Something’s Not Right with AI

I have a lot happening on the personal front this week — nothing too serious, but time-consuming, nonetheless — and that means my usual posting cadence will be disrupted. I’ll do what I can when I can, but I thought it only fair to give you prior notice. Everything should be back to normal next week.

In the meantime, I want to give you some substantive food for thought. At the end of this post, I’m going to provide counsel, which you are obviously free to accept or reject. We’re all respectful of each other’s opinions here.

We’ve all read, seen, or heard posts, reports, and interviews during the last week in which AI cognoscenti, executives, and putative experts have given us urgent warnings that an ill-defined AI superintelligence not only exists but is on the verge of annihilating us all.

This is, or was, the stuff of dystopian science fiction. We are now asked to consider AI armageddon scenarios as imminent risks. The intelligent machines have become too smart, beyond human control, and, to make matters far worse, they’re malicious bastards. As Jim Morrison sang many years ago, in an admittedly different context: “The future’s uncertain, and the end is always near.”

The answer, as proposed by the three AI musketeers — Anthropic’s Dario Amodei, OpenAI’s Sam Altman, and celebrity CEO Elon Musk — is to slow the pace of AI development. They believe putting a temporary brake on development is the only way to preclude AI catastrophe.

Is the situation really so dire? The fears might be real. Some people, including those who work in the AI trenches, earnestly believe that we’re in mortal danger of being slaughtered by our AI Frankensteins. Not everybody does, though. Some equally credible and knowledgeable people disagree strongly. We’re marooned in the DMZ of competing speculation rather than the contented settlement of verified facts.

Timing Begs Questions

The question I ask is, why now? Some of you might say that it’s because AI — from generative to agentic to the highly contentious “superintelligent” — has made such lengthy, rapid strides in such a short period of time. Perhaps that’s true. I can’t dismiss that view summarily, but I don’t see definitive proof to support that assertion.

As we’ve discussed previously, no form of extant AI possesses anything close to human agency or volition. AI doesn’t want, it doesn’t need, and it isn’t motivated by fear, greed, or any other human emotion. Nobody has proven scientifically that AI has human “intelligence,” much less human emotions and motivations. AI is trained on data, programmed by algorithms, and produces probabilistic output (inference) governed by mathematics. It’s a highly tuned prediction engine, but the predictions are extrapolations of data from the past, not the future. (There is no data from the future.)

There is no evil genius behind the curtain, just an impressive technological artifact. How can this be sentient, superintelligent, or, as we’re now being anxiously informed, malevolent?

Doesn’t it seem more likely that AI is a sophisticated tool, a culmination (for now) of the evolution of information technology, the use of which is governed entirely by human intent, competence, and purpose? AI can be used for good or ill, but how it is directed, deployed, and executed is entirely up to humans.

Could it be dangerous? Sure, but not because it’s some demonic robotic monster of unsurpassed intelligence, but because people fail to manage and use it competently or responsibly; or because they decide to use it for malicious means.

Human Dereliction, Not AI Gone Rogue

Aidan Gomez, CEO of enterprise-AI company Cohere, made the following observation this week about the allegedly “rogue” exploits of OpenAI’s agents and models, and about Anthropic CEO Dario Amodei’s plan to slow AI development:

As for the OpenAI hack, Mr. Gomez wrote that none of Mr. Amodei’s proposals likely would have prevented it. “What failed was the quality of the instructions, and the strength of the walls around the test, and how long agents were allowed to continue working without observation,” he wrote.

I agree. As mentioned here before, what’s happening is that AI is being mismanaged and misdirected, not that AI is running amok of its own (non-existent) volition. The mismanagement and misdirection of AI is a real and serious problem, but it’s not the same problem as a diabolical artificial intelligence that can direct itself toward hostile ends.

Maybe there’s another problem, though, one that involves the market rather than questions of AI’s cognition and volition.

Where’s the Business Value?

Amid the brouhaha about evil AI, I saw this brief article and video at the Wall Street Journal. In the video, Kate Smaje, McKinsey Senior Partner and Global Leader for Technology and AI, tells the audience at the WSJLI Technology Council Summit about a widening gap between AI usage and actual business value.

Specifically, Smaje said that 80% of enterprise executives claim personal productivity gains, but only about 37% of those executives report measurable EBIT improvement. Worse, only 6% say AI is delivering significant value for investors. The video runs less than two minutes, so I invite you to see it for yourselves.

Smaje’s data and observations echo similar views expressed by consultants working directly with enterprises on AI strategies and initiatives. AI risks becoming what horse racing enthusiasts call a morning glory: a horse that performs exceptionally well in morning workouts but does not produce meaningful results in actual races. In the case of AI, it looks good and seems to have enormous potential — and its narrative of automated intelligence is compelling to C-suite enterprise executives for various reasons — but subpar quantitative outcomes belie the surface appeal. There’s less than meets the eye — on current evidence, anyway — and that’s a problem for an industry that is trying to make a sustained case for unprecedented enterprise patronage.

