Self-learning AI agents will reshape web experiences, say top scientists

Artificial intelligence is on the brink of a paradigm shift toward the “Era of Experience,” according to a new paper by renowned AI scientists David Silver and Richard Sutton. This emergent phase will see AI systems rely less on human-provided data and instead improve themselves through direct interaction with the world. The concept has profound implications for enterprises, which must prepare to build applications that accommodate not just human users but autonomous AI agents that will increasingly navigate digital environments independently.

The big picture: Silver and Sutton, scientists with established track records of accurate AI predictions, argue that progress from supervised learning based on human data is slowing, necessitating a fundamental shift in approach.

• Their vision suggests AI systems will soon gather their own experiential data, reducing dependency on human input for continued advancement.
• The validity of their predictions is supported by features already emerging in today’s most advanced AI systems.

Four dimensions of the shift: According to the authors, future AI systems will transcend human-centric limitations across several key areas:

• Instead of disconnected episodes, AI agents will develop continuous streams of experience that progress over extended timescales, similar to human development.
• These systems will act autonomously in real-world environments, continuously gathering observational data.
• AI agents will design their own dynamic reward functions that evolve over time rather than relying on static human-defined parameters.

What it means for enterprises: Companies will need to build digital infrastructure that accommodates autonomous AI agents as primary users alongside humans.

• This includes developing secure and accessible APIs specifically designed for agent interaction.
• Creating “discoverable” agents that can be found and utilized by other AI systems becomes a priority.
• User interfaces will need redesigning to provide appropriate access to actions and observations for AI agents.

Why this matters: If Silver and Sutton’s vision materializes, the digital landscape will soon host billions of autonomous AI agents performing tasks across the web.

• This represents a fundamental shift in how digital services are consumed and how applications must be architected.
• Enterprises that prepare for this transition early will have competitive advantages in an ecosystem where AI agents become primary users of digital services.