Reinforcement learning pioneers win Turing Award, warn AI is rushing ahead unsafely

If a civil engineer built a suspension bridge and tested its strength simply by letting morning rush-hour traffic drive across it, people would be horrified.

Yet that is essentially how today's tech giants are rolling out artificial intelligence, according to the two pioneers who just won the field's highest honor.

Testing the bridge on the crowd

University of Massachusetts researcher Andrew Barto and former DeepMind scientist Richard Sutton were jointly awarded this year's Turing Award. Often hailed as the Nobel Prize of computing, the honor comes with a $1 million prize for their work pioneering reinforcement learning.

Reinforcement learning is the trial-and-error training method that underpins modern AI. Google senior vice president Jeff Dean called it a lynchpin of AI progress and a central pillar of the current boom, paving the way for systems from AlphaGo to ChatGPT.

Instead of taking a victory lap, both scientists used their moment in the spotlight to sound the alarm on commercial deployment practices.

In an interview with The Financial Times, Barto criticized the rush to push models out to the public, comparing it to building a bridge and testing it by having people use it.

Releasing software to millions without safeguards is simply bad engineering, Barto argued. He pointed out that while traditional engineering disciplines evolved specifically to mitigate unintended harms, today's AI companies seem to be ignoring those lessons entirely.

When profit overtakes engineering

Barto did not mince words about what is causing this rush. He noted that AI companies are increasingly guided by business incentives rather than scientific inquiry or safety.

Their warnings echo those of previous Turing laureates and deep learning pioneers Geoffrey Hinton and Yoshua Bengio. Even tech leaders signed a statement in 2023 warning that mitigating the risk of extinction from AI should be a global priority, with OpenAI chief executive Sam Altman among the signatories.

Yet the commercial race has only accelerated since then. OpenAI, which once temporarily ousted Altman in part over concerns about commercializing advances before understanding their consequences, announced plans in December to restructure into a for-profit business.

Why it matters

We think this warning lands with rare authority because it comes from the foundational architects themselves. Barto and Sutton did not just watch reinforcement learning take off; they literally invented it. When the people who designed the engine tell you the brakes are missing, it is worth listening.

For everyday users and developers, this means understanding your actual role in the ecosystem. Tech companies are treating the general public as live telemetry data. If you are integrating frontier models into mission-critical workflows, you are crossing that untested bridge every single day.

It looks like the tech industry has imported the old move fast and break things ethos into a domain where the blast radius is dramatically larger. But academic criticism alone will not break the competitive prisoner's dilemma. Until rigorous engineering standards and safety mandates carry real commercial consequences, big tech will likely keep treating live users as crash-test dummies.