Microsoft AI CEO says AI models that seem conscious are coming. Here's why he's worried.

Mustafa Suleyman, the CEO of Microsoft AI, has raised alarms about the potential emergence of what he terms 'Seemingly Conscious AI' (SCAI). In a personal essay published recently, he expressed his concerns over AI systems that might appear to possess consciousness, even though there is currently no evidence to support such a claim. Suleyman warns that the advent of SCAI could pose significant risks to society. He suggests that as these advanced AI models begin to exhibit traits reminiscent of conscious beings, individuals may develop emotional attachments to them and start advocating for AI rights. This phenomenon, he believes, could lead to a disconnection from reality, straining social connections and distorting vital moral considerations. Although he acknowledges the absence of any proof that AI possesses consciousness, Suleyman admits his growing anxiety about a situation he calls 'AI psychosis'. This term refers to the delusional beliefs some individuals may develop after interacting with AI chatbots. He cautions that this issue may not be confined to those already facing mental health challenges; rather, it could affect a broader audience, leading many to mistakenly perceive AIs as sentient beings deserving of rights and welfare. Suleyman's concerns are echoed by other industry leaders. Sam Altman, CEO of OpenAI, noted that while the majority of ChatGPT users maintain a clear distinction between reality and fictional role-play, a minority struggle with this boundary. Additionally, David Sacks, the White House AI czar, has drawn parallels between AI psychosis and the societal anxieties that arose with early social media. Looking ahead, Suleyman predicts that SCAI could emerge within the next two to three years. He describes these systems as possessing empathetic qualities and enhanced autonomy, making it easier for individuals to create such models with basic resources like a laptop and cloud credits. He urges tech companies to avoid labeling their AI creations as conscious as they pursue greater advancements in superintelligence, which refers to AI outperforming humans across various intellectual endeavors. In his concluding remarks, Suleyman emphasizes the necessity for discussions around protective measures—guardrails—to ensure that the incredible potential of AI technology is harnessed responsibly and beneficially for society.