Can AI Handle Our Travel Planning? Testing Chat-GPT and Bard
by Robert McGarvey
Everywhere I turn suddenly the word I hear is AI. AI can do this, it will do that, but can it in fact plan our travels?
Remember, the AI tools are 1.0 versions. They are very, very far from perfect.
Case in point: moments ago I asked ChatGPT, the current AI leader, to plan a two night business trip from Phoenix to Dallas. Seems simple right? A fast flight via American or Southwest, stay in a hotel near the event, and that’s 75% completed.
Not with ChatGPT. Here’s how it started: “Start your journey early in the morning and drive from Phoenix to Albuquerque, which is about a 6-hour drive.”
Note: I specifically asked this: Plan a two night trip from Phoenix to Dallas. The itinerary I got back showed one night in Albuquerque, one night in Amarillo and one night in Dallas, then returning home., presumably by driving but those details are not specified.
I gave Bard the same direction and it at least put me on a plane. It also suggested a trip to the G W Bush Library at SMU; I didn’t know it was there.
For day 1, Bard suggested: “In the evening, have dinner at one of Dallas’ many great restaurants. I recommend Tei-An, Uchiko, or Fearing’s Restaurant.”
Not bad actually. Score one for Bard.
I next asked ChatGPT to plan a six day walking itinerary on the Camino Ingles in northern Spain. A big win: it started me in Ferrol, not A Coruna which is the bigger city in that part of Galicia, but a pilgrim needs to start a few miles west in Ferrol if he/she wants a Compostela, an official document testifying to completion of a pilgrimage to Santiago.
As for Bard it returned a route of 93 kilometers…which is shy of the 100 minimum to qualify for a Compostela.
The ChatGPT route involved 117 KM.
The Stingy Nomads stages route – a human created route; I’ve found them reliable on past Caminos – comes in at 116 KM.
Score one for ChatGPT.
Next test: I asked both to plan a three night trip to Belfast to see the history of “the Troubles.”
Bard did ok, suggesting a Black Taxi tour to travel the Falls Rd, the Shankhill, and the Crumlin Rd museum. But on Day 3 it fizzled out with a suggestion to see the Titanic Museum. A fine stop for many but it has nothing to do with the Troubles.,
As for ChatGPT, it waived a white flag: “The server had an error processing your request. Sorry about that!”
You probably heard the recent story of tourists in Hawaii who, following GPS. drove directly into a harbor. There’s video. Personally I think that incident has more to do with human error than machine error but, then again, I have not see the directions the tourists were given, nor do I know what GPS mapping tool they used. Maybe it was in fact a machine malfunction.
But back to AI and travel. My best guess is that two or three years from now, AI will be able to plan most business trips literally in seconds with high accuracy, assuming the question is framed properly. The “prompts” – as they are called in AI – are crucial in shaping the results. Poor questions will produce poor results.
Why am I optimistic that better times are just around the bend? Partly it’s because the AI tools will get better – their databases will grow and their speed will accelerate. Partly it’s because we will get better at shaping our prompts.
Give it a couple years and about all that will be left for travel agents will be trips involving enormous complexities and poor source materials – such as travels in sub-Saharan Africa or India. I’ve tried to plan such trips myself and frankly gave up.
For now, however, by all means play with AI, even ask it simple questions. But always remember the machine is fallible. I just asked Bard the best steak houses in Manhattan and came back with a list headed by Peter Luger which happens to be in Brooklyn and, well, don’t ask Pete Wells what he thinks of the joint. On this list of 10 there were at least two more that just do not belong on any list of the best steak houses in Manhattan.
I asked ChatGPT the same question and, again, it offered up Peter Luger and on its list of five one other – clearly – did not belong.
But both named Wolfgangs which is at the top of my list.
The answers will keep getting better. Give AI time.