The Experts Were Wrong About AI. Again.

 

Let’s explore how artificial intelligence has consistently outperformed the predictions made by experts in recent years. While many anticipated that significant AI milestones would take decades to achieve, models have frequently reached these benchmarks in just a year or two.

Key Areas of AI Progress (Underestimated)

• Mathematical Problem Solving: In a 2025 survey, experts estimated a low probability for AI to contribute to solving a Millennium Prize problem by 2027. However, AI reached this benchmark in September 2026.
• Maths Olympiad: Experts previously projected that AI would reach gold-level skills in the International Maths Olympiad by 2030, a milestone that was actually achieved in 2025.
• Theorems and Publications: A 2023 survey of over 2,000 AI researchers predicted it would take 22 years for AI to prove theorems publishable in top mathematics journals. In reality, preprint servers are now being flooded with AI-authored papers, leading to new challenges regarding content responsibility and author comprehension.

 

Moving forward, physics and other theoretical disciplines may not require a new Einstein, but rather a systematic approach to utilizing AI to read and synthesize the vast existing body of scientific literature.

 

• Programming Skills: AI's programming capabilities have far exceeded expert predictions, demonstrating rapid improvement in a short period.
• LifeCode Bench Pro Performance: Experts previously predicted that the best AI score on this difficult coding test would rise to 14% by the end of 2026. However, by May 2026, AI had already achieved a score of 53.8%.
• Task Completion Time: In April and May 2026, experts estimated that by the end of 2026, an AI would be able to reliably complete a coding job that takes a human expert about 3.4 hours. While the survey was still ongoing, a new AI model had already reached a time of 3 hours and 6 minutes

 

This rapid progress suggests that for digital tasks like coding, where institutional and physical hurdles are minimal, AI development is moving much faster than human analysts anticipated.

• Cybersecurity: A significant acceleration in AI's cybersecurity capabilities, marking a clear departure from the conservative timelines previously projected by experts.
• Rapid Capability Surge: In 2023, cybersecurity experts widely predicted that AI would not be capable of autonomously finding and exploiting system vulnerabilities for several years. However, AI reached this milestone in 2024, far ahead of schedule.
• Autonomous Cyber Defense: While there is no single agreed-upon definition, the field is evolving toward agents that can move beyond simple threat detection to engage in active defense measures, such as system hardening and recovery. Ref: CSET (Center for Security and Emerging Technology)
• Modern Cybersecurity Dynamics: Current cybersecurity now involves a race between AI-powered attacks and AI-powered defenses. While AI platforms are highly effective at detecting anomalies at machine speed, they often still require human oversight to distinguish between legitimate system maintenance/deployments and actual malicious activity.

 

Why Experts Were Wrong

• Academic Bias: Many experts surveyed work in academia, an environment characterized by slow, methodical progress. They struggle to account for the velocity of market-driven economies, especially when those industries are supercharged by hundreds of billions of dollars in investment.
• The Disconnect Between Bits and Atoms
• Digital Success: Tasks that rely purely on software such as coding, mathematical proofs, and scientific writing face almost zero friction. These areas have seen rapid breakthroughs because they bypass physical constraints.
• Physical Constraints: Conversely, domains requiring physical labor, massive infrastructure changes, or complex institutional shifts (like building power plants, mining, or changing corporate workflows) face significant, time-consuming hurdles that do not scale at the same speed as digital algorithms; institutional and logistical hurdles that will take decades.

 

Misplaced Predictions (Overestimated)

• Job Automation: expert predictions regarding job automation have been largely incorrect, with the reality of workplace changes falling significantly short of past forecasts
• Overestimated Impact: In 2020, the World Economic Forum projected that nearly 50% of work would be automated by 2025. A follow-up report in 2025 revealed the actual rate of automation was closer to 22%.
• Specific Example: Roles that were expected to be automated early on, such as truckers and taxi drivers, have seen very little disruption so far, contrary to widespread predictions

The transition of physical, labor-intensive industries is much slower than the adoption of software-based AI tools. While digital tasks (like coding or writing) evolve rapidly, changing workflows in physical companies, building infrastructure, and updating supply chains face significant institutional and logistical hurdles that take decades to overcome

