Latest / Deep Values: Value Investing Deep Dives / Aswath Damodaran: Why AI Success Might Crash Markets
Transcript
- 0:00Welcome to the Deep Dive. We are really thrilled to have
- 0:02you with us today. For the last couple of years,
- 0:04we've all been sold this really sunny, utopian upside of
- 0:09artificial intelligence. Oh absolutely.
- 0:10The whole frictionless future narrative.
- 0:12Right. We've been hearing it from
- 0:13salespeople, tech advocates, visionary founders, and they all
- 0:17paint this picture where AI just solves our hardest problems and
- 0:21ushers in this golden age of leisure.
- 0:23Yeah. But recently, if you've been
- 0:25paying attention to the markets or the broader business world,
- 0:28the cracks are definitely starting to show.
- 0:30They are. And it brings up a really
- 0:31counterintuitive question. What if the problem we are about
- 0:34to face isn't that AI fails to deliver on its promises?
- 0:37What if the disaster scenario happens because AI works too
- 0:41well? Exactly.
- 0:42It is a fascinating tension to observe right now.
- 0:44I mean, the public narrative is visibly shifting from unbridled
- 0:47optimism to some very real, very tangible economic anxieties.
- 0:51And the mission of today's deep dive is to cut right through
- 0:55that hype. We are going to figure out
- 0:57exactly how AI might disrupt the global economy, which jobs are
- 1:01actually on the chopping block, and most importantly, how you
- 1:04can prepare yourself for what comes next.
- 1:06Yeah. And to do this, we are drawing
- 1:08heavily on a brilliant, incredibly thorough breakdown by
- 1:11the renowned financial expert Aswath Damodaran.
- 1:14But before we get into the meat of that analysis though, they do
- 1:16have a little bit of housekeeping to share with you
- 1:18to access more advanced researches, applications and our
- 1:21latest published personal Intelligent Multi Agent Deep
- 1:24value Intelligence. Welcome to subscribe on Deep
- 1:27Values dot AI and Research dot deep values dot AI.
- 1:30Definitely check that out. Also to track total market
- 1:32temperature. Welcome to Search and download
- 1:34iOS Apple Buffett Indicator from Apple App Store.
- 1:38And as a quick disclaimer before we begin, all of this list is
- 1:41based on publicly available sources and may be subject to
- 1:43revisions or differing interpretations.
- 1:45All contents from this channel is for educational and research
- 1:48purposes only and does not constitute investment advice.
- 1:51Conduct your own due diligence before making investment
- 1:53decisions. And if you like to access wider
- 1:55topics, including life and investing philosophy, please
- 1:58subscribe to our sister channel at Deep Values.
- 2:01Don't get to like this video? Much appreciated.
- 2:03Always appreciated. Speaking of deep values, the
- 2:05market is aggressively recalculating the value of AI
- 2:09right now. OK, let's unpack this.
- 2:12Where exactly did this shift in the AI narrative actually start?
- 2:17Well, the catalyst really happened during the most recent
- 2:19round of Big Tech earnings reports, we saw these massive
- 2:23technology companies announce 10s of billions of dollars in
- 2:26new ongoing AI investments. Staggering amounts of money.
- 2:30Right now, if this were a year or maybe a year and a half ago,
- 2:32the market would have thrown a parade.
- 2:33The stock prices would have surged on the sheer promise of
- 2:36innovation. But this time.
- 2:37They didn't, no. The market pushed back.
- 2:39Stocks actually went down. Investors are suddenly asking
- 2:42the hard questions about the return on all this capital
- 2:44expenditure and, more importantly, the collateral
- 2:47damage it might cause to the rest of the economy.
- 2:49Which leads us right to this really jarring report that
- 2:52Damodarin highlights, published by Satrini Research.
- 2:55They didn't just put out a vague, ominous warning about
- 2:58robots taking over. No, they got highly specific.
- 3:01They circled a specific date on the calendar, June 30th, 2028,
- 3:05just a couple of years from now. Yeah, the Citrini AI assessment,
- 3:08it's what is essentially A macroeconomic doomsday scenario.
- 3:12By that exact date, we are talking about unemployment
- 3:15rising rapidly above 10%. They predict the broader markets
- 3:19will be down 40%. Wow, the economy effectively in
- 3:23a state of severe shock and shambles.
