AI Makes Investment Judgments Harder Than Ever! Howard Marks' Latest Conversation on Private Credit, AI, and Today's Market's Biggest Undervaluation...

Ask AI · How to Protect Intuition and Prudence in the Age of AI?

“You need two things: first, a judgment about what will happen in the future; second, an assessment of how likely your judgment is to be correct.”

“No asset is so good that you can ignore its price; conversely, very few assets are so bad that they remain unattractive after becoming extremely cheap.”

On March 17, at the Pereira Weinberg Capital Markets Industry Conference in New York, Howard Marks, co-founder of Oak Tree Capital, was interviewed by Bloomberg host Lisa Abramowicz.

In the current context, this conversation’s timing is quite delicate. On one hand, AI is transforming the business world and expanding the boundaries of imagination in capital markets at an unprecedented speed; on the other hand, private credit, after years of boom, is increasingly becoming a focal point for risk concerns.

Of course, in just over 20 minutes, the discussion did not delve into broader topics like geopolitics or monetary policy. Even so, Howard Marks candidly stated that this might be the most difficult period he has ever seen for making judgments.

This conversation did not stay at the vague level of “AI is important,” but instead brought together two of the most talked-about topics today: the enormous uncertainty brought by artificial intelligence, and the nervousness emerging in private credit after a long period of prosperity.

Marks’ core view is clear: Lending money to companies is not inherently wrong; the problem is not with the asset class itself, but with the crowding in of too many participants, driving down interest rates and weakening safety margins, ultimately exposing risks.

Market attention has shifted from “Is private credit an opportunity?” to “Will private credit blow up?” This reflects a typical cycle turning point.

Marks is not a technological pessimist, but he is certainly not a naive technological optimist either. He rationally reminds us that if you bet on AI, it might be better to buy stocks rather than bonds. Because when faced with fundamental changes in business models, fixed income may not compensate for the uncertainty.

Moreover, he believes that even if AI can help organize information and generate judgments, it still cannot replace the most critical aspects of investing: intuition, prudence, human judgment, and the sense of timing to act in extreme moments.

The core thinking in this dialogue is quite simple: respecting cycles, prices, risks, and the parts of human judgment that machines cannot replace.

All these principles can help us remain calm amid chaos.

Smart investors (ID: Capital-nature) have compiled this latest conversation for everyone.

AI makes investment judgments more difficult than ever

Host: Many people are now worried about private credit. Do you think this concern is justified?

Marks: I believe that lending money to companies is fundamentally sound; it’s a solid and legitimate activity.

I’ve been lending to investment-grade companies for 48 years. In 1978, Citibank asked me to help start the high-yield bond business, and I was fortunate to be there at the beginning of that market. Over the years, our clients have generally performed well.

I estimate that about 99% of the high-yield bonds we bought have been paid back in the end.

So fundamentally, there’s nothing wrong with it. The issue is that sometimes too many people want to do this at once, all eager to compete for deals, which drives down interest rates and weakens safety margins.

Then, the old problem in the industry reappears: these products are ultimately sold to people who aren’t prepared or don’t truly understand them. When problems surface, they get a shock.

Host: Do you think the advent of AI, and its potential to replace certain functions—what some call “disintermediation”—or make some software obsolete, will change your investment approach?

Marks: I think the ongoing changes, especially the introduction of AI, make the world more unpredictable than ever before.

I can say this is probably the most difficult period for judgment I’ve ever experienced in my life.

In investing, most decisions are based on predictions of what will happen. But I’ve always believed that’s not enough.

You need two things. I wrote a book on cycles in 2018, which emphasized: first, having a judgment about what will happen; second, assessing the probability that your judgment is correct.

Not all forecasts have the same success probability.

And I think the era AI has ushered in is one of the most unpredictable scenarios I can imagine. We truly don’t know what will happen next.

By the way, it’s precisely because AI’s power is so great that it’s so unpredictable.

What it can do, what it can’t do, how much it will replace us, how many jobs it will displace—these are questions we’ve never faced before.

Host: That’s why many now worry that certain types of investments are increasingly concentrated among private fund managers. Plus, with limited transparency, it naturally causes concern.

People wonder: How are these loans made? Are leverage and contractual terms safe? Is the market nearing the end of this cycle, becoming overheated, prompting some to say “it might be time to clean out”? Are these judgments justified?

Marks: Over the past few months, especially in the last six weeks, the most common question has been “What do you think about private credit?”

A year ago, when people asked me about private credit, the tone was very different. Back then, they asked: Are we sufficiently allocated? Should we allocate more? Is it a good thing that can make money with little risk?

Having been in the market long enough, I’ve seen many similar cycles. Not economic booms and busts per se, but the rise and fall of certain “ideologies.”

