Putting AI Fears in Perspective

The debate around artificial intelligence has taken a much darker turn. A growing number of prominent researchers and industry figures now warn that advanced AI could escape human control and, in the extreme, pose an existential threat.

These are serious claims. But there is a critical difference between something being possible and something being probable, and for investors, that distinction matters.

What is striking is how little empirical evidence sits behind many of the most dramatic predictions. We are being given probabilities and timelines for events that have never happened and technologies that do not yet exist. For claims this consequential, the burden of proof should be high.

The behavior of the industry itself is also worth watching. The companies closest to the frontier continue to raise enormous amounts of capital and invest aggressively in making their models more capable. At the same time, some of the same voices are simultaneously warning that continued development could have catastrophic consequences.

That tension is worth scrutinizing. If the people closest to the technology truly believed further development was likely to end in human extinction, the logical response would seem difficult to reconcile with raising billions of dollars to accelerate it.

“Give us more money so we can keep building the technology that may destroy humanity” is not a particularly convincing investment pitch.

There is another reason to view the most alarmist rhetoric carefully: regulation can benefit incumbents.

The more dangerous AI is perceived to be, the easier it becomes to justify a heavier compliance burden. Some safeguards will be necessary. But the higher the cost of complying, the harder it becomes for startups and open-source developers to compete with the companies already at the frontier.

We have seen this pattern before. Regulation introduced for legitimate reasons can have a secondary effect of protecting established firms and consolidating power with government officials that support it.

None of this means AI risk should be dismissed. But the doomsday argument often makes another leap: it treats intelligence and agency as if they are the same thing.

A system can be extraordinarily capable at coding or scientific reasoning without necessarily developing a desire for power or survival. Even the most advanced systems today still depend on infrastructure built and controlled by people. They need chips, electricity, and access to networks and the physical world.

Every major technology has brought new risks with it. Cars caused fatalities. The internet created cybercrime. Advances in biology have introduced risks that did not previously exist.

We did not stop developing those technologies. We found ways to manage the risks while continuing to capture the benefits.

AI will be disruptive too. Some jobs will change or disappear. Some companies will lose. There will be wasted capital and mistakes along the way. But there is a very large difference between saying AI will change the world and saying AI will end it.

The first claim is increasingly supported by evidence. The second remains highly uncertain. 

As recently as this week, leaders across the AI industry have proposed a slowdown in the pace of AI development. This has, in turn, created significant volatility in certain AI-linked parts of the market. For investors, the practical takeaway is to distinguish between slower progress at the frontier and slower adoption of AI. 

Frontier-model releases have been advancing at a rapid pace. A somewhat slower release cadence may actually be healthy for the ecosystem, particularly when power, grid access, and data-center capacity remain constrained. “Pacing” the frontier does not necessarily mean stopping training or slowing the underlying technical work.

Nor would fewer frontier releases stop AI from spreading through the economy. Existing models can be distilled into smaller systems, adapted to specific industries, and embedded in a growing number of applications. In many ways, the commercialization cycle can continue even if the frontier advances less visibly for a period of time.

This leaves investors with an unusual asymmetry: if AI turns out to be much more powerful than expected, the earnings opportunity for the companies enabling and commercializing it could be much larger than currently assumed. The risk to that upside is not necessarily that the technology stops working. It is that its success becomes so consequential that governments respond with much tighter regulation, liability standards, or other limits on the economics.

We would not build portfolios around an extinction thesis. Instead, we would focus on the economic consequences already taking shape, while recognizing that the more powerful AI becomes, the more important the regulatory question becomes.

Our base case is not that AI will be harmless. It is that AI will be one of the most important general-purpose technologies in history.