The Tony Hawk Paradox: When Impossible Tricks Became Routine, and What That Suggests About Reality Itself
- David Borish
- 18 minutes ago
- 8 min read
The Tony Hawk Paradox
In April 2025, Tony Hawk sat down with Lil Wayne to talk about the video game franchise that carries his name. Discussing the original 1999 title, Hawk said the development team included certain trick combinations because they were fantastical. “It was like, this would be amazing, but you can't do it in real life. Now they do that.” He named Nyjah Huston, Shane O'Neill, and Yuto Horigome, and explained that they grew up believing those tricks were possible because they had performed them in a game.
I was reading Rizwan Virk 's The Simulation Hypothesis when I heard that interview, which is the only reason the sentence landed the way it did. My work sits in enterprise AI: deployment economics, open-weight model viability, capability forecasting. I had spent two years building the Exponential Replacement Curve (ERC), a framework that projects when AI systems cross specific autonomy thresholds using METR's measured task-horizon data. The thing that kept bothering me about the ERC was not that it worked. It was how cleanly it worked. The curves were too regular for a process supposedly driven by funding cycles, talent movement, and research luck.
Hawk was describing something structurally identical. A capability appeared in simulation first. Physical reality caught up roughly fifteen years later. If you asked me last year if I would ever write a book about simulation theory, I would have replied with a confident no, but the pattern kept showing up in places it had no business appearing, then when I started researching and quickly realized that my idea was actually novel, I felt compelled to go further.
What makes the skateboarding case unusual
Most games relate to reality in one of three familiar ways. Gran Turismo simulated existing racing technique with enough fidelity that the GT Academy program converted players into professional drivers. Flight simulators have trained pilots since the Link Trainer in the 1920s, replicating what aircraft already do. SSX Tricky went the other direction with Uber Tricks so physically absurd that no snowboarder has ever attempted them, and none ever will.
Tony Hawk Pro Skater fits none of these. The Neversoft designers created combinations they believed could not be done, included them anyway for entertainment value, and then watched a generation of professionals learn to do them. Ralph D'Amato, who produced the series from 1999 to 2007, put it plainly in the documentary Pretending I'm a Superman: when you see someone do something, it becomes possible in your mind, and skaters are now performing tricks the game inspired them to try.
The 1080 is the cleanest example. Three full rotations. In 1999, Hawk's 900 at the X Games was treated as the ceiling of human capability, and the crowd went silent when he landed it. The game let players execute a 1080 with a few button presses. In 2012, twelve-year-old Tom Schaar landed one on a physical board. Schaar was born two weeks after the game shipped. He grew up in a world where the 1080 was normal.
This is not the Bannister Effect. When Roger Bannister broke four minutes in 1954, sixteen runners followed within three years, but those runners always had the physiological capacity. What fell was a belief. The Tony Hawk tricks were not believed impossible. The designers understood them to be impossible, and included them for exactly that reason.
The same sequence, chosen deliberately
The pattern would be a curiosity if it stopped at skateboarding. It does not.
Demis Hassabis, who won the 2024 Nobel Prize in Chemistry for AlphaFold's protein structure predictions, described DeepMind's methodology in the documentary The Thinking Game in one sentence: “games are always just the proving ground for their algorithms.” Master Go, then apply the techniques to protein folding. Master chess, then apply them to materials science and energy systems.
AlphaGo's Move 37 against Lee Sedol in 2016 is worth sitting with. A shoulder hit on the fifth line, a placement human players had considered suboptimal for centuries. Go has been played for more than 2,500 years, refined through generations of masters. The simulation found something that accumulated tradition had missed, and within a few years professional opening theory had shifted to incorporate it.
Robotics shows the sequence in its purest form because psychology drops out entirely.
When 60 Minutes visited Boston Dynamics in January 2026, correspondent Bill Whitaker performed jumping jacks in a motion capture suit. More than 4,000 digital copies of Atlas trained on the movement in simulation for six hours, with randomized challenges added: slippery floors, inclines, stiff joints. Those 4,000 copies accumulated roughly three years of collective practice in a world that does not physically exist. The result was uploaded to the physical robot, which performed the movement correctly on the first attempt.
The Tsinghua and Galbot LATENT system extended this in 2026 by teaching a humanoid to play full-court tennis. The researchers could not capture clean training data, so they collected rough fragments from amateurs: a forehand here, a lateral shuffle there, recorded in a space seventeen times smaller than a real court. Simulation composed those fragments into fluid athletic sequences. The physical Unitree G1 achieved 91 percent success on forehand returns and 78 percent on backhands, performing rallies that were never present in the training data.
Robots do not have psychological barriers. There is nothing to overcome. The sequence runs simulation first, physical second, because that sequence works.
Where the curves stop looking accidental
The book's third section documents capability progression across domains that share nothing except trajectory.
The high jump record has improved 45 centimeters since 1912. The mile has dropped more than 30 seconds since 1913. Marathon times have fallen nearly half an hour since the distance was standardized in 1921. Different physiological demands, different training regimes, similar curves.
The most useful finding here is negative. Macnamara, Hambrick, and Oswald's 2014 meta-analysis in Psychological Science tested the deliberate practice thesis across hundreds of studies and found that practice accounted for 12 percent of performance variance overall, 18 percent in sports, 26 percent in games, and under 1 percent in professional domains.
