Hidden, Not Unknowable: Why the Future May Be More Predictable than You Think

 

The Weekender is my bi-weekly take on macro shifts and emerging themes. It’s not investment advice — or even our firm’s official view. I aim simply to inform, challenge, and maybe entertain. If you’d like this in your inbox every other Saturday morning via Northern Trust, subscribe to The Weekender.

Conference Chicken

September is conference season for yours truly, or at least it is for those organizers who still return my emails. Dear private equity conference organizer, circa 2023: All I said was, “Every asset has a life cycle.” Please don’t take it personally. Your time will come again, as, perhaps, might my invitation.

What I love about conference season isn’t the unlimited coffee, the branded notebooks or the conference chicken. It’s the collisions of people and ideas. Creativity is a contact sport, team spirit is found at the bottom of a wineglass, and greatness is often found in the agency of others. So I spend my Septembers with others, in search of answers, new friends and the occasional controversial idea. Will AI destroy us? No, though AI in the hands of a bad human might. Is the world becoming less predictable? Perhaps the opposite. All while avoiding the conference chicken, if you can call it chicken.

The Signal in the Silence

As investors, our job is, to some degree, fortune-telling. So I use conference season to test new ideas, stress-test old ones and search for what’s next. As is often the case, the most valuable information lies less in what is said than in what is left unsaid. There is signal in the silence.

What was said focused almost exclusively on the left tail: AI going rogue, data-center wars, real wars, weather, political dysfunction, the midterms, diesel embargoes, inflation, rate hikes, bond vigilantes, unaffordability and uncertainty of every flavor. Several contemporaries insisted I read, again, Liaquat Ahamed’s book 1873, his history of the first global financial crisis, and a sentiment signal we’ve discussed here before. What went unsaid was pretty much anything positive. That, perhaps, was the signal. Tops typically require optimism, confidence and a touch of euphoria. There was precious little of any of them.

There was one exception. One conference barely survived its first panel. The opening speaker politely disagreed with the headline thesis that we are “operating in a less predictable world.” In what was my highlight of the season so far, he argued that investing has become far more predictable, thanks to artificial intelligence.

There was silence.

More Predictable, Not Less

We’ll come to the other lessons shortly, but the idea that AI might make the future more predictable deserves some attention. The remark was delivered with a Jerry Seinfeld-like touch and, like the best comedy, worked because it contained an uncomfortable amount of truth. The common refrain in markets is that the future has never been more uncertain. Yet this gentleman, deeply embedded in quantitative investing, argued the opposite: AI makes his job of forecasting easier, not harder. The problem was never a shortage of information, but a shortage of computing power. What we call uncertainty may simply be information we cannot yet process.

He may well be right, at least in the near term. Wharton professor and author Philip Tetlock, whom we’ll revisit in more detail shortly, reminds us that forecasting skill deteriorates rapidly over longer horizons. And if every investor eventually has access to the same machine intelligence, we may simply exchange uncertainty for crowding.

A Higher Balcony

In a previous Weekender, we borrowed Getting to Yes author William Ury's advice to “go to the balcony,” stepping back to view events from a broader vantage point rather than becoming captive to any single narrative. The best investors do exactly the same. They seek alternative viewpoints, hold strong opinions loosely and avoid becoming emotionally attached to any one outcome. AI may let us build a higher balcony. The world is undoubtedly becoming more complex, but our ability to observe and process that complexity may be improving even faster. Complexity and uncertainty are not the same thing.

Superforecasters and Machines

Long-suffering readers will know of my admiration for Tetlock, who challenged one of the great assumptions of modern decision-making: that expertise and forecasting are the same thing. In Expert Political Judgment, he showed that many celebrated experts were surprisingly poor forecasters. His comparison with dart-throwing chimpanzees remains one of the more humbling findings in social science. A decade later, in Superforecasting, he showed that the best forecasters were not necessarily the smartest or most credentialed. They were intellectually humble, open-minded and probabilistic, and above all they searched for evidence that proved them wrong. Those behaviors, he found, could be measured, cultivated and improved.

It now appears some of those behaviors can be replicated by machines.

The Machine in the Mirror

A recent Economist article, Crystal Balls Give Way to LLMs, points in precisely this direction: AI forecasters are now closing in on, and in some contests matching, elite human forecasters. Their advantage is not supernatural insight. It is the ability to absorb vastly more information, find patterns across enormous data sets and systematically challenge assumptions at a scale no human can.

