Riders using AI for trip planning

[Björn Radstrom]

Winnipeg Transit has recently received several anecdotal reports that high school students and 1st/2nd year university students are using ChatGPT and other AI engines to plan transit trips on our network, instead of using our own Navigo trip planner, or the Transit App.

One of the complaints we received is that the Blue Line is the only transit route serving the University of Manitoba. This is completely untrue, but by trying to use several AI engines to plan trips, we discovered that they all give really bad information. Some tell people to use routes that no longer exist, some tell people to use that have never existed (the AI basically invented a Route 675, which we have never had), and if they provide a real trip, it’s usually a bad option. The fact that some of the AI engines direct everyone to the Blue Line probably explains why we are seeing so much overcrowding on that route!

I asked everyone I know who has high school or university-aged kids to ask them about their use of AI. The result of that totally-unscientific poll is that, although it’s certainly a minority of young people using AI to plan transit trips, it’s enough that it’s causing me concern… especially seeing as it will likely only grow.

Is this something anyone else has noticed?

0

Discussion

[Jon MacMull]

This is super interesting. AIs will pull from all available sources for that information, and that information is prioritized based on authority and credibility. It will prioritize official website content from the transit agency, google or other authoritative sources. This is similar to SEO, but the AI may grab info that is outdated and in some cases “hallucinate” in order to form conclusions. This heightens the need to ensure all info on your website is up-to-date and complete, and that old pages are fully removed. Also important to find external sites with misinformation and seek it’s removal. Using SEO on-page tactics, you can likely increase the priority of the “correct” content and the hope is that it will help the AI get things right.

0

[Björn Radstrom]

We already do exactly what you’re suggesting. We remove old information and make sure new info is up to date.

But even if AI could understand that (which it has proven it can’t), we still have two weeks every season where the active schedule for the season that is ending is still live, but new schedules are also posted for the upcoming seasonal schedule.

Using AI for trip planning just doesn’t seem to be feasible, especially not with the current disastrous state of AI.

This ended up getting a bit of media traction here in Winnipeg. See attached image.

0

[Jon MacMull]

Seems like a good approach. Not much else you can do at the moment except provide caution/warning.

0