In the domain where practice matters most, three quarters of the difference between performers is unexplained by how much they practiced. Genetics, coaching access, and developmental environment absorb some of that gap. None of them explains why capabilities that were impossible for all humans become possible for some and then common for many, on timelines that look scheduled.
The Flynn Effect follows a similar shape on the cognitive side: roughly three IQ points per decade through the twentieth century, across dozens of countries, at a rate genetic change cannot produce.
Technology curves show it in the cleanest form. Solar cost fell from roughly forty dollars per watt in 1977 to under ten cents by 2025. Genome sequencing went from ninety-five million dollars in 2001 to about two hundred dollars by 2024. And in AI, METR's measured task horizon has followed an exponential fit with an R-squared of 0.93. In social science, anything above 0.5 counts as strong. A 0.93 fit across a process supposedly driven by contingent research breakthroughs is the number that made me start taking this seriously.
The strongest objection, which I have not fully answered
Occam's Razor cuts against me, and the book says so directly.
Conventional explanations require only documented factors: better nutrition, better training science, broader talent pools, compound engineering gains. Each is independently verified. None requires postulating a simulation. Adding a metaphysical layer on top of explanations that already work is exactly the kind of move parsimony exists to prevent.
My response is that parsimony favors conventional accounts within any single domain and looks different across domains. Explaining athletics, cognition, technology curves, and the games-preceding-reality pattern conventionally requires four separate stories with no connecting mechanism. Whether one unified explanation carrying a large ontological cost is more parsimonious than four unconnected explanations carrying none is a genuine philosophical question, not a settled one. Reasonable people land differently on it, and I say so in the book.
The Flynn Effect reversal is the objection I find most honest. Bratsberg and Rogeberg's 2018 PNAS study of Norwegian conscription data found IQ gains peaked and then declined, with both movements occurring within families, indicating environmental rather than genetic causes. Similar reversals appear in Denmark, Finland, the UK, and Australia. If capabilities unlock on a schedule, why would cognitive scores go backward?
The most defensible answer draws on Flynn's own interpretation. He argued the gains never reflected rising innate intelligence but adaptation to changing cognitive demands, a shift toward abstraction and hypothetical reasoning. In the unlocking framework, the environment is not separate from the schedule; it is the mechanism. When the demands shift direction, the metrics tied to the earlier demands stall. I flag the risk in that reframing openly, because it is exactly the kind of move that can absorb any outcome, which is what the unfalsifiability charge warns against.
Athletic plateaus are similarly real. Berthelot's 2010 PLOS ONE analysis found no improvement in 64 percent of track and field events since 1993. But records have continued falling since that study: Duplantis broke the pole vault record four times in 2025, Kipyegon took the women's 1500, Kiptum's 2:00:35 marathon in 2023 broke what had been treated as a wall. Records fall in bursts separated by plateaus. And the stagnation figure covers traditional Olympic events, not skateboarding, climbing, or gymnastics, where progression has continued and where the game-preceded-reality pattern is strongest.
The predictions
An argument like this is worthless without exposure to failure, so the book ends with five ERC outputs, each carrying a stated threshold, a projected window, and the condition that would prove it wrong.
Autonomous vehicles serve as retrospective validation. Applied to rider-only mileage paired with peer-reviewed crash data, the framework identifies the safety threshold as crossed more than a year before the conventional discourse acknowledged it, and the peer-reviewed literature has since caught up.
The four forward predictions cover humanoid robotics, brain-computer interfaces, quantum computing, and virtual reality. Humanoid robotics is modeled against the intersection of cost parity and capability maturity for general factory work. Brain-computer interfaces are measured in throughput, with conversational speech parity as the threshold. Quantum computing is tracked in stable logical qubits rather than raw physical counts, with cryptographic relevance as the line. Virtual reality is tracked in pixels per degree, with the hard threshold defined as retinal resolution sustained across a full field of view rather than at the center of the lens.
Each domain gets two windows: a cautious one using the unadjusted doubling rate from public data, and a compressed one applying the multiplicative logic across the three concurrent exponentials operating in that field. Quantum computing produces the most aggressive output relative to consensus, and the aggressiveness comes entirely from the compression logic rather than from anything I have added by hand.
In the book I give the specific dates, the falsification conditions, and the check-back schedule for all five, so a reader can score them without taking my word for anything. If three or more fail, the framework as applied is wrong and I will say so. Check back in 2030.
What the book is actually asking
I am not claiming proof. The broad simulation hypothesis is unfalsifiable and I do not pretend otherwise. What I am claiming is narrower and testable: that capabilities appear to unlock on schedules readable in advance, that games and simulations consistently precede physical reality rather than following it, and that this pattern shows up across domains with no connection to each other.
The practical consequence has nothing to do with metaphysics. In my work with Fortune 500 clients, the most common planning error is treating capability timelines as unknowable and hedging in every direction. The curves suggest otherwise. Reading them well is a planning advantage whether or not the philosophical interpretation holds.
The patterns are documented and verifiable. The predictions have dates attached. If they fail, that tells us something; if they hold, that tells us something else. Either outcome is more useful than an argument that cannot be lost.
To get notified of the book's release date, and for a chance to win a signed copy of the Tony Hawk Paradox, sign up at TonyHawkParadox.com
