Manifesto

The research base and the method live on the about page ; the published results live at MediationBench . This page is the bet we are placing on them — opinion, clearly separated from evidence.

Why this matters now

These ideas have circled HAI.AI for about ten years. In the abstract, the function of HAI.AI is to generate and maintain good interpersonal commitments, and to make cooperation possible where it was hard before.

Like many people who dedicate their careers to change, the founder has worked at startups in education technology, mental health care, and environmental air-quality data, building new technologies, validating their efficacy, and running into the same wall every time: the effectiveness of the institutions we rely on. How do you serve an institution that cannot function sufficiently? Institutions full of good people doing good work, where, inside and out, no one is as happy or as effective as they could be. We understandably want to keep our jobs and our identities, but change is constant.

We may need new institutions for large, critical human services. The bigger the institution, the bigger the disruption needed. AI will bring that disruption, and an arms race with it. We are only beginning to understand the challenges ahead. Meanwhile, we have to build what we will be fighting for.

The world we see coming

Foundation models are powerful and will become more so. In the everyday world, at home and at work, people will increasingly use AI running on smaller, distributed computers, serving as personal assistants. Even in a world of very powerful models, individuals need their own AI and their own representation. Where today we have email, in the future agents will talk, argue, exchange, and transact on our behalf. That world needs infrastructure. It needs a harness, and it needs to be observable and trustworthy, held to high standards of personalization.