“Experience-driven AI doesn’t remove the need for human input—it changes where and how that input matters most.”
In their recent paper, Welcome to the Era of Experience, David Silver and Richard Sutton (Google DeepMind) articulate a clear thesis: AI’s next breakthrough won't come from bigger datasets or models trained on human demonstrations—it will come from experience.
They argue we are entering the third era in AI:
This paper resonated with us. It doesn’t dismiss the importance of human knowledge—but it challenges the idea that it's enough. When models are trained only on what humans already know and document, they’re limited by our blind spots and biases. Silver and Sutton argue that to move beyond imitation—to solve open-ended, dynamic problems—AI needs to engage directly with its environment.
In many of the most challenging AI applications, the real world doesn’t come neatly labeled. Data is often messy, incomplete, or constantly in flux. In these cases, systems can’t rely on static inputs—they have to learn dynamically, in context, and often under pressure. AI must learn on the fly, from limited feedback, under high stakes. The classical supervised learning setup—clean input/output pairs and static evaluation metrics—doesn’t cut it.
The Era of Experience idea validates something we’ve seen firsthand: that many of the most promising AI systems are moving toward long-horizon learning and reasoning. They're:
These aren't just "bigger models"—they’re different kinds of agents entirely.
Silver and Sutton call for a more autonomous, goal-driven, experience-based approach. But that raises its own set of questions: How should these agents be supervised? How do we know when they’ve learned something useful? How do we ensure their experiences are safe, meaningful, or generalizable?
There’s also a practical gap: very few environments today are set up to support this kind of learning. Most enterprise and research workflows still optimize for static datasets and predictable performance—not for systems that explore, fail, and iterate. That’s part of the infrastructure problem we think a lot about.
This paper doesn’t just suggest where AI might be headed—it suggests a different way of thinking about intelligence. One that’s less about replication and more about discovery. Less about modeling human knowledge and more about building systems that develop their own.
We don’t think this transition will be easy. It requires new tooling, new evaluation methods, and maybe new mental models for how we work with AI.
Crucially, it also requires people—especially domain experts—to guide, interpret, and shape these systems as they learn.
Experience-driven AI doesn’t remove the need for human input; it changes where and how that input matters most.
We agree with Silver and Sutton: if we want systems that go beyond human capabilities, we’ll need to give them room to have their own experiences. But as AI continues to evolve, it’s not about humans stepping aside—it’s about stepping in at the right moments to help steer discovery.
The future of intelligence isn’t just about replicating what we know—it’s about expanding what’s possible, together.
At Perle, we’re building tools and infrastructure to support this shift—so that AI systems can learn more effectively from experience, with domain expertise woven in where it counts most.
Reach out if you’re exploring these challenges too—we’d love to connect.
References:
Silver, D., & Sutton, R. (2024). Welcome to the Era of Experience. DeepMind.
No matter how specific your needs, or how complex your inputs, we’re here to show you how our innovative approach to data labelling, preprocessing, and governance can unlock Perles of wisdom for companies of all shapes and sizes.