Imperial College London researcher Calvin Tsay has won an ERC Starting Grant worth up to €1.5 million over five years for CODA, a project developing AI methods for making decisions under uncertainty. The aim is to move beyond systems that predict what may happen towards methods that help decide what to do, and assess how dependable those decisions are. That distinction matters in fields such as engineering and energy, where a plausible forecast is not enough if a system must act reliably.
Watch Desk analysis
What happened
Tsay, an associate professor in Imperial’s Department of Computing, will lead the CODA project. Imperial says the team will develop optimisation and learning methods for complex real-world decisions, with applications including engineering, energy systems and scientific discovery.
The Imperial announcement describes a five-year research programme, not a finished system or a tested deployment.
Why it matters
Many AI systems are built to predict outcomes from data. But choosing an action under uncertainty is a different problem: the system must weigh possible outcomes and provide reason to trust its decision. That challenge becomes especially important where an unreliable choice could undermine the very engineering or scientific work AI is meant to help.
CODA’s focus is on mathematical foundations and methods, rather than a new consumer product. The grant gives Tsay’s team time and funding to work on that underlying problem, but the announcement does not establish that the resulting methods will deliver reliable decisions in practice.
Our read
This is a serious research bet on a less flashy, more consequential question than whether AI can make another prediction: can it help choose well when the answer is uncertain? The useful test will be whether CODA produces methods that can be evaluated in real applications, not simply a more confident-sounding output.
What to watch
- What optimisation and learning methods CODA develops over the five-year project.
- Whether the team demonstrates and evaluates those methods in engineering or energy applications.
- What evidence the researchers provide that their decision-making methods are dependable.
Discussion spark: For AI used in engineering or energy systems, what should count as convincing evidence that a decision is dependable: strong mathematical guarantees, successful real-world trials, or both?
Sources and evidence
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.