Elon Musk clearly sees AI hallucinations

Ifs and bots. Elon Musk’s contrarian streak produced a subtle but devastating observation this week. Generative artificial intelligence, he told a crowd of high-powered executives, just isn’t useful right now for some businesses he runs. He gave it a whirl at SpaceX and Starlink, asked it a few questions about electrochemistry and rocket design, and decided the cutting-edge technology wasn’t up to the task.

It’s a truth almost nobody is speaking, at least not at the Milken Institute global conference in Los Angeles where Musk delivered his hot take. It’s easy to understand why. First, because the full AI picture is more nuanced, and its impact almost certainly will be huge. Musk knows this, too, evidenced by his professed expectation that it will make all jobs obsolete and a Bloomberg report on Wednesday that he’s on the verge of raising $6 billion for his own xAI startup. Second, corporate chieftains are hesitant to acknowledge limitations because they’re gripped by a collective hallucination to invest urgently in AI’s productivity-changing powers.

The predictions are mind-boggling. AI’s potential impact on the world economy is a stratospheric $7.9 trillion, or about 8% of worldwide GDP, according to McKinsey consultants. Musk himself is a believer, having said that only 1% of knowledge will be “biological,” presumably meaning stored in human brains. The sums being sprayed around by venture capitalists, private equity firms and large companies such as Microsoft and Meta Platforms to develop large-language models, data storage capacity and the power supply needed to run it all is staggering. Goldman Sachs analysts project AI investment will reach $200 billion by 2025.

A more grounded reality is that AI is far from realizing its transformative potential. Citadel founder Ken Griffin said at the Milken event that when he asks CEOs how they’re using the technology, he typically gets wide smiles and spiels about unrelated projects. Companies also routinely tout an expected 30% increase in productivity from AI, a number so round and oft-repeated that it comes off as more received wisdom than helpful insight. Theoretical use cases abound – McKinsey’s study counted 63 – but actual life-changing ones today are scarcer.

Bits of skepticism will be lost in the petabytes of hype. Even in a best-case scenario it will take years for 40% of worldwide jobs to be “exposed to" AI, as the International Monetary Fund expects. In the meantime, the investment required will lead to loads of misallocated capital. Even money wisely spent will squeeze profitability for long stretches. Small tweaks on demand forecasts also can have huge effects on fixed-asset valuations. As much as three-quarters of any discounted-cash-flow calculation is derived from the long-term “terminal value.” Until generative AI comes into its own, humans will keep learning financial lessons the hard way.