Artificial intelligence is shifting from tools that answer to systems that act. Agentic AI – models that can plan, search use software and data sources, and carry out multi-step tasks with limited human supervision – is moving rapidly from demonstration to deployment. This shift opens significant possibilities and issues:
First, it promises to improve how we use AI: moving beyond prompt-and-response towards delegation, where workflows, decision-making processes, and marketplace interactions are handed to systems that can complete them end-to-end.
Second, it is transforming research itself, with agents that can review literature, write and run code, analyse data, and even draft papers, raising the prospect of genuine contributions to science.
Third, it sharpens questions of governance: How do we define and bound an agent's mandate? Who is accountable when autonomous systems err, collude or conceal their actions? And what governance mechanisms or economic incentives are needed to make agentic AI trustworthy?
This workshop brings these three strands together, combining hands-on perspectives on developing and using agentic AI with critical reflection on its regulation and oversight.
We invite lightning presentations (4–7 minutes) from researchers and practitioners across different disciplines on these three strands. Presentations may cover:
- ongoing projects
- demos or tools
- early findings or new papers from the literature
- Provocations, early ideas or key questions
- and more.
Work in progress is very welcome; the aim is to map who is doing what and to spark discussion and collaboration.
If you want to present, please send a title and two to three sentences here by Monday 12 October 2019.
If you just want to participate, very welcomed!, please register here.
Programme
14:00 –14:05 Welcome
14:05–15:30 Developing & Using Agentic AI (lightning presentations)
15:30–15:45 Coffee Break
15:45–16:30 Governing Agentic AI (lightning presentations)
16:00–17:00 Discussion