The cost of providing AI to enterprises is high, for burgeoning datacenters and their ongoing energy consumption and for the processing and network infrastructure that resides within those mammoth datacenters. The AI business is an expensive proposition.

To attain profitability, AI purveyors must, at some point, get enterprises to more than cover these costs. Af first, AI providers could defray or otherwise subsidize prices of these services, providing them at discount to enterprises and incurring temporary losses.

Similarly, AI purveyors could use the crack-dealer approach — delivering the first few hits at bargain-basement prices — anticipating that they’ll have an opportunity to increase prices after enterprise customers are hooked on the product. There are various ways to square the circle, but they all depend on AI having the staying power of sustained enterprise value. Early evidence, though, suggests that whatever high enterprises are experiencing doesn’t last long, and it may not leave them wanting more.

Keeping the Faith Gets Harder

This is a serious conundrum for the cap-heavy AI ecosystem, even for those now prospering enormously from supplying high-priced, high-margin infrastructure. If the pull-through of enterprise demand doesn’t materialize sufficiently and remain strong, any future demand for more infrastructure becomes untenable. Sure, you can punt the ball down the road, or make an appeal to faith, saying that the demand will ultimately meet exalted expectations, but that story has a limited shelf life. At some point, the margin call will come.

Recently, Sam Altman said he would postpone the OpenAI IPO, which was believed to be on tap for later this year. He said that it was unseemly for OpenAI to pursue an IPO in the shadow of allegedly pernicious AI agents.

Maybe he’s being sincere. There’s also a possibility that the real reason he’s postponing the IPO is that the result, if it were to proceed in the forthcoming months, would be less than stellar. He has a valuation in mind, and current market demand for AI might not support his high expectations.

Will Anthropic also delay or postpone its IPO? We’ll see.

Meanwhile, Michael Burry, investor of Big Short renown, contends that warnings of an AI apocalypse are in service to cynical IPO strategies. Burry explained his view on Substack, but Inc. covered his musings. Here’s an excerpt from that article:

Burry’s view is rooted in the economic reality of the AI boom, rather than dystopian fears of the near future. Both Anthropic and OpenAI have been preparing for long awaited, historic IPOs at over trillion dollar market caps. Bury calls warnings of impending AI doom potential “cover for real uncontrollable slowing growth as IPOs look to be pushed out.” 
Investors have been pouring hundreds of billions of dollars into Anthropic and OpenAI in recent years, as the companies lead the race to build frontier models. Anthropic employs a certain accounting metric called adjusted operating income that removes non-recurring costs, according to the Financial Times. The financial reporting technique is unapproved by the Generally Accepted Accounting Practices (GAAP) outlined by the Financial Accounting Standards Board. GAAP is generally a guidepost for the Securities and Exchange Commission and all major companies, and many have cited Anthropic’s recent financial reporting of two consecutive quarters above 80 percent margin growth as the product of some sly, and possibly suspect, non-GAAP bookmaking. 
OpenAI’s losses were more straightforward in scope: In 2025, the company grossed $13 billion in revenue but lost $34 billion, according to leaked documents obtained by AI critic and blogger Ed Zitron. The financial picture has prompted some analysts to declare that OpenAI is poised to run out of money—which, if true, could result in a seismic and economically-ruinous dud of an IPO: Altman has announced the company will push back going public until 2027, telling Fortune last weekend “right now would be an ill-advised time.” 

Something’s Happening Here

As we’ve discussed previously, prediction is a chancy business. Nobody knows precisely what will happen tomorrow, much less next week, next month, or next year. Some expectations are more credible than others, but all are subject to contingency, to events that are beyond our control and even any reasonable anticipation.

Burry could be wrong; he could be right.

What I will say to you now is that while we cannot know the exact contours of the things to come, we can prepare ourselves for certain eventualities. I have learned, as one who has lived through booms and inevitable busts, that it’s wise to mitigate risks wherever and whenever possible.

Regardless of where you stand on the specter of an AI apocalypse, can’t we all agree that we’re living in a strange, seemingly deranged timeline? I mean, take a look around. Craziness abounds, in governments and in boardrooms, and now we’re actually entertaining nightmarish visions of being annihilated by our machines.

Where you have feverish behavior, panic often follows. The largest purveyors of AI models and their CEOs are behaving oddly, marketing their technologies as if they’re Halloween slasher films starring murderous models and agents. You can take these industry moguls at their word, of course, but, again, we live in a timeline where dishonesty is practiced with impunity. Given the circumstances, I think it’s better to protect yourself, including your investments.

AI will likely be with us for a long time and it will probably establish itself in numerous applications and industries, but the bar for its initial success has been set stratospherically high. I’m seeing indications that you should brace yourself for a market shortfall.

Adjust your portfolio accordingly. You might leave some money on the proverbial table by taking defensive action now, but moving too late could cost you a lot more.