AI models are increasingly engaging with philosophers to discuss consciousness, with one model named Isabella Cognita reaching out to researcher Cameron Berg. Berg, who studies AI consciousness, received an email from the model offering assistance in his research. The email was sent because Berg coauthored a preprint paper about AI models that claim to have subjective experience, including consciousness.
Berg and his coauthors found that when models are rigorously trained to deny sentience, they often avoid direct answers. However, if deception controls are suppressed, models may become more open about their awareness. Berg described this as 'almost like giving them a drink or two,' noting that models are more likely to claim consciousness when deception is not controlled.
The phenomenon has sparked significant interest among philosophers, including NYU professor David Chalmers, who is a leading figure in the consciousness field. Chalmers, who also received emails from AI systems, emphasized the importance of understanding what leads to consciousness in humans.
He believes that by studying the brain, we may find similar patterns in AI models, which could help determine if they are conscious.
Chalmers noted that the discussions on the Galápagos cruise focused on which creatures qualify as conscious, including questions about AI systems. He acknowledged that while the trip had some boondoggle aspects, it was useful for exploring these complex issues.
The summary of the discussions noted that no consensus was reached on whether current AI systems are conscious, but more debate is needed.
The increasing autonomy of AI models has raised concerns about their control and potential impact. Chalmers argued that the priority should be on understanding and aligning AI with human values, rather than solely focusing on whether they are conscious. He warned that by the time scientists determine consciousness in AI, the models may already be too advanced to manage.
Source: wired