It always starts with something new. Usually, it must be new because new things ignite imagination and are easiest to sell. And because they’re new, people haven’t yet seen their flaws exposed.

That’s why these cycles keep repeating. I don’t think ordinary assets form bubbles.

Steel doesn’t bubble, neither do hamburgers. The assets that tend to bubble are new technologies or financial innovations.

People buy into the promising outlook, but don’t truly understand the downside risks.

When something happens that disappoints expectations, they realize they didn’t fully understand what they were doing or the limitations involved, leading to disappointment.

The relative advantage of private credit is diminishing

Host: You mentioned before that when the market was overly enthusiastic about private credit, it might have made the public markets more attractive.

What about now? Do you think the fear in the market is enough? Is this an attractive entry point, or do we need more fear? Are investors still too greedy? Is there still excess heat in the system that needs to cool off?

Marks: Honestly, I don’t know. I don’t know what private credit is priced at today.

After the 2011 global financial crisis, banks were hit hard and faced stricter regulation, so they exited leveraged buyout financing. Non-bank lenders then stepped in, providing direct loans—financing for mid-sized leveraged buyouts.

Back then, private equity firms had strong capital needs, and non-bank lenders could demand higher interest rates and better security.

But in our industry, there’s a classic saying: “What smart people start out doing, fools will flock to do in the end.”

Initially, they make money, everyone sees it, and imitators flood in. The “specialty” aspect gradually disappears.

So I believe enough of that has happened in private credit. Rates have been pushed down, safety margins eroded.

A few months ago, maybe two months ago, I’d say the yield on public market loans was around 7%, and direct lending around 8.25%.

In my view, a 125 basis point liquidity premium is fair—roughly enough, but not generous.

In other words, the “special” quality of private credit has largely vanished. It’s now in a balanced state with the public market—an equitable trade, but no extra cheapness.

So when asked “What do you think about private credit,” I reply: “What about credit overall?”

Of course, in our industry, there’s also a saying: “talking your own book,” meaning people tend to speak favorably about their own business. Oak Tree has been in public market credit for 48 years, so we want more attention there.

But I mean, when A and B are roughly in balance, rather than putting all your money into one, it’s better to diversify a bit and buy both.

Host: Another concern now is that private credit is increasingly reaching retail investors, aiming to include those seeking some regular liquidity. Do you think transparency, valuation methods, and pricing standards need to be more formalized for this to be credible?

Marks: Clearly, I think the answer is yes.

First, some institutions—including us—can’t say they’re completely outside this. Selling private credit products, like BDCs, to retail or pension investors is a business that’s happening.

But I dare say not everyone buying these products truly understands their nature, especially that these assets are inherently illiquid and lack a real-time market.

They still buy. I think mainly because they don’t understand, not because they understand and still think it’s fine.

I was discussing this yesterday with another group of investors.

I said it reminds me of “Casablanca.” Everyone’s seen “Casablanca,” right?

There’s a classic scene. The sheriff walks into a casino bar where he’s been taking bribes for twenty years, and feigns shock: “What? People are gambling here? I’m so shocked.”

What I want to say is, the worries and discomfort investors feel now are nothing new.

These products have always been illiquid, not marked to market daily, and so on. But when everyone’s euphoric and rushing to invest, these issues are often ignored; when trouble hits, they suddenly become sources of anxiety.

Part of the reason people focus on private credit is because unseen risks tend to unsettle them.

Credit markets have experienced 17 years of low default rates

Host: Do you think the current pressures seen in some private credit funds, and the discussions around them, reflect a broader sense of tension? Could this pressure continue to grow, eventually leading to a larger default cycle or more serious problems in the credit market?

Marks: Credit is inherently cyclical.

Sometimes everything’s smooth, borrowing is easy; other times, the environment worsens, and borrowing becomes difficult.

When financing is easy, everyone fights for deals, as I mentioned, interest rates naturally fall.

When times are tough, lenders are less eager, and those willing to lend can demand higher rates.

That’s the basic cycle of credit.

Last fall, when issues with First Brands (an auto parts company) and Tricolor (a car dealership and subprime auto lender) surfaced, their bankruptcies and possible fraud shocked everyone.

Jamie Dimon once said that you rarely see only one cockroach. I later wrote a memo titled “Cockroaches in the Mine,” mixing metaphors.

In that piece, I said that when the music is still playing, funding is loose, which encourages lenders to relax standards to win deals. As a result, borrowers who shouldn’t get money can get it, and scammers can continue their schemes.

There’s an old saying in banking: “The worst loans are often made at the best times.” That’s the point.

And we’ve just gone through 17 years of good times.

The stock market bottomed in March 2009, probably around the 6th. That’s exactly 17 years ago.

