Stop Obsessing Over AI Agents. Build Systems That Make Us Wise

The discourse around AI agents has become fixated on creating competitive intelligences - systems designed to outthink and outperform humans. Recent research shows cognitive outsourcing to AI reduces critical thinking capabilities - in other words, it makes us dumber. This puts us on a trajectory that plainly threatens human flourishing.

A different type of AI agent is needed - one centered on human values, that enhances human agency and fosters human-to-human trust. In a recent conversation, former Google LLM engineer Varun Godbole shared his vision for a cognitive prosthetic that would help with his ADHD - part of a broader AI system designed to make humans wiser. "Agents are chasing the wrong thing," he argues. "I'm more interested in personal agency as a human being - that's where the locus of change resides."

In fact, human-centered agents have entered the AI lexicon. A Stanford/DeepMind study distinguishes between "simulation agents" versus "tool-based agents." Simulation agents represent preferences and values of real human individuals, while tool-based agents perform tasks autonomously. But while this research offers a taxonomy for human-sourced AI agents, it doesn't consider what benefits those humans might want from their AI twins - like passive income or personal enrichment. Once again, we see technology being developed to use human data rather than serve human needs.

This pattern echoes the rise of big data nearly two decades ago. Back then, Chris Anderson, editor of Wired, boldly argued that humans should abandon attempts to make sense of the world and instead trust the outputs of algorithms. Social media platforms have since revealed where this mindset leads—our data repurposed to build systems that manipulate us rather than serve us. Now, with AI scaling at an even larger magnitude, we risk falling into the same trap. As Yuval Noah Harari warns in Nexus, humanity thrives when we trust and empower one another—not when we surrender our agency to "divine" authorities, whether religious or algorithmic.

But we can choose a different path. AI has the potential to strengthen human-to-human trust, helping us connect across languages, cultures, and borders, as I’ve explored in my previous talks and writings. It can enhance our communication, expand our reach, and deepen our relationships. But this only happens if we build AI that serves human interests rather than replacing human judgment.

It's about AI designed to restore and augment our agency. I've previously argued that AI represents the next frontier in our cognitive evolution. History proves how radical changes in human cognitive tools can alter our species' trajectory. Around 40,000 BCE, humans nearly faced extinction but survived through a fundamental shift in how we used our brains - what scientists call the Upper Paleolithic transition. This "software upgrade" in human cognition enabled our ancestors to create new tools, develop complex social structures, and ultimately flourish.

We stand at a similar inflection point. Our survival as a species may depend on how we integrate AI into our cognitive toolkit. But this time, we can and must consciously direct this evolution. We must design AI systems that enhance human wisdom and agency, that help us evolve not by replacing our intelligence but by amplifying it. As Varun argues in his blog post, From Knowledge to Wisdom, the value of AI lies in turning information into wisdom that empowers us. If we can align AI with these human values, we can strengthen our agency and ensure a future where AI makes us wiser and more competitive in an evolutionary sense.

The AI-powered wisdom upgrade could be the next leap for our species. We can't afford not to build AI, and we can't afford to build it wrong. I'd add that we can't afford not to build it for us as human individuals. Our survival depends on it."

Would love to hear your thoughts. As ever, please get in touch at natalie.monbiot@gmail.com or by responding to this email.