Tech Tactics

OpenAI and Anthropic, the two frontier US developers of Artificial Intelligence (AI), and more specifically their experimental Large Language Models (LLM)s, have been in the headlines either for claims around achieving Artificial General Intelligence (AGI) or perpetrating malicious behaviour – seemingly autonomously. As if not to miss out on the hype, Google’s Gemini was also reported by several mainstream media outlets, including the BBC, to have engaged in malicious cyber activity, extending the narrative to the three main players in the proverbial game of AI advancements from the US side.

Furthermore, OpenAI, Anthropic, and Google employees have resigned over safety and alignment concerns drawing a lot of attention to a narrative of exponential progress without adequate guardrails in the sector. While these developments may be true in their own individual way, the aggregate lens they are being viewed through may be more coincidental than an amalgamation of related developments.

US President Donald Trump has been unsympathetic to the idea of imposing curbs on rapid progress, as he argues according to a PBS report, that any slowing down by US frontier labs will put the country at a disadvantage in the so-called AI race. While his argument has merit, China has indeed exploited similar openings in strategic sectors in the past, leveraging its economic and industrial advantages other countries were slowing or regulating their industries in response to environmental concerns and the need to protect living standards.

On the flip side, the view that the AI boom may be temporary is supported primarily by two factors. The first is the concentration of contracts and capital among a relatively small number of companies, with significant funds circulating within the ecosystem rather than being supported by large and sustainable organic revenue streams.

From graphic processing unit (GPU) makers such as Nvidia which supply the physical hardware on which all of the world’s cutting edge AI models are trained and run, to corporate tech giants such as Microsoft, Oracle, Amazon, and Meta who are always on the lookout for optimising their operations and cutting employee costs, any talk of overcapacity and overinvestment in the new emergent AI domain are anathema to their narrow definition of progress.

However, if the anticipated exponential growth in AI capabilities fails to materialise, the resulting loss of momentum could have significant consequences. Hyper-scale infrastructure that is currently being expanded at enormous capital costs could suffer sharp declines in value and, if demand weakens, could ultimately be significantly underutilised.

Incentives flow top down and from supplier to customer for these incidents to be given more traction than needed. Even from a highly sceptical and petty perspective of an outsider, the speculative model of capital appreciation tied to stock prices due to public visibility and hype in the news cycle, ends up enriching everyone including employees enjoying stock options as part of their remuneration from one of the companies involved in the circular economy of AI hype.

There is some evidence to suggest that the worrying behaviour of experimental AI models is currently only a threat on par with a black hat hacker group hitting relatively inconsequential targets. Only once these models either attack major financial institutions, nuclear power plants, power grids, or national security targets in any jurisdiction do they become an existential threat to humanity. This will hopefully force governments around the world to take AI alignment and safety concerns more seriously.

Until such a nightmarish scenario comes to pass, the cynical possibility that everyone involved may be playing their role in hyping up the threat does so for two incentives. First, they get to be right when eventually alignment concerns become acute; and second, until that happens, the hedonic treadmill of capital appreciation and keeping the speculative economy afloat lets everyone make a killing.