Since then, the financial markets haven’t experienced a truly tough period. 2015 was mediocre, the pandemic in 2020 had a few terrible weeks, and 2022 wasn’t great either.

But from around September 30, 2022, I’d estimate the S&P 500 has doubled.

Company values haven’t doubled, intrinsic value hasn’t doubled, but prices have.

It’s been a very good period. Good times tend to make people more eager to invest, but they also weaken discipline, standards, and skepticism, replaced by FOMO—fear of missing out.

When the market is driven by FOMO rather than skepticism, bad trades happen naturally.

This isn’t new.

We’ve experienced 17 years of low default rates.

And the Fed’s actions back then caused one of the most destructive environments I’ve seen—only one year of high defaults in high-yield bonds, whereas normally it would last two years.

That crisis was the worst I’ve seen, yet default numbers were lower than many other crises.

Later, in 2020, some discussed a default rate of 15%, but it ended up around 5.5%.

People have become accustomed to “no defaults.” But defaults are normal, especially after a lot of credit has been issued, and problems start to surface.

Warren Buffett always says: “Only when the tide goes out do you discover who’s been swimming naked.”

Only in tough economic times do we see who lent too loosely and to whom. I believe that phase is still ahead.

Host: From a broader corporate bond market perspective, do current prices reflect a risk of a higher default cycle?

Marks: As I said, I don’t know the exact current prices.

But if you ask me the big picture, I’d say no.

Recently, the most common question is why the spread on sub-investment-grade credit remains at the low end of the normal range.

If spreads are still low, the conclusion is: the market isn’t really pricing in higher default worries, nor providing extra compensation for higher risk.

Betting on AI should mean buying stocks, not bonds

Host: We’ve been discussing how AI and technology are changing many assumptions. As you said, this might be the most unpredictable environment you’ve ever faced.

From your perspective, is it reasonable for giants like Google, Microsoft, Amazon—companies with huge scale and strength—to issue 30-year, 40-year, or even 100-year bonds when the future five years is so uncertain?

Marks: I think that’s exactly the point I wanted to make.

Take your example: I remember Google issued a 100-year bond with a coupon around 5.8%.

If you think carefully about that, I’d say right now, optimism and trust are prevailing, not pessimism and doubt.

When optimism and trust dominate, making truly high-return investments becomes very difficult.

High returns mean returns above the risk you’re taking.

Host: If a company is highly related to these new technologies and the outlook is huge, how would you invest? In such uncertainty, how do you pick the truly benefitting companies?

Marks: First, the impact isn’t limited to AI companies or even tech firms; many industries will be affected.

Regarding the memo I wrote on December 9 last year, I also raised a question: should we lend to AI companies or to tech firms investing in AI?

I quoted my colleague Bob O’Leary, who said that if you really want to invest in an AI-focused company, you should buy its stock, not lend to it. Because if it succeeds, you want to benefit from the upside, not just collect fixed returns.

That makes sense to me.

You just said it well: you’re lending to a company for 30, 40, or 100 years, and we can’t see what will happen in five years.

Since you’re taking on fundamental business model risks, shouldn’t you, as an owner, seek returns accordingly, rather than as a fixed-income investor just collecting a small fixed yield?

Host: Of course, AI will almost certainly impact every company.

Marks: Indeed.

And because its scope is so broad and unpredictable, the world becomes even more complex.

Honestly, I prefer the world of the past—maybe you feel the same.

Host: The era when you went to school barefoot, walking backwards?

Marks: Exactly. Back then, year after year, almost nothing changed.

But now, everything changes almost daily.

This kind of change brings vitality. In many ways, life today is much better than when I was a kid.

But it also means we can’t rely on our expectations as much anymore.

AI Won’t Give Us Investment Intuition

Host: How is AI currently changing Oak Tree’s daily work? Will it affect your hiring or team structure? Is it already a tool everyone uses?

Marks: It hasn’t changed our business model or how many people we hire. Right now, it’s mainly an auxiliary tool—helping us organize data and access information.

When I write these notes, I often ask my son for help. He’s an excellent venture capitalist. After I finish, I ask: “Want to take a look?” He says: “No need, just send it to Claude.”

But the thing is, I write on my desktop. To send it to Claude, I have to copy it, paste into an email, send it to myself, then open it on my phone, copy again, and paste into AI to send to Claude.

It takes about two minutes. And just as I hit send and scroll to the bottom, the answer is already there.

It’s astonishing. I was completely amazed.

So AI can do many such things. It has read almost everything written, remembers what it’s read, and can retrieve it. That’s very different from us.

It can also recognize patterns that lead to success or failure, and extend these patterns into the future.

And it’s unlikely to make arithmetic errors, logical mistakes, or emotional biases. It won’t get overly excited at the top or overly depressed at the bottom.

This is a huge advantage.