- 3:26But the mind bending twist here in the Citrine report isn't that
- 3:29AI turns out to be a massive bubble that bursts, right?
- 3:32It's not a failure of the tech. It is the exact opposite.
- 3:35The doomsday is triggered because AI succeeds, it works
- 3:38incredibly well, and it's adopted incredibly fast.
- 3:40So it's a victim of its own success.
- 3:41Exactly. The scenario projects that AI
- 3:44replaces highly paid white collar workers at a blistering
- 3:47pace. We're talking lawyers, mid level
- 3:49bankers, management consultants, the.
- 3:50People who usually feel perfectly safe from automation.
- 3:53Right. And because those workers lose
- 3:55their jobs so suddenly, they lose their incomes.
- 3:57When a massive swath of the upper middle class loses their
- 4:00income overnight, consumer spending just drops off a Cliff.
- 4:04And that consumption void takes the broader economy down with
- 4:07it. It is a story of catastrophic
- 4:09success. I mean, we have spent the entire
- 4:1121st century worshipping at the altar of disruption, but
- 4:14destruction comes with a very real cost to the people working
- 4:17in those disrupted spaces. Here's where it gets really
- 4:20interesting, because that report clearly hit a nerve on Wall
- 4:23Street. You had Goldman Sachs, Moody's,
- 4:26JP Morgan all jumping into the fray, but the most surprising
- 4:29rebuttal came from Citadel. Yes, Citadel.
- 4:31Citadel was a massive hedge fund and they notoriously avoid
- 4:35wading into these public speculative business debates.
- 4:38So why do they step in and how do they dismantle the Citrini
- 4:41doomsday scenario? Well, Citadel's assessment
- 4:44contests the idea of an overnight economic collapse
- 4:46using three very solid pillars. First, they looked at the
- 4:50current hard data on layoffs. Which is what everyone should be
- 4:52looking at. Right.
- 4:53If AI were causing a silver employment Cliff, you would
- 4:56logically see it in software engineering first, since coding
- 4:59is directly targeted by current AI capabilities.
- 5:02But we haven't seen a Cliff. We've seen about a 6% drop in
- 5:06software jobs over the last couple of years.
- 5:08Only 6%? That's not a Cliff at all.
- 5:10Exactly. Citadel points out that this is
- 5:12much more likely just normal attrition and market correction
- 5:16after a massive period of over hiring during the pandemic.
- 5:19So the data tells us it's a dip, not a devastation.
- 5:22What is their second pillar? History Citadel looks back at
- 5:25past technological disruptions like the PC revolution in the
- 5:291980's, the dawn of the commercial Internet in the late
- 5:3190s. Right, those didn't happen
- 5:33overnight either. They took real, agonizing time
- 5:35to integrate into the corporate world.
- 5:36They always move in a pattern of three steps forward, 2 steps
- 5:39back. thinkaboutthe.com Boom. You had the initial hype, the
- 5:43massive crash in 2001, and then a slow, grueling decade of
- 5:46actually building the infrastructure that changed the
- 5:48world. So disruptions never happen in
- 5:51the highly compressed overnight timeline that the Citrini
- 5:54scenario suggests. They just don't.
- 5:56And that brings us to their third pillar, which is
- 5:58macroeconomics. How does Citadel justify the
- 6:01math of a massive tech roll out without an economic crash?
- 6:05Well, when you disrupt an existing business and replace
- 6:07human workflows with a new technology, the driving force
- 6:10for that change is a massive productivity benefit.
- 6:13You are producing more output for less cost.
- 6:16Makes sense? Citadel points out that
- 6:18historically, we have never seen a major disruption that creates
- 6:21A productivity benefit of this magnitude simultaneously caused
- 6:24caused the market and the broader economy to suffer as
- 6:27severely as Satrini predicts. You might have localized short
- 6:30term adjustments in specific sectors, sure.
- 6:32Right, but a massive society wide productivity boom does not
- 6:36historically equal an economic wasteland.
