Google DeepMind’s in-house philosopher Henry Shevlin says deliberately creating conscious AI now would be “pretty reckless”. Speaking to the Financial Times, he argues for a better understanding of consciousness and wider public discussion first. Source
Shevlin compares it to having a child or adopting a pet before knowing how to care for it. The care arrangements aren’t exactly a detail to sort out after the adoption. He wants researchers and the public to consider how a conscious system would be treated. Source
Shevlin’s work at DeepMind includes machine consciousness, human-AI relationships and readiness for artificial general intelligence. The FT reports that he also retains a post at Cambridge. Source
An impressive answer isn’t the same as an inner life
The FT explains consciousness through philosopher Thomas Nagel’s account of subjective, first-person experience: whether there is something it feels like to be a system. Source
One approach at DeepMind, the FT reports, is to examine whether existing theories of consciousness apply to AI. “Consciousness itself is something we just don’t understand,” Shevlin says. He doesn’t rule out a new model exhibiting something like it. The latest chatbot doesn’t get a consciousness certificate on that basis. Source
Non-conscious doesn’t mean safe
Shevlin also warns against treating greater capability as automatic evidence of consciousness, while acknowledging that greater intelligence could provide stronger links to it. Source
His caution about creating conscious AI isn’t a claim that consciousness itself makes a system dangerous. Nor does he think avoiding it guarantees safety: “It’s equally not the case that as long as we don’t make conscious AI it’ll be safe.” Source
Shevlin told the FT: “If we do, at one point, decide that we’re going to build conscious AI systems, I think we should be ready for it.” Source
Comments