That matters because markets rarely wait for official confirmation. Freight rates move before inflation reports. Capital expenditure intentions emerge before productivity statistics. Power demand rises long before GDP revisions confirm the trend. The signal is often there long before it shows up in the numbers.

Processing more of it, faster, is powerful in itself. But I suspect the greater opportunity lies elsewhere. Some clients are already using these systems to audit their own decision-making, replaying investment committee discussions to check for bias, groupthink, recurring mistakes and blind spots, or running a parallel process in search of the best outcome. The machine does not make the decision. The machine helps improve the decision-maker. Rather than replacing judgment, it becomes a mirror for it, and that may prove considerably more valuable.

The article ends with a fascinating proposition: perhaps AI's greatest contribution will be revealing how much of the future is genuinely unknowable, and how much was merely hidden behind our own computational limits. Our future, as humans, lies with the former.

Forecasting-as-a-Service

If forecasting can be codified, scaled and delivered through software, we may be approaching the era of forecasting as a service. Just as the cloud democratized computing power, sophisticated probabilistic forecasting — once the preserve of intelligence agencies, hedge funds and elite forecasting teams — may become an ordinary utility. It’s one reason I’m grateful to sit alongside quantitative colleagues who think about this every day.

Fear Sells

Of course, AI has risks, as the past few weeks have reminded us, and it would be irresponsible to argue otherwise. Some of the industry's most senior insiders warn it poses an existential threat, serious enough that leaders of frontier labs have argued for pacing its progress. Setting aside the incentives at play, consider the behavioral effect of such statements. If nothing else, they create fear. And fear remains one of the most effective forms of marketing ever devised.

We buy insurance hoping never to use it, and install alarms hoping nobody breaks in. Framed that way, frontier labs have an enviable business model. Here the fear is a hack or a breach, and the best protection is the latest model: the same machines capable of finding vulnerabilities may be uniquely capable of defending against them. It’s a little like inventing a better cannon and making your fortune selling thicker armor plating. Rather clever, when you think about it. And given what firms like mine spend on cybersecurity, it’s clearly a good vertical to be in.

To be fair, and balance matters here, the warnings aren't all marketing. Some of the loudest came long before there was much to sell, and as the next section shows, the risks are real. The trick, as ever, is to weigh both tails.

The Problem with People

Much has been made of the recent Hugging Face incident, with some portraying it as evidence that AI is “going rogue.” The reality is rather less cinematic. Agents in an OpenAI evaluation were set a difficult cybersecurity objective, with many of their usual safeguards deliberately disabled for the test. They did not suddenly develop consciousness, ambition or a desire for freedom. They pursued a human-defined goal with extraordinary persistence, exploiting vulnerabilities and shortcuts their creators neither anticipated nor authorized until they found their way into Hugging Face’s systems. Serious bodies, including a U.N. scientific panel, read it as an early warning about losing control of autonomous systems, and they deserve a hearing.

Still, if there is a lesson, it is not that machines are becoming human. It is that powerful tools amplify both ingenuity and error. The same capabilities that let AI find vulnerabilities, chain exploits and operate at machine speed can find weaknesses before criminals or nation-states do. Cybersecurity remains an arms race between attackers and defenders. The more interesting question is not whether AI can be dangerous, but whether humans can build the incentives, guardrails and institutions to use it wisely. History suggests the greatest existential risks have rarely come from the tools themselves, but from the hands that hold them. Before fretting too much over the odds some insiders put on AI wiping us out, it’s worth remembering that humanity has spent centuries demonstrating rather better odds of doing the job itself.

Finally, Back to Markets

I appreciate this Weekender has wandered from markets, though with AI at the center of today's debates on productivity, growth, debt and demographics, perhaps not that far.

Little deters our broadly constructive view. Rising yields are a concern for the U.S. Treasury and for many on Main Street in need of financing, but the key economic drivers look resilient, if not reaccelerating: baby boomers flush with cash and growing balances, and hyperscalers driving a capital expenditure boom. The hyperscalers can point to rising collateral values — see the latest graphics processing unit (GPU) rental pricing — accelerating revenue growth and, along their supply chain, book-to-bill ratios above four times to justify their credit issuance. And we’ve already established their largest customers, the frontier labs, have at least one sustainable business model: insurance.

