It’s a good thing artificial intelligence is developing so quickly. The natural variety seems to be sputtering.
Consider, for instance, the panicked reaction last week to the news that DeepSeek, a Chinese upstart, has devised a cheap, smart AI model that appears to vastly undercut the existing players. Investors immediately wiped nearly a trillion dollars off the market value of U.S. tech stocks, dragging the Nasdaq index to a painful loss. Then they changed their minds and propelled the index to healthy gains over the next few days.
The market’s rapid shift from despair to complacency doesn’t make a lot of sense. What it does demonstrate, more than anything else, is that AI remains an investing story built on lazy assumptions and hopeful speculation.
The investing hypothesis isn’t even internally consistent, as the initial reaction to DeepSeek demonstrated.
Think about it: If you consider AI to be a good thing, and you figure more of a good thing is better than less, then news of cheap AI should have been cause for general rejoicing. Olivier Blanchard, the renowned MIT economist, pointed this out in a note on X: the announcement of DeepSeek’s achievement was probably the largest positive total factor productivity shock in the history of the world, he declared.
In other words, DeepSeek may signal a future in which cheap, abundant AI makes the economy of every nation more productive. If you truly believe in AI’s revolutionary potential, what’s not to like about that?
The only real casualty of the DeepSeek unveiling was the simplistic notion that the future of AI would necessarily be dominated by a handful of prominent U.S.-based companies such as Nvidia, OpenAI and Alphabet. But that always seemed unlikely, anyway. Anyone who knows a bit of history realizes that the classic pattern is for a smart innovator—IBM, say, or Intel or AOL—to get off to a commanding lead in the early stages of a technology, then see its paramount position eroded by younger, cheaper, more global upstarts as the years go by.
This seems to be happening already with AI, and it is not necessarily bad news. If DeepSeek’s success makes investors more realistic about how competitive the sector is likely to become, and how unpredictable its development is likely to be, it will have performed a valuable service.
Investors, though, seem determined to shrug off those implications. The rebound in tech stocks speaks to the widespread desire to believe that we can still forecast how AI will develop and who will control the sector. This seems absurdly optimistic.
For starters, nobody knows how genuinely intelligent AI models will become. Economist Dan Davies, who recently wrote a book on management cybernetics, points out that ChatGPT and other popular AI apps are really “unoriginality machines” that merely recycle existing output at high speed.
On a more hopeful note, some AI models do seem capable of genuine insight. AlphaGo, an AI model devised by Google subsidiary DeepMind to play the classic board game Go, managed to beat human masters by playing in a style that seemed “utterly alien” to them, Davies notes. By dint of playing against itself a zillion times at computer speeds, AlphaGo discovered strategies that had eluded human players operating at slower flesh-and-blood velocities.
Yet, despite such isolated successes, researchers still seem far away from developing Artificial General Intelligence (AGI)—machines that can turn their attention to different subjects and learn and think about them in the same way a human being does. “I think it’s definitely sensible to be skeptical of AGI,” Davies writes.
Even garden-variety AI may not be an immediate panacea. People should expect “that AI, like every other information technology, will end up creating complexity as well as processing it, that the robots will get in each other’s way just like we do, and that consequently we are going to systematically overestimate the benefits of the technology during the initial phase,” Davies cautions.
Maybe the best lesson to take away from the DeepSeek ruckus is the need for humility. We simply don’t know which companies or countries will dominate AI or how powerful the ultimate effects will be.
A bit of perspective might help. Timothy B. Lee, who writes the newsletter “Understanding AI,” spent two days testing DeepSeek’s model and came away impressed. The Chinese product is as good as Google’s new Gemini 2.0 Flash Thinking model, he concluded. It is also nearly as good as OpenAI’s state-of-the-art o1-pro model, but much cheaper.
All three models are much better than the models that existed just a few months ago, in that they correctly answer some problems that stumped their predecessors. They all realize that filling a car with helium will not cause it to float away. They all know that 9.9 is larger than 9.19.
However, a simple brain teaser can still be a challenge. A puzzle that involved someone duct-taping the upper surface of a ham sandwich to their walking stick and carrying it to another room stumped both the Google and DeepSeek models because they did not realize the sandwich would fall apart.
Until AI can reliably cope with a ham sandwich, investors might want to refrain from assuming they can confidently predict its real-world impact over the next few years.
Ian McGugan writes about markets and economics. His work has appeared in the Globe and Mail, the New York Times and Bloomberg/BusinessWeek. He was founding editor of MoneySense magazine.
The Canadian Family Offices newsletter comes out on Sundays and Wednesdays. If you are interested in stories about Canadian enterprising families, family offices and the professionals who work with them, but like your content aggregated, you can sign up for our free newsletter here.
Please visit here to see information about our standards of journalistic excellence.