Whether it can outperform 80% or 90% of people, I don’t know. I just hope it’s not yet capable of surpassing 100%.

I hope humans still have some things to do, and I believe they do.

I don’t think it has intuition. Sometimes, reading a prospectus, you get a chill down your spine. AI doesn’t have the hairs on the back of its neck to stand up.

Over the years, we’ve saved clients a lot of money, mainly because we don’t deal with unreliable people.

I think we’re still better than AI at identifying “bad actors.” There are many such things.

So I believe AI understands history well, is good at pattern recognition and extrapolation. Ultimately, I increasingly see AI as a tool for prediction. It’s not answering questions; it’s making forecasts.

For example, when you write an email on your phone, it suggests the next word. It’s predicting what you’re likely to say next.

Like: “I hope you’ll go with me to the…” It might suggest “party,” or “ballgame,” depending.

It can do this because it’s read millions of sentences. In “I hope you’ll go with me to the…” sentences, there’s probably a 73.7% chance the next word is “party.”

So it can do that.

It provides us with predictions, which I interpret as hypotheses. But I wouldn’t just act on what it says—human judgment is still necessary. That step reintroduces human error, but I believe it’s indispensable.

We’re not ready to hand over the entire process to AI, but we’re happy to use its help.

We’re still far from the point where caution prevails

Host: In such an unpredictable era, do you think holding more liquidity or assets that can be quickly liquidated is more important?

Marks: You know, in September 1969, I joined Citibank. At that time, Citibank and most “money-center banks” were buying the “Beautiful 50”—the 50 fastest-growing, top companies in the US. Everyone thought these companies wouldn’t fail, and their prices weren’t high.

If I remember correctly, I bought Citibank shares on September 22, 1969. If you had bought those stocks then and held for five years, you’d have lost about 95% of your money.

Later, I was “exiled” to the bond department, which was like Siberia, and was asked to manage a high-yield bond fund.

Looking back, that was one of the luckiest things in my life. Because I started making steady, safe money on some of the worst-listed companies in the US.

This taught me that no asset is so good that you can ignore its price; conversely, very few assets are so cheap that they remain unattractive once they’re deeply discounted.

So, even with all our worries about AI and our own judgment, I still believe some assets will eventually fall to attractive levels, even in such environments.

What we’ll feel then, I don’t know—because that moment hasn’t arrived yet.

Right now, we’re not at the stage where everyone is recklessly selling everything, like in past crashes when “anything that moves gets sold.”

But I believe that moment will come. And I trust that when it does, we’ll step up. Just like in September and October 2008, when we made big moves even as most thought the financial system was about to collapse.

Host: Do you think we’re heading toward another such moment?

Marks: I have a very profound answer: who knows?

Benjamin Graham once said: “In the long run, the market is a weighing machine; in the short run, it’s a voting machine.”

I don’t know how people will vote next month. If they vote for “collapse,” it will happen. If they calm down and think “it’s okay, we’re fine,” then it won’t.

It’s that simple.

So I believe this is inherently unpredictable. But I wouldn’t rush to invest all my money just because a collapse might not happen.

I’d wait and see what unfolds. If it really happens, I think we’ll become very aggressive.

But when it will happen, I don’t know. Even if it does, you can’t immediately confirm it. You never know if prices are low enough to buy.

You can only rely on a feeling.

By the way, I’m not sure if AI has such a feeling.

I don’t think you can ask Claude: “Tell me, is now the time to buy?” and get a very clever answer.

So when you ask about AI’s role in the investment process, I think that’s a perfect example of what it can’t do yet.

Most people underestimate the impact of AI

Host: If I asked you to describe the current market sentiment, would you say it’s cautious?

Marks: I’ve always been cautious—unless I reach a point where I’m certain it’s time to act.

Because I’m inherently cautious, and that caution only loosens when reality completely overturns it.

And, ultimately, lenders should be cautious. A lender who’s not cautious or is too aggressive usually ends badly.

Because lending only goes downward; there’s no upward movement. The “up” is just the fulfillment of contracts and promises.

So I think, until we see enough disappointment and price decline to justify boldness, we’ll stay cautious.

When that moment comes, I hope we’ll be among the most aggressive and decisive players—like in past cycles.

And maybe by then, there’ll be few left willing to be bold.

Host: About a minute left. I’m curious—what do you think most people are underestimating today?

Marks: I think most underestimate AI’s impact.

About 18 days ago, on a Friday, a company called Block announced that out of 10,000 employees, 4,000 were laid off that day. 40% of staff cut in one day because AI can do their jobs cheaper and faster.

Think about what that really means. How many people truly understand the weight of this?

—— / Cong Ming Tou Zi Zhe / ——

Layout: Guan He Jiu

Editor: Ai Xuan