Seoul is a vibe today. Fitting that the conversation turns a bit more futuristic at ICML. Today is about trust and alignment. I look at this through a consequential but, honestly, narrow lens: the models' reasoning and judgment and how to attach accountability. There are a few sessions I am tracking that speak to the accountability question: whether we can see and contest what these systems actually do at the moment they do it — not whether we approved them beforehand.
But one of them takes this idea even bigger: how can AI enable democracy? The pitch is that AI can lower the barrier to real participation (help people articulate a view, get balanced information, follow a complex debate) so more voices get in, while humans still make the actual choices.
Funny to fly to Seoul to end up back where I started: who decides, and can we prove what happened and what information they had? The scale changes, from an agent in a workflow to a citizen in a debate. More as the week unfolds.