- 6:38So we have these two massive financial powerhouses completely
- 6:41at odds. One says economic ruin by 2028,
- 6:44the other says it's a slow, manageable, historically normal
- 6:47evolution. Why is it so hard for the
- 6:49smartest, highest paid people in finance to figure out what will
- 6:52actually happen? What's fascinating here is that
- 6:54finance and economics have a systemic blind spot when it
- 6:58comes to dealing with true uncertainty.
- 7:00The industry relies heavily on what we call point estimates.
- 7:03Estimates. Yeah, a point estimate is when
- 7:05an analyst makes their best guess on revenue growth, margins
- 7:09and risk and then spits out one exact specific number for
- 7:13companies future value. But picking one exact number for
- 7:17a future event guarantees you will be wrong. 100% of the time.
- 7:21But people in finance have been historically resistant to
- 7:24adopting probabilistic statistical approaches like
- 7:27giving a range of possible outcomes for a few
- 7:29interconnected reasons. Like what?
- 7:31In the past it was simply too computationally expensive to run
- 7:34complex multi variable simulations.
- 7:37Today, obviously, that excuse is gone.
- 7:39A second reason is that many finance professionals are
- 7:41actually quite uncomfortable with deep statistics beyond
- 7:44basic bell curves. But the third and most powerful
- 7:47reason is purely psychological. Let's dig into that psychology.
- 7:51What is stopping a brilliant Wall Street analyst from just
- 7:54admitting that the future has multiple possible outcomes?
- 7:57The fear of looking weak. Really just ego.
- 8:00It's the culture. If a highly paid financial
- 8:02analyst goes to a demanding client and gives them a wide
- 8:05range of probabilities, openly admitting how uncertain the
- 8:08future of AI is, it is almost universally viewed as a sign of
- 8:12weakness or lack of conviction. The culture demands alpha, so
- 8:17they make these highly specific aggressive point estimates and
- 8:20present them as absolute truth to project confidence.
- 8:23And that psychological trap leads to these dramatic binary
- 8:26models like the ultimate worst case scenario.
- 8:29Exactly. And Damodaran argues that
- 8:31modeling a true worst case scenario is practically useless
- 8:34for decision making. I mean, if you build a model
- 8:37where the entire economy collapses and unemployment hits
- 8:3920%, it paralyzes you. No investment looks good in that
- 8:42environment. It might generate clicks and
- 8:44shock value, but it doesn't actually give you any actionable
- 8:46steps to take today. Precisely that is why Damodarin
- 8:49introduces a much better mental model for evaluating all these
- 8:52wild AI claims. He calls it the three P test I.
- 8:55Love this framework. It's so useful when you hear a
- 8:57scenario about the future of AI, you need to filter it through 3
- 9:01distinct buckets. Is it possible, is it plausible
- 9:04and is it probable? Let's define those buckets for
- 9:07you listening, because this is a tool you can use everyday.
- 9:09First, possible just means it isn't a fairy tale.
- 9:12The laws of physics allow it. Citrine's rapid doomsday.
- 9:15Technically it is possible. Next is plausible.
- 9:19This raises the bar significantly.
- 9:21For a scenario to be plausible, there needs to be some
- 9:24historical precedent or some existing measurable data to back
- 9:27it up. And finally, probable.
- 9:29This is the highest hurdle. A probable scenario means the
- 9:32data shows it is already happening and based on current
- 9:35observable trajectories it is the most likely outcome.
- 9:38So think about the headlines you scrolled past this morning about
- 9:40AI taking over the world. Which bucket do you think the AI
- 9:43doomsday falls into? Do they pass the possible test
- 9:46or are they actually probable? This raises an important
- 9:48question, because to answer what is probable, we have to
- 9:51establish the framework of how AI disruption will actually
- 9:54unfold in the real world. Damodarin breaks it down into
- 9:57two core variables. First, will AI broadly displace
- 10:01workers entirely, or will it just act as a powerful tool for
- 10:04them to use? The classic displacement versus
- 10:06tool debate. Yes, and the second variable is
- 10:09speed. How quickly will this disruption
- 10:11happen? Are we talking months or
- 10:13decades? To find the probable answer, we
- 10:15have to look past the Twitter hype and look at the data we
- 10:18have right now. And we do have data.
- 10:20The Federal Reserve in Saint Louis actively tracks AI
- 10:22adoption and its productivity effects, and while survey data
- 10:26can be a bit noisy, the trends are clear.
