• Home
• Technology

Canadian AI pioneer and Turing Award winner Rich Sutton has warned that the technology industry’s growing reliance on synthetic data to train artificial intelligence models could prove to be a “big mistake,” arguing that data generated by AI systems cannot replace the complexity of real-world human experience.

Sutton Challenges the Rise of Synthetic Data

In a podcast interview with Business Insider, Sutton said synthetic data — information generated by algorithms or AI models rather than collected from real-world sources — cannot adequately reproduce human behavior, experience or interactions with the environment.

“We simply cannot create synthetic data that reflects the minds of others or human behaviors,” Sutton said, arguing that artificial substitutes cannot capture the full complexity of human experiences and genuine interactions.

Instead, Sutton advocates what he describes as “real-world experiential data,” in which AI systems gather information by directly interacting with the world and learning from the outcomes of those interactions.

AI Industry Faces a Shortage of Original Data

The debate over synthetic data has intensified as technology companies face growing limitations in accessing new, high-quality data generated from real-world sources.

As AI models increasingly consume data produced by other models, companies have been looking for scarce datasets that can provide access to original information. Google recently reached a $10 million deal to obtain internal data from Spirit Airlines, while OpenAI has also been seeking exclusive datasets.

Sutton argues that the industry’s focus should shift toward systems capable of continuously learning through direct experience rather than depending primarily on massive collections of pre-existing information.

Oak Lab Focuses on Learning From the Real World

Last month, Sutton and his colleague Khurram Javed launched Oak Lab, a startup developing AI systems designed to continuously learn through interaction with the world and collect the results of those experiences.

The approach reflects Sutton’s broader view that AI systems need exposure to real-world environments rather than relying exclusively on predefined datasets or simulated experiences.

“You cannot rely on synthetic data to train a robot to interact with its environment, for reality holds variables that cannot be enumerated, and any simulation remains a vast simplification,” Sutton said.

Oak Lab Remains in Its Early Stages

Oak Lab is still in its early stages and has not announced any funding round or investors.

Sutton’s criticism comes as AI companies continue to explore synthetic data as a way to address the limited supply of original training material, while researchers and technology firms increasingly examine whether systems can learn more effectively through direct interaction with the real world.