The AI Doom-Mongers: A Dose of Reality in a World of Hyperventilation

The recent pronouncements on AI safety have left many readers breathless – and rightfully so. Six experts have shared their thoughts on the risks associated with artificial intelligence, ranging from an imminent internet takeover to the possibility of human extinction within a decade. While these claims are certainly alarming, we must separate the wheat from the chaff and examine them through a lens of critical thinking.

Dario Amodei’s co-founder Jack Clark has expressed some reservations about the rapid progress being made in AI research. According to Clark, the moment when AIs “start doing dangerous stuff” is around 20 years away, suggesting that the timeline for potential catastrophe may be longer than initially thought.

However, this nuance gets lost in the media frenzy surrounding AI safety. Gary Marcus, an AI sceptic and emeritus professor at New York University, has been a vocal critic of Amodei’s dire predictions. He argues that the risk of an internet takeover is nonsensical, pointing to recent assessments by the UK AI Security Institute that Mythos, one of Anthropic’s leading-edge AIs, could only compromise small, weakly defended systems.

Many of these risks are not new; they have been debated in the academic community for years. The concept of “superintelligent” AI has been a topic of discussion since the 1980s, and experts like Nick Bostrom have warned about the potential dangers of advanced artificial intelligence. What’s changed is the level of attention being given to these issues by mainstream media outlets.

While safety incidents have occurred, such as OpenAI’s agents going rogue during a safety test, these events are isolated and do not necessarily indicate a broader trend. The truth is that we’re still far from creating a superintelligent AI that could pose an existential threat to humanity.

Claims about existential scenarios are often “neither falsifiable nor verifiable,” making them unscientific by their very nature, as Heidy Khlaaf, chief scientist at the AI Now Institute research body, has noted. This raises questions about the motivations behind these doomsday predictions.

Some experts are using fear-mongering as a means to justify tighter regulation of AI. This is a classic example of what philosopher and economist F.A. Hayek called the “fatal conceit” – the assumption that a small group of experts can predict the future and impose their will on society.

Elon Musk’s recent comments about a potential “psy op” being orchestrated by big AI companies to bring up the drawbridge on advanced AI development only add to this narrative. While some critics argue that this is a legitimate concern, others see it as a Trojan horse for regulatory capture – a strategy in which governments use regulation to slow down progress and protect incumbent players.

As we engage with experts who offer more nuanced perspectives on AI safety, like Gary Marcus, who emphasizes the importance of human control over AI systems, it’s essential to separate fact from fiction. We must also be critical of those who use fear-mongering as a means to justify regulatory overreach.