- 10:28AI adoption is definitely climbing, but it is not
- 10:30exponential yet. The curve hasn't suddenly gone
- 10:32vertical. And here's a really surprising
- 10:35data point from that Fed research.
- 10:37People are actually using AI more in their personal lives
- 10:40than they are at work. That is fascinating.
- 10:42It tells you a lot about the friction inside large
- 10:44corporations. Plus the time savings at work
- 10:47are currently quite modest. The highest reported time
- 10:50savings are in company management and technology roles,
- 10:53but even there we're talking about marginal efficiency gains,
- 10:56not an overnight revolution. Right, by combining those two
- 10:59variables, speed and displacement, Damon Darren
- 11:02builds A disruption matrix. This helps visualize the future
- 11:06in the top left quadrant. If you have speedy disruption
- 11:09combined with mass displacement, that is the Citrini doomsday
- 11:12scenario. Imagine entire corporate legal
- 11:14departments wiped out in 18 months.
- 11:16Exactly. Massive short term benefits for
- 11:18AI companies, but catastrophic short term costs for the broader
- 11:21economy due to rapid job losses. Then you're the next quadrant.
- 11:25Slow disruption with displacement.
- 11:26He calls this the delayed detonation.
- 11:29What does that look like in practice?
- 11:30Think of the accounting industry.
- 11:32AI might not replace every CPA tomorrow, but over a 10 year
- 11:34period the software slowly eats away at the entry level auditing
- 11:37tasks. Firms stop hiring junior
- 11:39accountants. A decade later, firm sizes are
- 11:42slashed in half. So the benefits grow slowly and
- 11:45the devastating cost to the workforce are pushed into the.
- 11:47Future correct. On the other end of the matrix
- 11:50you have slow disruption where AI acts purely as tools.
- 11:54This is the disruption, Yeah, it has a very low macroeconomic
- 11:57impact. Microsoft Copilot just writes
- 11:59slightly better emails for us. We get faster spell check, but
- 12:02nothing fundamentally changes about the labor market.
- 12:04So when we run this entire matrix through the three P test,
- 12:07we arrive at the probable outcome.
- 12:10The probable verdict is that disruption will certainly
- 12:12happen, but it will be highly uneven and localized.
- 12:16AI will completely displace workers and some specific
- 12:18businesses, and it will provide game changing tools in other
- 12:21areas. But overall, the shift will be
- 12:23incremental, messy and slow. Which brings up a great point
- 12:26about human nature. One of the main reasons this
- 12:29disruption will be slow is simply human reluctance.
- 12:32People, especially older workers or those in established legacy
- 12:35industries, fiercely resist changing their daily habits.
- 12:39Oh, absolutely. Plus, for all the magic of AI,
- 12:42it still has major hallucination glitches to work out before any
- 12:46sane CEO hands over total autonomy to a machine.
- 12:50If we look at the history of the PC revolutionorthe.com boom,
- 12:55none of them happened in a straight frictionless line.
- 12:58Humans are naturally resistant to paradigm shifts.
- 13:01AI will face those exact same cultural and psychological speed
- 13:04bumps. So what does this all mean?
- 13:05Like, let's bring this down to the individual level.
- 13:07We need to talk about the vulnerability checklist.
- 13:09Yeah. How do you, listening to this
- 13:11right now, evaluate your own career?
- 13:13Who is actually going to get hit first by AI displacement?
- 13:16The prime targets for AI displacement share a few
- 13:19distinct characteristics. First, AI will aggressively
- 13:21target young companies with young employees.
- 13:24See, that feels completely really counterintuitive to me.
- 13:26You would assume ancient, bloated corporations would be
- 13:28the most vulnerable to a sleek new technology.
- 13:31What is the logic there? It comes down to institutional
- 13:33inertia. Younger companies have
- 13:35significantly less bureaucracy. They're agile.
- 13:38A startup with 50 employees can decide on a Tuesday to replace a
- 13:42human workflow with an AI agent and implement it by Friday.
- 13:45Oh I see. Yeah, a legacy corporation with
- 13:4810,000 employees has compliance departments, HR protocols and
- 13:52decades of red tape. Bureaucracy acts as a temporary
- 13:56shield against rapid AI displacement.