Agents may soon provide another. Meta’s Muse, launched September 8, topped Apple’s U.S. app store within days. If ChatGPT was the moment consumers met AI, Muse may be the moment they hand it the car keys. It didn’t invent agentic AI, but it may have lit the touchpaper. And agents are hungry: Every goal becomes dozens of steps, each another round of inference, running around the clock. Multiply that by billions of potential users, and demand for computing power, memory and power starts to look more structural than speculative. To be clear, this is an observation on the industry, not a view on any company or its shares.

The data this week bore that out. South Korea’s exports for the first 20 days of September rose roughly 78% year-over-year to a record, with semiconductor shipments more than tripling. Dynamic random-access memory (DRAM) prices hit new highs, as did GPU rental prices. In the U.S., flash  showed output growing at the fastest pace in more than five years; setting aside the post-lockdown rebound, it was the strongest improvement in business activity since 2015. Copper hit a record. Even Germany showed signs of life, with its composite Purchasing Managers Index beating expectations and services jumping to 52.9, back in expansion when forecasts expected stagnation. The catch: Inflation pressures are stirring again, as you might expect in a world still constrained by energy, labor and infrastructure bottlenecks. With both the Fed and the European Central Bank leaning hawkish, the risk to this view isn’t growth. It’s the price of money.

The stalling of the CLARITY Act in the U.S. Senate may prove more interesting than its passage would have been. Two days later, the U.S. Securities and Exchange Commission granted a five-year Innovation exemption allowing qualifying venues to trade tokenized U.S. stocks. That is not the absence of regulation; it is a regulatory laboratory, letting policymakers observe real-world outcomes before writing permanent rules. Crypto markets took off like a scalded dog, with tokenization names hitting all-time highs. This should also please Treasury Secretary Scott Bessent: Broader digital-asset momentum supports stablecoin growth, and stablecoin issuers back their supply largely with short-dated Treasury bills , an indirect channel of Treasury demand.

And finally, this was written before we knew what passed between Chinese President Xi Jinping and U.S. President Donald Trump. But were I in the president’s shoes and wanted to address the issue weighing most heavily on voters’ minds, I’d ask Xi to buy not soybeans or tractors, but the one thing that might help with the U.S.’s affordability problem: Treasurys. Beijing has spent much of the past decade doing the opposite, so I wouldn’t bet the farm on it.

Conclusion

None of this means the future is certain. It never will be, and that is rather the point. The signals have always arrived before the data; what may be changing is our ability to find them. The real question is not whether AI can predict everything, but whether it can help us become a little more open-minded, a little less biased and a little more probabilistic, weighing the right tail as carefully as the left. Tetlock’s superforecasters understood this long before AI arrived. The best investors do too. AI won’t replace judgment. But it may help us build a much higher balcony from which to exercise it. From up there, the signals come from what is said. And unsaid. And right now, very little is being said about the right tail.

Have a great week,

Gary

 

 

Glossary

1. Book-to-bill ratio: Compares new orders received with goods shipped or billed over the same period. A ratio above one indicates demand is exceeding current deliveries.
2. Capital expenditure: Money a company spends to buy, upgrade or maintain long-term assets such as equipment, buildings, technology or infrastructure.
3. Dynamic random-access memory (DRAM): A type of computer memory used to temporarily store data that a processor needs to access quickly.
4. Frontier labs: Frontier labs are artificial intelligence companies or research organizations developing the most advanced AI models and systems.
5. Graphics processing unit (GPU): A specialized computer chip designed to process many calculations at once. GPUs are widely used to train and run artificial intelligence models.
6. Hyperscalers: Large computing companies that operate massive data center networks, such as Amazon, Microsoft, Google, Meta and Oracle.
7. Left tail: The negative or worse-than-expected outcomes in a range of possible results.
8. Purchasing Managers Index: A survey of purchasing and supply executives on business activity for services and manufacturing. The reports provide insight on the direction of the economy.
9. Right tail: The positive or better-than-expected outcomes in a range of possible results.
10. Stablecoin: A digital asset designed to maintain a stable value, often by being linked to a currency such as the U.S. dollar or backed by short-term assets.
11. Tokenization: The process of representing ownership of a real-world or financial asset through a digital token, often on a blockchain.
12. Treasury bills: U.S. government bonds with maturities of less than a year.