Can AI invent something as valuable as AI?

THIS WEEK’S NOBEL PRIZE IN ECONOMICS

This week’s Nobel Prize in Economics went to three academics - Mokyr, Aghion, Howitt - for studying Innovation as the central feature of sustained economic growth.

Readers of this blog should not be surprised by this finding as we have written extensively about Innovation in the context of investing - for example: “Where Does Alpha Come From”, “Investment Process vs Innovation”, “Two Centuries of Creativity”. It’s also just a very intuitive finding.

Interestingly, one of the three Nobel Prize winners, Peter Howitt, was one of my professors during my undergraduate studies at Brown University. Our economics department was well known for studying models of long-run economic growth.

The department chair and my thesis advisor, Oded Galor, developed the first model to explain the shift from centuries of Malthusian stagnation (pre-1800) to modern, innovation-driven growth. He recently published a fascinating book “The Journey of Humanity” with many of his findings.

For curious readers, here is my 2004 honors thesis, written under the supervision of Professor Galor, exploring a related question—why life expectancy has risen so dramatically over the past two centuries. Yes, readers will recognize that’s where my passion for “Everything Long Run” was born.

[As a side note, today’s popular quest for Longevity might benefit from revisiting a hypothesis I proposed in my thesis: humanity’s life expectancy increased so dramatically in the late 19th and early 20th centuries not only because of scientific breakthroughs, but also because longer lifespans offered evolutionary advantages. If we are to extend human life even further, there may need to emerge new evolutionary pressures or contexts that favor or enable longer-living humans—perhaps even allowing us to “download” the knowledge required for scientific innovations that would enable survival over extended lifetimes.

For example, if our planet becomes inhabitable, multi-century space flights to seed life on distant planets could make long-lived humans evolutionarily advantageous—or some other context might provide a similar benefit to living, say, 300 years.]

During that senior-year thesis experience, I felt for the first time the electrifying freedom of being allowed to invent something entirely new within an established field like economics. Up until that point, economics had been a strictly textbook-driven discipline—an experience I later brought to Wall Street, developing a “not by the textbook” quant model that outperformed the market for many years.

In fact, our elective seminar with Professor Galor had no textbook at all. On the chalkboard, he would draw a single chart: first a long horizontal line, then a sharp kink upward around 1800 into a diagonal line. He’d label it as log-scale GDP per capita, then look pensively at our small class of ten and ask one question:

“What unified model can explain this entire history of human growth, not just the subperiods?”

That kind of powerful, clear question is the root of innovation. After months in the library, a researcher trying to answer it would inevitably experience those “aha moments” that feel like a light bulb turning on inside. These moments feel biologically real—both exciting and relaxing at once. (Anecdotally, My Oura ring even shows green relaxation dots during these times, as if the pressure to discover something new is released, even as heart rate rises from excitement.)

This is the process where, out of a gazillion possibilities and permutations, your mind attracts one intuitive idea—and your body instantly recognizes its value. Innovation isn’t just about novelty; it’s about value. People who work creatively, from scientists to artists, often describe this feeling.

If innovation requires the body to feel the creative moment, can AI—without a body—ever truly innovate?

---

Can AI truly innovate?

Now, back to the question in the title of this blog. Today’s AI is an extremely powerful machine trained on historical probabilities. It excels at tasks like summarizing writing—rearranging words and concepts in the most likely patterns.

But can AI truly innovate on its own, without human involvement?
Isn’t something truly new defined by having a historical probability of zero?

Humor is a great example. Can AI create a genuinely fresh and funny joke? A good joke, by definition, isn’t stale—it’s surprising, relevant, and releases tension in ways only humans can fully experience. So far findings (and experience) suggest that AI struggles with humor.

Can AI come up with The One research question, and discover The Insight that is both new and valuable?

Advocates of AI will say yes. And maybe, someday, it will be possible when AI is trained not just on data but on human intuition and the biochemistry of the “aha moment.” There are times when it feels like AI is already there, so precisely capturing the nuances of my intent. But soon those moments fade like mirages—projections of true intelligence.

For me, anything genuinely new, artistic, creative, and “aha-worthy” still happens inside the human body and not inside the machine. Then we bring those ideas into AI for refinement, expansion, and execution.

---

The shadow of innovation

One important “shadow” aspect of innovation that Professor Howitt studies is creative destruction—the process by which the new replaces the old, often causing the latter to vanish. This evolutionary mechanism has propelled growth and human progress.

Yet many thought leaders today worry that AI could eventually extend this process to humanity itself.

It seems to me that if AI ever truly takes over the innovation mantle from humans—and without any human input begins producing breakthroughs as valuable as AI itself, the Internet, antibiotics, the steam engine, the Beatles, or the Mona Lisa—then we might indeed become the displaced, evolutionarily extinct branch of biology.

Until that day, let’s continue creating and innovating—but also building into every innovation design the support mechanisms for those negatively affected by such innovation: the displaced, the disrupted, the destroyed.

Because someday, that could be us.

And if AI has been trained on designs that include compassion and support for those replaced, maybe it will learn to keep the humans around.

Would love to hear from readers - comment to this blog’s related post on LinkedIn or X.
Have you experienced real innovation or true creativity from AI?