- 13:58That is fascinating. Bureaucracy is a Moat.
- 14:00OK, what about the types of work being done?
- 14:02Who is next on the target list? Second are high gross margin
- 14:05businesses. If an industry has massive
- 14:08profit margins, AI startups are going to aggressively target it
- 14:11because there is so much excess profit to be captured by
- 14:14automating the work. Right.
- 14:15Follow the money. Exactly. 3rd and perhaps most
- 14:18obviously are jobs built entirely on mechanical and rule
- 14:20based tasks. But this leads to the most
- 14:22crucial differentiator Damodarn identified system protection.
- 14:26Yes, system protection. This is the ultimate hidden
- 14:29shield against AI. Software engineering is highly
- 14:33vulnerable right now because it is entirely rule based and
- 14:36crucially, it has incredibly weak system protection.
- 14:39There are no major regulatory moats, no government licensing
- 14:42boards stopping a tech startup from using an AI to write Python
- 14:46code. Compare that to real estate
- 14:48appraisers or mergers and acquisitions bankers.
- 14:50Right. By definition their day-to-day
- 14:52jobs are incredibly mechanical and rule based.
- 14:55They are essentially plugging numbers into fair value
- 14:57accounting models. But they are highly protected.
- 15:00Very protected. They are protected by courts of
- 15:01law, by rigid accounting rules and by intense, well funded
- 15:05lobbying efforts. It's the ultimate reality check
- 15:07for the tech utopians. A CEO doesn't hire an M and a
- 15:10banker and pay them millions of dollars just to get a
- 15:12mathematical valuation, and AI could do that in three seconds.
- 15:15Exactly. ACEO hires a banker for a
- 15:18fairness opinion to protect themselves from getting sued by
- 15:20shareholders in a court of law. And right now a judge wants a
- 15:25human neck to ring if something goes wrong.
- 15:27A judge is not going to accept a fairness opinion generated by an
- 15:30algorithm. Until the legal system
- 15:33fundamentally rewrites its rules, which takes decades,
- 15:36especially when human lobbyists fight back Hass, those jobs are
- 15:39protected by the system, not by the complexity of the task
- 15:43itself. If we connect this to the bigger
- 15:44picture, we have to look at the unspoken toll this shift is
- 15:47going to take on society. Every major technological
- 15:51disruption of the last 40 years has consistently tilted the
- 15:53playing field toward massive corporations, worsening wealth
- 15:57and income inequality. And there is a real profound
- 16:00psychological cost here that we rarely talk about on corporate
- 16:02balance sheets. Jobs aren't just a way to pay
- 16:04the mortgage. For many people, they are the
- 16:06very foundation of their self esteem, their status and their
- 16:09identity. We are looking at a scenario
- 16:11where highly educated white collar workers are about to face
- 16:14the exact same crushing psychic damage that blue collar
- 16:18manufacturing workers face during the massive automation
- 16:20and offshoring waves of the late 20th century.
- 16:23It is a profound shock to the system to spend 20 years
- 16:27perfecting your specialized expertise, paying off massive
- 16:30student loans, only to be told that a piece of software can
- 16:33replicate your life's work for fractions of a penny.
- 16:36Which brings us to one of the most fascinating and frankly,
- 16:39slightly terrifying anecdotes from the modern source material.
- 16:42Oh, the demodarian bot. Yes, a colleague of his at NYU
- 16:45Stern essentially built a demodarian bot.
- 16:48They took every video lecture he has ever recorded, every
- 16:50textbook he's written, every blog post he's published, and
- 16:53fed it all into a large language model to create a digital
- 16:56interactive clone of his brain. Having an AI actively trying to
- 17:00replace you is quite the motivator.
- 17:02It really is. To stay ahead of his own digital
- 17:04clone, Damadaran developed a four step survival guide.
- 17:07It is a practical strategy to outrun your digital replacement,
- 17:10and it is something everyone listening should evaluate their
- 17:13own habits against. Step one, be a generalist, be a
- 17:16dabbler. For the last 50 years, the
- 17:19global economy has fiercely rewarded hyper specialization.
