Hi! I’m an interdisciplinary scientist and founder working to make sure that AI reduces destructive human conflict. I’m currently Senior Scientist at the UC Berkeley Center for Human-compatible AI.

There is mounting evidence that current AI systems are going to make our disagreements and conflicts worse, not better. For example, AI models make people more certain they are right, tell different facts to different sides, and escalate to nuclear war 95% of the time in simulations. Meanwhile, these systems are mediating every kind of human relationship from marriage to business to war. This is a catastrophic AI risk: if the machines encourage our worst impulses, we will end up destroying each other.

I study how what our machines tell us makes human conflict worse or better. Conflict isn’t inherently bad; my goal is not to prevent all conflict but to ensure that it is constructive rather than destructive. I do AI and conflict theory, execute large field experiments and build products for better conflict outcomes. This work is related to “epistemic risk” and “cooperative AI” but it’s something different — conflict is not merely the absence of cooperation.

I have been active in the international peacebuilding community for some time and write a newsletter on better approaches to the culture war. For a decade I taught the double masters in computer science and journalism at Columbia Journalism School, where I  led the development of Workbench, a visual programming system for data journalism, and Overview, an open-source document set analysis system for investigative journalists. For a while I was an editor at the Associated Press and a data journalist at ProPublica. Before that, I did computer graphics R&D at Adobe Systems. I like to build big weird art.

You might be interested in:

• My publications and talks
• Blog entries on various topics
• Contacting me through this form this form. Or, find me @jonathanstray.