Pacing the AI frontier

Self-regulation will be challenging

Discussion on the safety of artificial intelligence (AI) has been going on for years. But the recent breach of AI research platform Hugging Face by two AI experimental models developed by OpenAI, which escaped their digital sandboxes, has concentrated minds on the potential for greater unanticipated disasters if creators lose control of their AI models. All of a sudden, the AI billionaires have come up with virtuous pronouncements on the need for voluntary restraint and the establishment of industrywide standards. At an employee meeting last week, OpenAI Chief Executive Officer (CEO) Sam Altman said he was willing to decelerate development to assess potential risks. Soon after Anthropic CEO Dario Amodei published a 3,800-word essay titled “We must pace the frontier”, to which Elon Musk, who has mega-bets on AI to power SpaceX, signalled his agreement. This belated public discourse recognising AI’s serious civilisational risks is certainly welcome. But it is difficult to escape the notion that the call for self-restraint is little more than performative virtue signalling by the Big Tech titan.

 

For one, rather than reflecting a suo motu realisation of AI’s dangers, these responses are a reaction to a global conversation that has been circulating among researchers and developers at some of these companies. It all began when an Anthropic researcher quit his job last week, predicting that AI was “racing straight to self-improving superintelligence and gambling with our lives”. Soon enough, his predictions went viral, first among his peers in other Big Tech firms and then globally. Eventually, it spilled over into the political sphere, where contrasting responses offer examples of the limits of regulation and restraint. Former United States (US) President Barack Obama urged the Democratic Party to frame a coherent governance agenda to tackle the issue, a reflection of the liberal-left approach. But US President Donald Trump equated AI critics to “very negative forces”, which is a comment typical of the extreme right-wing. Complicating any political or corporate agenda to curtail AI development is the China factor. It is no secret that Chinese developers are not far behind their Silicon Valley counterparts in the AI race, suggesting that AI regulation must transcend borders to encompass a mutually agreed global regulatory system. How far this is possible between two fiercely competitive superpowers is an open question. As Mr Amodei acknowledged, China is the “toughest dilemma” when it comes to a proposed AI slowdown. 

 

High-minded sentiments about AI slowdowns may be tempered by the sheer scale of investment being poured into its development. According to one estimate, major AI companies this year issued about $320 billion in bonds, which equals a staggering 68 per cent of new long-term US Treasury borrowing and are already influencing interest rates. Meanwhile, Anthropic and OpenAI have blockbuster initial public offerings (IPOs) in the works. The former, slated for October/November, is eyeing a valuation of $2 trillion. Now Mr Altman has suggested that a 2026 market debut for OpenAI, targeting a mammoth valuation of $1 trillion, may be postponed over growing AI safety concerns. This scale of funding points to deep all-round systemic risks in addition to broader doomsday concerns expressed by developers. A voluntary slowdown in AI development would be a desirable development. How far Big Tech chooses to align self-restraint to the profit motive is the real question.