- 17:23If you were the absolute best at one tiny esoteric niche, you are
- 17:27safe and highly paid. But now AI is the ultimate
- 17:30specialist. Your best defense is to move
- 17:32fluidly across different disciplines.
- 17:34Connect the dots that a highly specialized AI wouldn't even
- 17:36think to look at. Step 2.
- 17:38Use both hemispheres of your brain.
- 17:40Combine cold left brain logic with warm right brain
- 17:43storytelling. AI is incredibly, undeniably
- 17:46good at logic and data processing.
- 17:48But if you can marry that analytical rigor with human
- 17:50empathy, nuance, and narrative, you become exponentially harder
- 17:53to replace. Step three, flex your reasoning
- 17:56muscles. This is a big one for our daily
- 17:57habits. It is so easy right now to just
- 18:00open ChatGPT and ask for the answer to a complex problem.
- 18:03But if you do that constantly, your cognitive muscles atrophy.
- 18:06You have to force yourself to practice the process of
- 18:09reasoning, to endure the struggle of figuring things out
- 18:12from scratch so that you maintain your critical thinking
- 18:14skills when the bot inevitably makes a mistake.
- 18:17And finally, Step 4, which might be the most counter intuitive
- 18:21piece of advice in the modern working world.
- 18:23Maintain an idle mind. I love this point.
- 18:26We live in a hustle culture that demands constant, measurable
- 18:29productivity. But letting your mind wander,
- 18:32having unstructured, idle time to Daydream, is a uniquely human
- 18:36source of creativity. It's a space where true
- 18:39disruptive innovation happens. An AI does not Daydream, it only
- 18:43processes when it receives a prompt.
- 18:45It's a vital reminder we must fiercely protect our idle time.
- 18:49But of course, there is always the lingering skeptical
- 18:51question, what if we do all this preparation?
- 18:53What if we become generalist, we practice reasoning, we adapt,
- 18:56and the AI revolution turns out to be a total fizzle?
- 18:59What if it never really arrives the way they promised?
- 19:01Yeah, that is where Damodarin brings up a brilliant historical
- 19:04analogy to frame our thinking Pascal's Wager.
- 19:07Blaise Pascal. Right, the 17th century French
- 19:10mathematician who created a logical argument for believing
- 19:13in God. Pascal said if you believe in
- 19:15God and God exists, you go to heaven.
- 19:18Infinite upside. If you believe in God doesn't
- 19:21exist, you've lost very little. But if you don't believe and God
- 19:24does exist, you go to hell. Infinite downside.
- 19:27Therefore, the expected value of believing is entirely positive.
- 19:31How does that translate to the AI debate?
- 19:34We can apply the exact same mathematical logic to artificial
- 19:37intelligence. You should act right now as if
- 19:40AI is absolutely going to take your job.
- 19:42Assume the bot is coming for you.
- 19:43Assume the worst. Right.
- 19:45If you prepare by diversifying your skills and learning to
- 19:48leverage the technology, and the disruption does come, you are
- 19:50ready to survive and adapt. But if you prepare and AI turns
- 19:54out to be a massive fizzle, well, you haven't lost anything.
- 19:57In fact, by broadening your skills and improving your
- 19:59reasoning, you have simply made yourself a much sharper, more
- 20:02dynamic, more capable professional.
- 20:05The expected value of preparing for AI is completely positive,
- 20:08regardless of the outcome. Exactly, it's a no lose scenario
- 20:11for your personal development. Thank you so much for joining us
- 20:14on this deep dive. Keep asking hard questions, keep
- 20:16those reasoning muscles strong, and take a hard look at your own
- 20:19vulnerability checklist this week.
- 20:21We will see you next time. But before we go, I want to hand
- 20:23it over for one final thought to Mull over.
- 20:26We have spent this entire deep dive talking about AI driving
- 20:29hyper efficiency and productivity.
- 20:32But if the most advanced technology in human history is
- 20:34being designed specifically to optimize every single second of
- 20:37our working lives, what if our greatest competitive advantage
- 20:41in the future is our ability to be intentionally inefficient?
- 20:44What if daydreaming, making mistakes, going down weird
- 20:47rabbit holes, and being wonderfully, uniquely
- 20:49unproductive is the only true Moat human beings have left?
- 21:23None.