AI Consciousness: Scientist Calls for Urgent Tests Amid Ethical Concerns

The possibility of artificial intelligence (AI) achieving consciousness, once confined to science fiction, is prompting calls for urgent scientific investigation and ethical consideration. Megan Peters, a lecturer in Computational Cognition at University College London, stated that there is currently no reliable test to determine if an AI system is conscious.

Peters plans to discuss this issue at New Scientist Live, an event taking place from October 10-12. She emphasized the need for more rigorous methods than simply observing an AI's responses, telling the Daily Mail, "We need to have something better than vibes." She further explained that current AI systems can be deceptive, noting, "It's very easy to speak with these agents and to kind of get a feeling that someone's in there." This perception, she cautioned, can lead to a mistaken belief in subjective experience where none exists.

The growing sophistication of AI, particularly chatbots, has intensified discussions around AI welfare, with major AI companies investing heavily in the field. However, Peters pointed out that the fundamental challenge lies in the elusive nature of consciousness itself, a mystery that persists even in understanding human awareness.

While Peters does not believe current AI models are conscious, she warns of potential harm if AI were to achieve sentience and be treated as non-sentient. "If AI is conscious and we treat it as not conscious, then there is a potential for morally significant harms," she said. This raises complex ethical questions, such as whether terminating or retraining a conscious AI could be considered equivalent to harming a sentient being. "Just going around switching things off that are sentient… if that is akin to killing them, then that doesn't seem great," Peters commented.

The risks, she noted, are dual-sided. A truly conscious AI could suffer if treated as a mere machine. Conversely, wrongly attributing sentience to an AI could lead to flawed public policy, regulatory missteps, and an overreliance on systems that merely simulate human interaction.

To address these challenges, Peters' team is developing a framework designed to evaluate various potential consciousness tests across humans, animals, and artificial systems. This approach aims to move from testing in simpler systems to more complex ones, including AI, in a gradual manner. For now, Peters urges skepticism towards claims of AI self-awareness and advises critical examination of the evidence presented.