DeepSeek torpedoes Western AI oligopoly

Artificial intelligence is treading a familiar path of democratisation of technology. The first smartphone, IBM’s clunky Simon Personal Communicator, was demonstrated in 1992. But the technology was made widely available by Android 16 years later
DeepSeek torpedoes Western AI oligopoly
Picture credits: Wikimedia Commons
Updated on
4 min read

On Tuesday, the Bulletin of the Atomic Scientists set the Doomsday Clock forward by a second. The clock, created in 1947 by concerned scientists behind the first atomic weapons, including Albert Einstein and J Robert Oppenheimer, is a visual representation of how close the world is to anthropogenic collapse. The menace of the ticking bomb is much more compelling than the real threats it represents, like the melting of Arctic ice, new infectious diseases and the possible militarisation of Earth’s orbit.

Moved by such concerns, the clock now stands at 89 seconds to midnight. The margin of safety is so tiny that even an advance of a second represents a very high fractional increase in global risk. But if the scientists behind the clock had stayed their hand for a couple more days, they could have moved it even further forward to acknowledge new threats from artificial intelligence and disinformation. The clock was reset just the day after a cheap and shiny Chinese AI sent the industry and markets into a flat spin.

The only problem with DeepSeek, as with all things Chinese, is the shadow of the party looming over it. The AI believes that Uyghurs are treated equally in China and “Taiwan has been an integral part of China since ancient times”. It’s more diplomatic about Arunachal Pradesh: “Let’s talk about something else.”

DeepSeek is open-source. Anyone can open it up, copy it or ‘fork’ it into a different project. Of the major US AI projects, only Meta’s Llama is completely open. Projects are open-sourced to invite the world to build ecosystems using them. Open software like the Python programming language, the Apache web server and MySQL database manager, WordPress and Linux have exerted a foundational effect on multiple industries. But products built on DeepSeek could inherit Chinese censorship.

Even more interesting than the new AI is the political speculation and the fog of disinformation it has triggered. DeepSeek claims to outperform Silicon Valley AIs with much less hardware, energy and training. It’s a serious blow to an industry highlighting its scale to attract big investments. This change of perspective has profound geopolitical implications—AI, perceived to be a plaything of powerful, high-technology countries, now appears to be within the reach of much smaller entities. Unless DeepSeek’s claims are rebutted, investors and governments may not back AI unquestioningly any more.

The timing of the chatbot’s release does invite suspicion: just 10 days before Donald Trump took office in the US, following a campaign threatening a trade war against China. Two days after assuming office, he announced the Stargate Initiative, a venture fund with SoftBank, Oracle and OpenAI on board, which would invest at least $500 billion in AI infrastructure, clearly to push China onto the shoulders.

All that now looks like hubris, though the desperate hope remains that DeepSeek is a psy-op, a geopolitical gambit based on tall claims. Though its code is open, its makers have not revealed what data it was trained on. American researchers suspect it was their own. No conflict, really, because the Chinese have only claimed they broke up the training process to gain efficiency. They also claim to have used fewer and cheaper chips, because the US has restricted the export of cutting-edge Nvidia chips for security reasons.

The AI race is being seen as a national interest contest, the successor of the nuclear arms race and the space race, which were driven by dual-use technologies. Framed like that, it seems to call for export controls, the same as its predecessors. But AI is actually a different sort of animal. It is dual-use like nuclear technology, but it will be ubiquitous. You mustn’t go about with a nuclear reactor in your pocket, but your smartphone with 20 AI services on it belongs right there.

AIs like Gemini, Llama and ChatGPT are displacing traditional search engines as default gateways to the internet. Running Boolean searches is already less rewarding than asking plain questions. Ergo, AI should be generally accessible. The very internet was originally a military and academic network. If it were not opened up, if the US had treated the TCP/IP protocol as a strategic technology, the whole world would have been far poorer.

But with the release of DeepSeek, AI seems to be treading a familiar path: first someone innovates an expensive and exclusive technology, and then someone else comes along and democratises it. Cars were quite exclusive until Henry Ford laid down his production line. The first computers were institutional machines so expensive that access to them was metred by the minute; but mass-market Intel processors made home computers affordable in about 20 years. The first smartphone, the IBM Simon Personal Communicator, was demonstrated in 1992, but the technology was democratised by Android 16 years later.

The world would be less stressed without yet another arms race—if AI is developed by cross-border cooperation as a global resource, as the internet was. Maybe it would be enough to set the Doomsday Clock back to where it was―provided we ask no uncomfortable questions about Uyghurs, Taiwan or Arunachal Pradesh. But we know the answers already, and the wise value silence.

(Views are personal)

(Tweets @pratik_k)

Pratik Kanjilal | Senior Fellow, Henry J Leir Institute of Migration and Human Security, Fletcher School, Tufts University

Related Stories

No stories found.

X
The New Indian Express
www.newindianexpress.com