In Short:

Democracies are weaker than they have been for decades. A great wave is coming, and they are ill-prepared. AI agents may be cure as well as cause, but we cannot depend on them, nor can we simply trust that they will advance democratic values by default.

Abstract

Computational progress has always been Janus-faced for democracy. The spread and networking of computing power bolster the epistemic and communicative practices on which democracies rely [30, 38, 164]. Yet the same tools are among the most sophisticated instruments of coercion and control ever devised [47, 78, 86, 166].

Every landmark in computing—from the first digital machines to the PC, the internet, and now artificial intelligence (AI)—has provoked anguished reassessment of this tension [1, 16, 21, 29, 93, 99, 101, 120, 132, 157, 159, 160]. In 1984, Langdon Winner [159] described an enduring divide: “computer romantics” who dream that each leap forward will finally realize computing’s unkept promise for democracy; and skeptics, like himself, who think that more powerful technologies always serve the powerful first, best, and perhaps only.

For computing’s first half-century, the romantics seemed to have the better of the argument. Computational and democratic progress proceeded hand in hand. In the 21st century, however, this picture has darkened. Democratic ideals face acute pressure: just over a quarter of humanity now lives in electoral or liberal democracies, down from almost half in 2016. And while computational progress has accelerated, the public’s endorsement of the social role of computing in general, and of technology companies in particular, has recently faltered [98]. Through the mid-2010s, big-tech companies ranked among society’s most trusted institutions [36]. Since then, a cross-national backlash against platform power and digital harms [53] has spurred heavy regulation and, even in the United States, a bipartisan conviction that too few companies wield too much power.

Policymakers and the public today face another digital revolution. In the last decade, research progress in AI has taken off [17, 84, 104, 149]. We have already developed extraordinarily powerful and economically valuable analytical and generative AI tools. We are now on the cusp of building autonomous AI systems that can carry out almost any task that competent humans can currently use digital technologies to perform. Our democracies will soon be infused with AI agents.

In this paper, we explore how AI agents might benefit, advance, and complicate the realization of democratic values. We aim to consider both faces of the computational Janus, avoiding both Panglossian optimism and ahistorical catastrophizing.

We begin (Section II) by defining key terms and introducing our approach. We then explore AI agents’ democratic implications through three lenses. Section III examines how agents may interact with structural pressures already straining democratic institutions. Section IV identifies novel threats they could introduce. Section V outlines how to design ‘agents for democracy’ that reinforce, rather than undermine, those institutions.

Democracies are weaker than they have been for decades. A great wave is coming, and they are ill-prepared. AI agents may be cure as well as cause, but we cannot depend on them, nor can we simply trust that they will advance democratic values by default. Our urgent task is to rebuild and revitalize the institutions and practices that advance democratic values, restoring their resilience against the technological and social upheaval ahead [13].

---

About the author