Life comes at you fast, eh? Only a few weeks ago I was grumbling in this very slot about the way in which the big AI companies were stealing copyright material in unimaginable quantities and using it to train their models without so much as consulting the owners of the work, still less compensating them. The reaction of the tech bros to people calling them out about it has been a colossal shrug of contempt. Fussing about intellectual property is for the little people, they seemed to say. We can make a lot of money stealing your work, and our machines don’t work at all without stealing your work, so tough. And anyway, what are those little people going to do about it?
Then, along came DeepSeek. All casual-like, a Chinese hedge fund guy and his pals who had been tinkering with AI in their spare time for kicks released a free-to-use model which was not only a fraction of the price to run, using far less processing power and therefore being less environmentally ruinous, but which also worked as well if not better than the flagship products of Google and OpenAI. At a stroke, it wiped a trillion quid off the value of the western AI industry. Suspending, for a moment, our soberer consideration of the geopolitical and security implications of this development, we could all allow ourselves at least ten minutes to look at the egg-covered faces and cratered stock prices of the AI oligarchs and say something to the effect of: ‘Ha ha ha ha ha!’ Couldn’t have happened to a nicer bunch of guys.
There’s a pleasing piquancy, too, to the fact that DeepSeek (if I’m right in understanding this) kicks the backside of its competitors by being cleverer. Where Google and OpenAI proposed to improve their large language models by nothing more imaginative than brute force – essentially scaling up the computational muscle used to run them at vast expense, and outspending the competition – DeepSeek figured out a way to make the models work with more agility and efficiency. They used a scalpel rather than a hammer.
The glacé cherry surmounting this most just of desserts was the bad grace with which big tech responded to the Chinese eating their lunch. Almost immediately, OpenAI’s Sam Altman was whining that DeepSeek had cheated, and that it had trained its model on OpenAI’s work without permission. Well, fancy! This anti-copyright radical disruptor had apparently experienced an eleventh-hour conversion to the idea that there might be something to the idea of intellectual property rights after all. If it’s your work that someone is stealing for their own profit, and then using to compete with you at an advantage in the marketplace, suddenly it feels a bit less like all’s fair in love and war. They don’t, it turns out, like it up ‘em.
But when we’ve finished laughing, it bears remembering that my enemy’s enemy is not my friend. The enraging intrusion of AI into every corner of our lives is, it seems very likely, a result of tech companies having sunk too much money into it to ever admit it isn’t nearly as widely useful as they had hoped. So if they can’t sell it, they’re going to force it on us. You can see that in the way it muscles into Google search results; in its unwanted intrusion into Adobe products and nearly every customer service interaction we now have. And you can see it, outstandingly, in the way Microsoft forced its useless AI assistant Copilot into its Office 365 suite of programmes without so much as asking users if they wanted it. They made it all but impossible to turn off. And, what’s more, they ‘upgraded’ our subscriptions without telling us and charged us for the privilege. (I asked Microsoft directly how they justified this. After nearly a week, their PR company Edelman wrote ‘confirming that we do not have an official comment here’.)
The glacé cherry surmounting this most just of desserts was the bad grace with which big tech responded to the Chinese eating their lunch
So the DeepSeek moment may have put a deserved dent in the AI-industrial complex that’s pushing this rubbish on us. But it’ll probably (once they’ve all copied DeepSeek’s innovations) make them even keener to chase those sunk costs by pushing it even harder. And it does nothing about AI’s original sin – which is that, as already mentioned, its models rely for their very existence on the theft of other people’s intellectual property (technically, they could choose to train only on non-copyright material, but they would all end up sounding like 1890s flaneurs or something).
On that front, the government’s ‘consultation’ on AI and copyright continues to run; though nobody now has reason to suppose that it’s anything other than a political fig-leaf for a decision that’s already been taken. Barely even a fig leaf, such is the naked bad faith. Keir Starmer has already pre-empted the results by announcing a 50-point plan for making us a ‘world-leader’ in AI – a plan drafted, we’re told, by a tech venture capitalist – which will involve letting AI companies take anyone’s work for free unless the owner specifically asks them not to. Not an inconsequential tweak to the principles of property, but heigh ho.
And as for DeepSeek, we may enjoy seeing OpenAI getting a bloody nose; but the idea that if a Chinese company becomes the market leader we can expect it to show greater respect for privacy, data protection and intellectual property than its western competitors offers us the second big laugh.
Comments