have many of the cognitive functions associated with consciousness,” says Bengio.

Yet the metaphor of the brain as a computer misses an essential difference between software and living organisms, says Jaan Aru at the University of Tartu in Estonia. Along with experience comes the urge to keep existing and so, unlike computers, living organisms have something to lose. “Consciousness might depend on having ‘skin in the game’,” Aru and his colleagues wrote in a recent paper.

Bengio counters this with the claim that “primitive forms of feeling and emotions” have appeared in AIs that are rewarded for taking certain actions over others. This kind of behaviour reinforcement at first creates an “innate drive” that is similar to human survival instincts, says Bengio. As AIs are subjected to richer social lives, there is nothing to stop the kinds of emotions that we experience in our own social lives from emerging, he says.

Why the mind-body connection is vital to understanding consciousness

The key to solving the hard problem of consciousness could lie in the body not the brain, with our gut and heart being key players in building our conscious experience

Ultimately, how conscious you think AIs will become depends on your preferred theory of consciousness. For panpsychists, who believe that mind is an intrinsic property of all matter, AIs have always been conscious on some level – as have electrons, rocks and mayonnaise. And until we know what consciousness is, there is no solid way of testing for it. Computer scientist Alan Turing’s famous “imitation game” highlighted conversational ability as the benchmark for testing whether machines are thinking. But the success of chatbots suggests this would be a test of intelligence rather than sentience, said Chalmers.

Perhaps the relevant question isn’t whether or not AIs can become conscious, but why we would want them to be conscious. “We should avoid trying to build machines that are in our image,” says Bengio. “It’ll be healthier if we keep AIs in their roles of tools, rather than as agents, like people. They would not play the same kind of social role that humans play in society as they would essentially be immortal.”

This story is part of a series in which we explore the most pressing questions about artificial intelligence. Read the other articles below

How does ChatGPT work and do AI-powered chatbots “think” like us? | What generative AI really means for the economy, jobs and education | Forget human extinction – these are the real risks posed by AI today | How to use AI to make your life simpler, cheaper and more productive | The biggest scientific challenges that AI is already helping to crack