Checking In

It’s been a while since my last blog. I wanted to check in, and review some topics that I’ve covered in the past.

So much has changed in the year and a half since I last wrote. For example, there is now an “AI” button in my blog editing window that is offering to write this blog for me.

At the same time, so many things are “same as ever”. For example, everyone continues see “a lot of uncertainty in the markets” and things continue to look foggy ahead and clear in hindsight.

Uncomfortable IS PROFITABLE

Unlike the generally agreed upon yet not very useful concept of “market uncertainty”, the much less appreciated and more useful concept that also stays the same is that markets (stocks, factors etc.) continue to move in the direction that is most uncomfortable to predict, because the comfortable direction is already priced in.

For example:

• Last year, it was uncomfortable to predict that inflation will be solved by now.

• This year, it was uncomfortable to predict that a full recession will be avoided.

In September, I had the honor of speaking on Meb Faber’s podcast, where we discuss this idea in more detail. I gave an example that predicting a market doubling is much less comfortable than market halfling. What is comfortable is already priced in and so it’s only the uncomfortable views (if correct) that make money. That’s why forecasters are famously always wrong - even when they are correct, their forecasts are already priced in.

What’s the most uncomfortable view can you have today? - that’s the one to watch out for.

Long-Run Evidence alleviates ABANDONMENT RISK

On the podcast, we also discussed my latest academic paper on long-run asset allocation where I use almost a century of data (building on top of other long-run work of the past 10 years) for many asset classes and factors. I run a horse race between the popular asset allocation approaches from 60/40 to Risk Parity, Endowment Based, Factor Based and Dynamic Asset Allocation. At this point, it’s not news to anyone paying attention that Dynamic allocation historically crushed the other approaches on drawdown protection in traditional growth recessions.

However in 2022, things were different. Both stocks and bonds suffered a drawdown during inflation driven correction, so Dynamic approach came down as hard as the other approaches, in the 20-30% range. Yet the absolute drawdown was still much better than the max drawdowns of 60/40, which ranges from 30-70%.

STRATEGY timing REDUCES RETURN

Unfortunately, most investors continue to ignore historical evidence of crashes and dry spells of their chosen approaches. That leads them to sell out of their allocations when either of these two risks shows up.

Poor timing contributes to the difference between dollar-weighted returns (the ones investors actually get to earn) and the time-weighted counterparts. Watching my own clients add and remove assets from their accounts based on recent deviations from trend return confirms this human tendency. Sticking with a negative deviation is uncomfortable and yet was the right choice when the underlying source of return is reliable. Remaining invested in reliable approaches continues to be the most prudent answer - albeit much easier said than done.

FactorS BOUNCED BUT ALPHA decayED

Speaking of staying invested, my 2020 blog on why value investors should not give up was well timed. The bounce back from the extreme drawdown gave factor investors a short-term relief. However, my other point stands that the long-term “alpha” in traditional factor investing is likely gone. Because factors were not some "reliable premia” as the academics would have us believe, but were basic anomalies that generated alpha by identifying types of companies that were uncomfortable to hold. Proliferation of quants and smart beta, made these approaches comfortable and unprofitable. As a quant, this is hard to admit (and i have dedicated a lot of time to showing that these factors were real over the long run and not results of datamining). But ironically, it is the contrast of the flat return of last two decades vs the positive return of the prior two centuries, that is the most alarming evidence of factor’s decay.

For anyone who still doubts that factors can get arbitraged away just watch how the top hedge funds manage their capacity. Most of the top funds are closed to new investors and have been closed for a long time. These funds are giving up tens (if not hundreds) of millions of fees by not accepting the easily available additional demand. Something that traditional asset managers would gladly accept. Why do they forgo these easy profits? Because additional AUM would eat into their ability to generate alpha and hence their long-term profits. So if a hedge fund choses to close at 50billion to avoid the risk of alpha decay (where alpha is made up of hundreds of signals), how can traditional factors take in a couple trillion of AUM into a small handful of factors and still maintain their alpha?

• What’s is the main antidote to alpha decay? - innovation.

• What’s the main enemy of innovation? - bureaucracy.

ALPHA IN Intangibles

Intangibles are one way to generate innovative alpha as they continue to play a large role in company dynamics. There isn’t a week that goes by, that I don’t see a headline about some company’s culture deterioration causing a crisis or CEO style impacting company culture. These are hard to value assets that are material to the future fundamentals. Just because they are hard to measure, does not mean they are not important. In fact, from the perspective of alpha, this difficulty of measurement makes them even more valuable.

I recently gave a series of talks on intangibles investing at QuantStrats NYC and London, Nuedata, and in this Interactive Brokers webinar that you can watch for free. The webinar is filled with examples of how intangibles can be measured in innovative ways, using language. In one example, I show how we can trace Steve Job’s language in the Apple’s 10Ks. It points to the creative direction in which he took the company when he returned as the CEO in 1998.

Although intangibles investing has proven to be a huge source of alpha over the past two decades, this approach did experience underperformance in 2022. The rise in long-term interest rates affected companies with longer duration assets, which intangibles by definition are.