• The 'Intelligence Explosion': The predictions of an imminent 'intelligence explosion' (by 2027) is proven unrealistic, given the immense physical infrastructure required to power it.
• The Prediction: Forecast from Leopold Aschenbrenner, which posited an impending intelligence explosion by 2027. This theory suggested that AI would self-improve at an accelerating rate by building massive infrastructure, including hundreds of gigawatt-scale power plants and supercomputers, facilitated by robots designed by the AI itself.
• The Critique: The scenario assumes an overly simplistic feedback loop where AI can easily command physical reality; ignoring the massive logistical, resource, and institutional hurdles involved in building power grids, mining, and large-scale industrial robotics.
• The Reality Check: This "explosion" narrative contracts with the reality of how technical progress actually happens. While software tasks (coding, math) can indeed see near-zero friction progress, physical-world change (the 'atoms' side of the equation) requires decades of labor and coordination, making an instantaneous 'intelligence explosion' that transforms the physical world in a few short years highly unlikely.

 

Future Outlook: The Bifurcated AI Trajectory

I predict a bifurcated future for artificial intelligence, defined by the stark difference between digital-only tasks and physical-world implementation

• Rapid Growth in Software Domains: Areas like mathematics, coding, and scientific research are expected to see exponential progress. Because these tasks occur within software environments and face minimal institutional or physical friction, we are entering a phase where AI will rapidly tackle long-standing theoretical problems perhaps even accelerating the discovery of new science.
• The 'Physical' Bottleneck: In contrast, industries requiring massive infrastructure, energy, and material resources (such as power plant construction, large-scale mining, and industrial robotics) will likely evolve much more slowly. These transitions will take decades, tempering the likelihood of an overnight 'intelligence explosion' that drastically alters the physical world.
• The Future of Human Contribution: Despite rapid AI gains, the speaker argues that progress will remain accessible. The near future may be characterized by a shift where AI tools enable almost anyone to function as a researcher, leading to a period of democratized scientific discovery, even as the ultimate societal impact whether it leads to total job displacement or a post-work utopia remains an open, debated question.

The Productivity J-Curve: Why Your AI Investment Looks Broken (But Isn't)

 

Your AI initiative is costing more than you budgeted.

The dashboards are not showing the promised productivity gains. In fact, some metrics look worse than before. Your CFO is asking hard questions. Your board is getting restless.

You are probably not doing it wrong. You are probably exactly where you should be.

You are in the middle of a J-curve.

---

The Pattern That Repeats With Every Major Technology

In 1987, economist Robert Solow made an observation that haunted the technology industry for decades: "You can see the computer age everywhere but in the productivity statistics."

Personal computers had colonized every office in America. Companies had invested billions. Yet the official productivity numbers showed nothing. A decade of massive investment with no measurable return. This became known as the Solow Paradox, and it terrified everyone who believed in technology.

It turned out Solow was measuring the wrong part of the curve.

Erik Brynjolfsson, Daniel Rock, and Chad Syverson (economists at MIT, Carnegie Mellon, and the University of Texas) studied this phenomenon across multiple technology waves. They discovered a consistent pattern: General-purpose technologies always show a J-shaped productivity curve.

When a transformative technology arrives (the computer, the internet, now AI), measured productivity initially declines or stalls. Not because the technology is failing. But because organizations are investing heavily in complementary infrastructure, redesigning workflows, retraining workers, and building new business models.

These investments are real and expensive. They consume resources and disrupt existing processes. But they do not show up in quarterly productivity metrics. They show up as costs.

Then, after 18-24 months (sometimes longer), everything clicks. The new processes stabilize. The learning curve flattens. Suddenly, productivity surges past the original trajectory and keeps climbing.

The J-curve is not a bug. It is the signature of transformation.

---

Why the Dip Happens with AI (And Why It's Real)

When companies deploy AI, they think the deployment is the hard part.

It is not.

The deployment takes weeks. The reorganization takes months or years.

Here is what actually happens in the dip:

1. You have to rethink every workflow.

An AI system does not slot into an existing process. It forces you to ask: "If AI existed when we designed this workflow, would we build it the same way?"

The answer is usually no.

So you redesign. You eliminate steps. You restructure decision authority. You move work upstream and downstream. The old workflow dies. The new one is not yet stable. Measured productivity during this phase: down.

