When some experts warn about AI, it is certainly worth taking notice. But if the warning comes from the head of Microsoft, whose company has poured billions of dollars into artificial intelligence, we should be on our guard. And that is exactly what happened on Sunday, July 12, when Microsoft CEO Satya Nadella published an article on his blog in which he argues that companies do not pay for AI with money alone. Without realizing it, they also pay with something far more valuable: their own know-how. He called the entire issue the “reverse information paradox.”
The reverse information paradox
Nadella draws on a theory by economist Kenneth Arrow. Arrow once explained that sellers of information risk losing it simply by showing it to a prospective buyer. Artificial intelligence, however, takes this paradox even further. In this case, it is the buyer—the company—that is at risk. In the article, Nadella emphasizes that you pay for intelligence twice: “…once with money and once with the knowledge you must reveal for AI to be useful to you.” And that is absolutely true. The better the results you want from a model, the more information you have to give it.
And this is where the problem arises. While companies feel that paid tools make their work easier, the reality is more complicated. AI learns from exactly what you enter into it—from prompts, tools, and corrections through which the user guides the model toward a better and more accurate result. According to Nadella, it is precisely these corrections that contain corporate know-how that would otherwise remain private and inaccessible to competitors. Yet companies voluntarily hand it over through AI, because the more AI knows about them, the better the results it provides.
A double standard
Nadella also directs his argument at the creators of AI models. He considers it ironic that AI companies train their models on public data from across the internet while simultaneously contractually prohibiting users from engaging in so-called distillation—that is, training their own models on the outputs of commercial ones.
This is exactly what OpenAI accused the Chinese company DeepSeek of doing. DeepSeek allegedly distilled OpenAI’s models to improve its own AI models. The paradox, however, is that large AI companies engage in this kind of “distillation” on their users every day. According to Nadella, customers should therefore have the same opportunity.
What can be done? Nadella has some advice
Nadella advises companies to retain control over their data, prompts, and feedback, build their own “learning environment”, and, above all, avoid relying on a single AI provider. The ideal environment is therefore one in which companies can switch between models from different providers.
However, Nadella’s recommendations have not escaped his critics, who see a degree of self-interest in them. After all, Nadella is the CEO of Microsoft, which operates one of the world’s largest cloud platforms, and companies can create their own “learning environment” on its cloud platform, Microsoft Azure.
Other experts are also sounding the alarm
But Nadella is not the only one drawing attention to this issue. Similar warnings have previously come from investor Jason Calacanis and Palantir CEO Alex Karp, among others. And according to TechCrunch, companies are already responding, as interest grows in open-source models running on their own hardware. What is more, these models can handle roughly 90% of the work performed by large models, but at a fraction of the cost.
How companies will choose to work with AI in the future remains an open question. Nevertheless, the question of who owns the knowledge created through the use of AI is becoming one of the most important issues in the entire industry.
Source: SN Scratchpad, TechCrunch, Business Standard
