Cornell researchers have received a three-year, $600,000 National Science Foundation grant to develop AI models of human decisions in complex systems. The aim is to make predictions more interpretable and explicit about uncertainty, a useful ambition when people’s choices can change the infrastructure those models are meant to understand.
Watch Desk analysis
What happened
The project will combine Bayesian deep learning, microeconometrics and graph-based data to model decision-making in connected systems. Cornell professors Ricardo Daziano and Anna Scaglione are leading the work, which is intended to cover areas including energy, mobility and technology adoption.
The researchers point to the electric grid as a case where individual decisions can ripple outwards. People may respond to electricity prices or charging options, for example, while their collective choices affect demand and conditions across the network. The project aims to model that feedback, rather than treating people as a predictable input and the system as a fixed backdrop.
Why it matters
A model that predicts what people will do is not necessarily useful if it cannot explain the trade-offs behind its forecast, show how uncertain it is, or indicate whether it still works under a new policy or technology. Those questions matter when predictions could inform decisions about infrastructure people rely on.
The grant supports research, not a finished system or a demonstrated improvement in grid operations. Still, the brief is concrete: build AI that can represent human choices in changing, interconnected environments while making its uncertainty easier to see.
Our read
This is the more interesting side of AI for infrastructure: not simply forecasting a number, but trying to understand the choices and feedback loops that produce it. The researchers have set a worthwhile target. The real test will be whether their models remain interpretable and reliable when the network, policy or people’s behaviour changes. A confident forecast is easy to admire; a useful one should also know when to hesitate.
What to watch
- Whether the team publishes methods or results showing how it represents uncertainty.
- How the models perform on energy, mobility and other networked systems.
- Whether the work can transfer across different networks without losing interpretability.
Discussion spark: For systems such as the electricity grid, should AI prioritise the most accurate forecast, or a slightly less accurate one whose reasoning and uncertainty people can inspect?
Sources and evidence
- Source update (19 October 2026, 15:48 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.