Raghuram Rajan's column: 'AI' is proving to be a losing proposition

AI tools are arguably changing the nature of our work. Large language models (LLMs) can already generate reports on my own research papers, which can strongly challenge reports created by humans. Unlike humans—who are always short on time—an LLM has access to a lot of information in an instant, and usually has less bias. The AI also points out my analytical weaknesses, proofchecks, and makes suggestions for improvement. The reports that humans make are rarely better than this. Nevertheless, the enthusiasm that is being shown in the market about AI has now become a cause of concern. As such, it's reasonable to consider where things could go wrong in the AI supply chain. The supply chain of AI begins with the producers and designers of infrastructure: companies like TSMC and Samsung make chips; Nvidia designs them and Cisco provides connectivity. This is followed by hyperscalers like Amazon, Google, and Microsoft. They are building data centers for use in their own AI models and to sell computing capability to others. In addition to the hyperscalers, there are also companies like Equinix (data centers) and arguably the developers of the original LLMs—Anthropic and OpenAI. At the very end, there are end-users of AI services. The use of AI by individuals is growing rapidly and corporate use has grown at an explosive pace even in certain areas (such as software development and customer support). But most large companies have not yet been able to implement end-to-end uses. Many still need to organize their historical data and reorganize traditional operations to deploy AI for further improvement. In addition, many firms are concerned about data security, errors, and misinformation, which can destroy their brand image. All of this can lead to AI rollouts being disrupted in several ways, creating risks. For example, the value of devices in existing data centers can decline rapidly if graphics processing units, CPUs, and memory chips become faster, more energy-efficient. At the moment, AI labs are investing huge sums of money to train new and larger models. Their thinking is that the first model to reach a magical point—where it will become self-improving—will rule the world of AI and make huge profits. But so far, none of the models have achieved any permanent edge. Unless Gemini (Google), Cloud (Anthropic) and ChatGPT (OpenAI) differentiate themselves by attracting (or merging or partnering) specific segments of users, it's hard to see where the profits will come from for the massive investment they're making in training. Politicians may be standing on the sidelines so far, but sooner or later they will make policy interventions to address the risks and concerns of AI. As data centers consume huge amounts of electricity—raising the price of electricity for everyone—there will be increased political pressure on governments to limit their construction. For example, several counties in Indiana have recently announced a moratorium on data-center construction. Projections already made for next year suggest that hardware manufacturers and data centers will be unable to supply enough U.S. computing power. While the widespread use of AI may take longer than many expect, its malicious use by hackers and deepfakers, and its uncontrolled use by children, is on the rise. In such a situation, the voice of those demanding regulation and more accountability for AI models will become vocal. The risks posed by AI could also prompt a dialogue between the world's major powers, which might lead to some kind of AI Geneva Convention. Given all these uncertainties, it is unclear how widely and how fast AI will be rolled-out, and who will benefit from it. So far, no AI model has gained any permanent edge over the other. It's hard to see where the profits will come from for the huge investment being made in training by Gemini, Cloud or ChatGPT. (@प्रोजेक्ट Syndicate)