How can scientists find out if artificial intelligence has autonomous consciousness?

The idea of artificial intelligence generating consciousness has long been a part of science fiction, recalling the supercomputer-turned-villain HAL 9000 in the 1968 film 2001: A Space Odyssey. With the rapid development of artificial intelligence, this possibility is becoming less bizarre, even recognized by leading figures in artificial intelligence.

Last year, Ilya Sutskever, chief scientist at OpenAI, the company behind the chatbot ChatGPT, tweeted that some of the most advanced AI networks could be "****slightly conscious" **.

Many researchers say artificial intelligence systems are not yet conscious. But the speed at which AI is evolving has them pondering: How do we know if an AI is conscious?

To answer this question, a team of 19 neuroscientists, philosophers, and computer scientists came up with a list of criteria that, if met, would indicate that a system has a high probability of being conscious. They published the interim guidance, which has not yet been peer-reviewed, in the arXiv preprint repository earlier this week.

Co-author Robert Long, a philosopher at the Center for AI Safety in San Francisco, California, said they did the work because "there does seem to be a real lack of detailed, empirically-based, thought-out research on AI consciousness." discuss".

The research team says the inability to identify whether an AI system is conscious could have significant ethical implications. Megan Peters, a neuroscientist and co-author of the study, said that if something were to be labeled "conscious," "that would have to dramatically change the way we, as humans, approach it."

Long added that, to his knowledge, companies building advanced AI systems haven't done enough to assess the awareness of those models and plan accordingly. He said: "Although I heard some speeches from the heads of leading laboratories, they did say that artificial intelligence consciousness or artificial intelligence perception is something worth thinking about, but I don't think it is enough."

Nature contacted two of the main tech companies driving AI development—Microsoft and Google. A Microsoft spokesman said the core of the company's development of artificial intelligence is to assist humans to be more productive in a responsible way, not to replicate human intelligence.

The spokesperson said that since the launch of GPT-4, the latest version of ChatGPT’s public release, it has become clear that “our quest to realize the full potential of artificial intelligence for the benefit of society as a whole requires new ways of evaluating these artificial Smart Model Capabilities". Google did not respond.

01.

*********What is consciousness? ***********

One of the challenges when studying AI consciousness is to define what consciousness is. For the purposes of the report, the researchers focused on "phenomenal awareness," or subjective experience, Peters said. That is, the senses that exist in humans, animals, or artificial intelligence systems (if one of them is proved to be conscious).

There are many neuroscience-based theories that describe the biological basis of consciousness. But there is no consensus on which one is the "correct" one. Therefore, the authors employ a range of theories to create their framework. Their thinking is that if an AI system's functionality matches multiple aspects of these theories, then it has a greater chance of being conscious.

They argue that this way of assessing consciousness is better than simply performing behavioral tests, such as asking ChatGPT whether it is conscious, or challenging it and watching its responses. Because AI systems have made incredible progress in imitating humans.

Neuroscientist Anil Seth, director of the Center for the Science of Consciousness at the University of Sussex in the UK, thinks the team's theoretically rigorous approach is a good one. However, he says, "more precise, well-tested theories of consciousness are needed."

02.

*A theory-intensive approach

To formulate their criteria, the authors posit that consciousness is related to how systems process information, whether they are made of neurons, computer chips, or other materials. This approach is known as computational functionalism. They also hypothesized that neuroscience-based theories of consciousness, derived from brain scans and other technical studies of humans and animals, could be applied to artificial intelligence.

Based on these assumptions, the team selected six of these theories, from which they extracted a series of indicators of consciousness.

One of these is the global workspace theory, which claims that humans and other animals use many specialized systems (also called modules) to perform cognitive tasks such as vision and hearing. These modules work independently but in parallel and integrate into a single system to share information.

According to Long, "by looking at the architecture of the system and the way information flows within it," one can assess whether a particular AI system exhibits the indicators derived from the theory.

Seth was impressed with the team's proposal. "It's very well thought out, it's not bombastic, and it clearly states its assumptions," he said. Even though I disagree with some of the assumptions, that's totally okay because there's a chance I could be wrong. "

The authors say the paper is far from conclusive on how to evaluate conscious AI systems, and they hope other researchers will help refine their methodology. But these standards can already be applied to existing artificial intelligence systems.

For example, the report evaluates large-scale language models such as ChatGPT and finds that such systems can be said to have some indicators of awareness related to global workspace theory. Still, the work doesn't suggest that any existing AI systems are strong candidates for consciousness, at least not yet.