AI for public transit: connect data before automating decisions
A useful transport assistant should help an operator understand a disruption and decide what to do next. That requires a clear picture of the network, reliable timestamps and a practical way to communicate decisions.
G-ATAI analysis
Start with one operational question: which information does a dispatcher need during a delayed service? Map each answer to its source, update frequency and owner. Keep observations separate from forecasts so a predicted arrival never appears as a confirmed vehicle position. An interface should expose missing information instead of filling every gap with confident language.
Our suggested pilot combines a small set of trusted feeds with a read-only assistant. Compare the assistant’s incident summary with the operator’s assessment before enabling any action. Measure the time needed to understand a disruption, correction frequency and whether passenger messages remain consistent. Add automation only after the team can trace a recommendation, override it and recover when a feed stops updating.
Questions before deployment
- Who owns each feed and its freshness checks?
- Can an operator inspect the evidence behind a suggestion?
- Does the pilot improve decisions during actual disruptions?
Independent commentary inspired by MIT News. No affiliation or endorsement is implied.