 

2. You have to retrain everyone.

A developer optimized for writing code needs to learn prompt engineering, evaluate outputs, and verify correctness. A manager optimized for assigning work now needs to optimize for directing AI and managing hybrid teams. A strategist who synthesized research manually now needs to curate and judge AI-generated analysis.

None of these skills transfer cleanly. People are slower during the transition. New skills take time. Measured productivity: down.

3. You have to build new infrastructure.

You need guardrails. Monitoring. Security policies. Integration points. Data pipelines that connect AI to your systems. Probably new hardware. Definitely new tooling.

All of this is invisible in terms of user productivity. It is pure cost. Measured productivity: down.

4. Someone has to verify everything.

MIT studied this in writing tasks: a 10-hour task dropped to 6 hours with AI, yet the worker felt 40% more productive. But someone still has to read the output. Fact-check the numbers. Verify the analysis. A 2026 BetterUp Labs study found that 41% of workers received "workslop" (content that looks productive but requires rework). Each instance cost two hours of cleanup. Annual cost to a 10K-person company: $9 million.

The productivity gain was real. The verification cost was invisible.

The result: one manufacturing company in a 2025 Census Bureau study saw productivity fall 1.3 percentage points relative to non-adopters. At the tail, older firms with entrenched processes lost as much as 60 percentage points before anything turned around.

This is not accounting fiction. It is real organizational friction.

---

The Burnout Trap: When Recovered Time Disappears

AI makes work faster. It does not make it lighter.

A landmark UC Berkeley and Yale study followed a 200-person tech company after introducing generative AI. Adoption was voluntary. Employees loved it. They felt hyper-capable.

But the recovered time did not stay recovered.

Because AI made starting tasks effortless, workers filled every recovered minute with new work. The natural micro-breaks (the walk to get water, the moment of staring out the window between tasks) disappeared. Workers extended their workloads without being asked. Only 8% of the time saved by AI was reinvested in their own recovery.

The result: 45% of frequent AI users report burnout, versus 35% of non-users.

The CFO's dashboard showed productivity up 40%. The employees' experience was chaos compressed.

 

---

Where Organizations Get Stuck: The Valley of Death

The most dangerous moment arrives at 12-18 months in.

You have paid all the reorganization costs. You have disrupted your existing workflows. But you have not yet received the productivity benefits. The technology is suddenly toxic.

This is the "valley of death." Companies that quit at this point typically do not try again for 5-10 years. By then, competitors have lapped them twice.

The measurement problem accelerates the exit. When companies measure activity instead of outcomes (lines of code, hours worked, tasks completed), they create pressure to show immediate output volume. This incentivizes automation of existing inefficient processes rather than redesigning them.

The companies that make it through measure something different: business value per decision, strategic insight per analysis cycle, customer impact per shipped feature. Their dashboards look worse (fewer reports, fewer tickets, fewer hours logged). Their bottom lines look better.

---

How Companies Push Through: The Climb

1. Measure the invisible costs.

Track reorganization time, training investment, infrastructure build. Do not pretend these do not exist. They are real costs. They should be in your budget. When Brynjolfsson adjusted official statistics for intangible investments, US productivity was 11.3% higher annually than reported.

2. Extend your timeline.

Most companies underestimate how long the dip lasts. 18 months is optimistic. Plan for 24-36 months for real transformation. If it happens faster, that is a bonus.

3. Eliminate ruthlessly.

The highest-value outcome of AI is not doing existing work faster. It is eliminating work that should not exist. When you can delete an entire process, do it. The temporary productivity dip from reorganization is worth it.

4. Retrain continuously.

Your people need to learn new skills. Allocate budget specifically for this. Do not expect it to happen naturally. Fund experimentation time. Build psychological safety. Recognize teams that successfully redesign workflows regardless of immediate output.

5. Measure what matters.

Stop obsessing over the activity metrics that are currently down. Start tracking the outcome metrics that will surge: decision turnaround, customer outcomes, strategic velocity, innovation cycles.

6. Stay committed.

The companies that win are not the ones with the smartest AI strategy. They are the ones that refuse to quit during the dip. They push through, keep reorganizing, and capture the gains on the other side.

---

The Historical Parallel That Matters

The pattern is not new. When electric motors emerged in the early 1900s, factory owners simply replaced steam engines with electric motors. Nothing much happened. Factories remained vertically structured around central drive shafts.

For nearly 30 years, US manufacturing productivity stagnated despite superior technology.

