A Parable of Unintended Consequences

DeepSeek’s AI model upset the artificial intelligence status quo by matching the performance of leading systems developed with far more resources. Its announcement and circumstances of development reminded me of a few plot points in Isaac Asimov’s Foundation Trilogy, a pillar of modern science fiction first published in the 1950s – so we have another instance of reality mimicking fiction. Since there’s already been a deluge of commentary about the technology and its potential impact, some of what I write will likely repeat what someone else has already written. Mea culpa.

The Foundation Presaged Global Tech Conflicts

On to Asimov. If you haven’t read the Trilogy, you should – unless, of course, you don’t find themes like power, religion, mathematics, history, mercantilism, plutocrats, and statistical mathematics interesting. Here’s a quick synopsis of the initial premise that shouldn’t be too much of a spoiler: The eponymous Foundation endeavors to avoid a predicted 30-century-long dark age following the collapse of the Galactic Empire.

Who predicted what and how? Psychohistorians using statistical mathematics foresaw the collapse of the Galactic Empire and the ensuing decline of civilization. However, their model couldn’t account for the actions of the Mule, a mutant disrupter outside the scope of its training data. Cutting to the chase, Asimov narrates two struggles: 1) the Empire, in a conflict between its declining imperial power and the Foundation’s rising technological superiority and influence; and 2) the Mule, who, let's say for brevity, colors outside the lines. By the way, these novels were published in the 1950s, well before “long ago, in a galaxy far away” epics, but, of course, there’s only so many plots available to authors and screenwriters.

As this increasingly dogeared sci-fi genre demonstrates, upstart movements always face an imperatorial threat that limits their options, so they must make do with whatever resources they can get. This reality drove the Foundation to the efficiency and innovation that led them to miniaturize powerful technologies, much to the amazement of imperial citizens. For example, its walnut-sized personal atomic-powered force-shield could protect a single individual, while the Empire built “huge, lumbering powerhouses that will protect a city, or even a ship, but not one single man.” Technological prowess underpins its economic power against in its battle against the Empire’s military might.

Seeing a Parallel in DeepSeek and the Foundation

I was traveling by air shuttle up the Atlantic coast of the US on the day the DeepSeek news dropped, so I caught the story only in dribs and drabs on sporadic wifi. As it evolved, I understood the significance of what DeepSeek accomplished in the face of the tech Cold War between the US and China that severely limited the GPU computing power available to Chinese software developers. What DeepSeek in response did was analogous to the Foundation's miniaturization of hardware and other technologies in a Great Leap Forward for Chinese – and open-source global AI in general:

• DeepSeek took on the challenge to innovate. The restrictions on the purchase of Nvidia chips were designed to suppress technological advancement in China. Given those shackles on their work, the company encouraged innovation and thinking outside the box of 10,000-GPU engines – thousands of which were not available to them for love or money.
• It miniaturized the gargantuan AI models and built a practical application. The Foundation adapted technology force-shields, jump drives, and whisper ships to be smaller and thus more practical, while the Empire, unconstrained with respect resources, carried on business as usual. Similarly, DeepSeek didn't have to start from scratch, instead standing on the shoulders of Meta’s open-sourced Llama model. Smaller meant that they didn't need thousands more GPUs, ginormous data centers, and massive amounts of energy - and the realization of such disruption put a torch to stock market valuations of Nvidia and other AI chip makers, data centers, and the energy-generating companies that power it all.
• It likely reverse-engineered the bejeezus out of everything it could test. First-mover advantage accrues to OpenAI, but the DeepSeek team could simply brainstorm around aggregators like Poe to drill down into “ChatGPT, Claude 3.5 Sonnet, FLUX, Runway, ElevenLabs, and millions of others.“ Like the Foundation, it didn’t have to invent everything – or in the words of the great Lobachevsky, they “let no else's work evade your eyes...but remember to always call it research.“ DeepSeek acknowledges its debt to open-source software, so it’s made its oeuvre open to similar scrutiny.

What the Foundation and DeepSeek Mean to AI Development

The biggest takeaway from the DeepSeek announcement is that it deep-sixed the notion that today’s AI leaders are tomorrow's winners – whether it’s chips, juice, or code. Open-sourcing à la Meta's Llama means that viable models are ready for further development and exploitation by more developers and innovators around the world – thus boosting the oft-cited-in-Linked-posts Jevon's paradox that enabling more people will result in more ambitious attempts and breakthroughs.

Finally, it has domestic and geopolitical consequence in undermining the value of government technology embargoes in fields that have long involved development, support, and supply chains that gird the planet. Pushing developers to think outside the box doesn’t seem to have been one of the consequences considered by the sanctioning bodies. In other words, to quote Jurassic Park, “life finds a way.”