This past week I embarked upon two long car drives — from Philadelphia to Cape Cod and back again — with an almost-two year-old. He’s a remarkably good traveler who takes everything in his stride (as long as he has his puffy stickers and you agree to put Elmo on the radio from time to time).

The biggest challenge was finding places for him to eat well: restaurants that wouldn’t bat an eyelid at a toddler doing toddler things at the table, that also would have the kinds of food that he likes. (Yes, I packed snacks. But sometimes it’s time to sit down for a real meal.)

We made it work, but it was always a bit of a guessing game. Would X chain or Y mom-and-pop accommodate his needs? I found myself wishing that there was some kind of app that would just recommend somewhere to go — particularly as he could get hungry at a moment’s notice, and faffing around with restaurant reviews while I was driving was an impossibility.

So out of interest, once I returned, I asked ChatGPT for recommendations:

Find a restaurant suitable for attending with a toddler in Elkins Park, PA. Please just return the restaurant name and address; don't share any other information.

It returns:

The Creekside Co-op
7909 High School Rd, Elkins Park, PA 19027

That’s pretty good! Creekside is a local marketplace and brewery near where I live that absolutely is suitable for hanging out with a toddler.

I tried it in a few locations. In the part of Cape Cod where we were staying, it suggested the Moonakis Cafe, which is a superb choice. In the part of Oxford, England, where I grew up, it suggested The Victoria Arms — again, no complaints whatsoever. I’d eat there with my little one in a heartbeat.

So what if I want to make this a simple app, starting on the web?

It turns out that you can replace the name of the location in the query with its GPS coordinates, which simplifies matters a great deal. You can get the user’s current location via the web geolocation API: you can then plug that straight into the query, make an API call, and forward the user directly to a Google Maps view for the restaurant. Reverse geocoding APIs, which take latitude and longitude and return a human-readable name, are also available, and might be useful — but they tend to cost money, and the API call to ChatGPT is already an expense.

More refinements are possible, of course. Most crucially, it would be helpful to know if a given restaurant is actually open; on a road trip, it would also be very useful to find restaurants that are close to the freeway. I wouldn’t mind seeing a top 3 or top 5 list with the relative merits of each one. And being able to plan a trip around toddler-friendly eating spots in advance would be gold.

What’s sort of neat about the technology is that this is all actually very feasible — in an hour or two. Using AI as an engine in this way cuts out a lot of development time, not to mention the need to gather your own database. There are obviously ethical issues related to how these models were trained, which are sure to be litigated for years to come. But here we’re not creating any new content or replacing an artist: we’re making it easier to find our way to a local business.

The next time I’m on a road trip with my little one and I see that he’s starting to get hungry, I’ll be able to hit a button on my phone and get directions to a nearby place that will probably be appropriate for him to sit down and eat at. I think that’s cool.