The inflection point came in the 1920s, when managers fundamentally redesigned factories around electricity's unique properties. They realized electric motors allowed unit drive (giving each machine its own small motor). This enabled horizontal assembly lines, single-story layouts, and continuous material flow.

That redesign, not the motor, was where the productivity lived.

AI is walking the exact same path. Simply bolting AI onto legacy processes generates modest gains. The transformational gains arrive when organizations redesign around AI's unique properties: on-demand intelligence, natural language interfaces, and reasoning across domains.

The companies that understand this will look like they are moving backward for 18-24 months. Then they will move forward at a pace their competitors cannot match.

---

The Curve Is Not Optional

This is not a choice. The J-curve is baked into how transformation works.

You can invest in AI and accept the dip. You can stay committed through it. You can emerge on the other side with transformational productivity gains that compound for years. Adopters of computerization in the 1990s saw productivity improvements that lasted for decades, exceeded initial investments, and enabled entirely new business models.

Or you can quit. Declare it a failure. Let your competitors own the next decade.

There is no third option. You cannot avoid the dip. You can only decide whether to push through it.

Most companies will quit during the valley of death. That is why the winners will be so far ahead.

AI is a challenge of leadership instead of innovation

 

AI doesn’t fail primarily due to lack of ideas or technology. It fails because leaders don’t make the hard decisions AI forces into the open.

Innovation problems are about can we build it?
AI problems are about should we, where, and under what constraints?

That’s a leadership problem.

1. AI collapses the gap between decision and consequence

Traditional innovation lets leaders delegate:

• Engineers build
• Product experiments
• Leaders review outcomes later

AI doesn’t allow that comfort.

• AI executes decisions at scale
• Errors propagate instantly
• “Small” choices become policy

Leadership challenge

• You must decide in advance what decisions are allowed to scale.
• You own failures you didn’t personally approve line-by-line.

 

2. AI exposes organizational contradictions

AI systems force answers to questions leaders often avoid:

• Do we value speed or safety?
• Growth or trust?
• Consistency or discretion?
• Efficiency or employment?

Humans can navigate contradictions informally.
AI cannot.

Result

• Leadership indecision becomes model ambiguity.
• Political compromises turn into technical debt.

 

3. Innovation tolerates ambiguity. AI amplifies it.

Innovation thrives on exploration.
AI systems:

• Act even when uncertain
• Sound confident when wrong
• Hide edge cases until damage occurs

Leadership failure mode

• Treating AI like a prototype instead of an operational actor.
• Confusing model accuracy with decision readiness.

 

4. AI shifts accountability upward, not downward

In classic innovation:

• Failure belongs to the team.
• Leaders sponsor and shield.

In AI:

• Failures trigger legal, ethical, and reputational consequences.
• “The model did it” is not a defense.

Hard truth

You cannot delegate moral agency to software.

That accountability sits with leadership whether acknowledged or not.

 

5. The real bottleneck is not data or models; it’s permission

Most AI programs stall because leaders won’t decide:

• Which workflows can be automated
• Which roles change
• Which risks are acceptable
• When humans must override the system

Teams can build models faster than leaders can grant authority.

 

6. AI forces explicit value tradeoffs

 

Innovation asks: What’s possible?
AI asks: What is acceptable?

Examples:

• Fairness vs profitability
• Transparency vs performance
• Personalization vs privacy

These are normative decisions, not technical ones.

Only leaders can make them and be accountable.

 

7. AI success looks boring, not innovative

Well-led AI:

• Quietly prevents bad decisions
• Stops scaling the wrong things
• Reduces variance, not creativity

Poorly led AI:

• Demos well
• Fails publicly
• Surprises leadership

Innovation celebrates novelty.
Leadership values reliability.

AI rewards the second.

 

The core insight

AI is a mirror. It reflects leadership clarity or the lack of it - at machine speed.

If values, ownership, escalation paths, and risk tolerance are unclear, AI will surface that confusion faster than any other technology.

That’s why organizations with strong leadership but mediocre tech outperform those with brilliant models and weak governance.

 

A simple litmus test for leaders

If a leader cannot clearly answer:

1. What decisions this AI is allowed to make
2. What data it is allowed to use
3. What failure looks like
4. Who shuts it off
5. Who apologizes publicly

They are not leading AI.
They are experimenting with it.