Conor Pope in the Irish Times had this recently, organising, if that is the correct term, a holiday in Ireland using A.I. Now, having had reason to use so-called ‘A.I.’ closely and also, though not simultaneously, book various train journey’s across Europe recently it strikes me how useless the former is with relation to the latter. It took a process of hours to investigate routes, sign up to various rail providers, check connections, see that there were trains and those trains at the correct times and then purchase tickets.
Pope’s ‘A.I.’ assist was entirely cosmetic in that regard.
We asked ChatGPT to plan a weekend in Dublin for two adults and two children aged eight and 10, travelling from London on Friday, June 5th, and leaving on Monday, June 8th.
Less than 45 seconds later it was back to say it had it all worked out. Sort of.
Plenty of recommendations:
“The main London airports with affordable options to Dublin are Gatwick, Luton, and Stansted,” our AI wannabe travel agent started earnestly, if obviously. “By early June, you can find good-value returns on budget airlines if you book soon.”
It suggested an early-morning Ryanair flight from Stansted with a Monday return, and told us the total cost would be less than £200 (€231) .
The Maldron in Smithfield was recommended as the best place to stay, with the online oracle promising a room with a double bed and two singles for €510 in total.
But lots of odd discrepancies:
As itineraries go it was grand, but what intrigued us more than the activities were the suspiciously low prices. For three nights in the Maldron on these dates,the best deal we could find was €1,036, or twice what the AI bot told us.
The restaurant prices seemed out of whack too. When our fantasy family decided to have mains and desserts in Ely along with a fairly sober two pints of Guinness and two soft drinks, they easily wound up spending over €150 rather than the €80 ChatGPT told us to expect.
Even when using a supposedly more ‘tourist’ oriented ‘A.I.’ the results were… mixed:
It made some basic errors, telling us to have breakfast in Brother Hubbard on Capel Street – a long schelp from Croke Park – while the Bunsen it recommended was in Dublin 8, even though there are two Bunsens much closer to the Croke Park Hotel.
When we asked for a total cost of the trip, it said no. “Airial tracks flight and hotel costs but doesn’t provide detailed budgets for activities, meals, or other expenses. You’ll need to estimate those separately based on the suggestions I provide,” came the decidedly tetchy response.
Doesn’t this suggest that cutting out the A.I. and going and doing it by engaging directly with the elements of a stay – looking at maps, estimating distance, checking out hotels, etc, is actually more time effective?
But the fundamental problem is referenced in passing and in a slightly different context in the article:
Most AI bots aggressively search the internet, leaning heavily on positive reviews scraped off travel sites.
It’s not just the reviews, all the information is scraped from there too and it has no way of assessing is it correct or not because, of course, it does not and cannot analyse what it is at. It’s a fundamental error to place any store in what information is ladled out – or to assume that it is correct.
Sure, the article goes on to talk about the future, Agentic AI which is meant to act in a more proactive fashion, but why that should be any more trustworthy is not explained. Indeed explanations are not part of the process. One person is quoted as saying:
“If AI knows what you want and the kind of hotel you like to stay in and the experience you like, it can, over time, build up a picture of you,” he says. “At a conference recently I was told that we are not living in an era of change but in a change of era, and that really encapsulates it. It is fundamentally different way of thinking and doing things, and how travel companies react will be fundamental to their survival.”
What does this mean?
Surely the material reality is that there are hotels and they have rooms which they offer at various prices. There are locations and they offer services, sights, etc. There are means of transport and they run at certain times. Ultimately the decisions as to how to use those, or not, rests with conscious volition. For travel companies that remains exactly the same.
It is as if a layer has been interpolated between that reality and potential customers offering nothing and doing nothing that wasn’t available before (yet expending incredible amounts of energy while doing it). The current iterations cannot be depended upon. Future one’s do not as of yet exist. Yet we are told this is utterly world and economy changing.
One person is quoted as saying:
She reckons AI “definitely got me deeper” than more traditional Google searches.
But what is A.I. as is, other than a sort of cursory google search? And not an accurate one? One person notes that ‘when it comes to real-the information all the AI engines seem to fall flat on their ass’. Which is, if one thinks about it, incredible, given the emphasis placed on this. Again there’s this weird dynamic where people ‘use’ A.I. to do things they should do in the first place, and which they’ll have to do anyway.
Despite his misgivings, he uses AI regularly. “It does work, but you have to double-check things. If I’m going to a football match in Madrid and want to know what part of the city to stay in so I’m close to the Bernabéu, it will recommend places close the Metro line to the stadium. If you know the city well, you’d know this, but if you are not familiar with it, it can really help. And it can also learn things over time about how you like to travel, what you prioritise.
Can it? Is that additional information actually more than you’d find with a good map and guidebook?
As it happens for train travel across Europe, or the world, you don’t need A.I. You can go here, The Man in Seat Sixty-One. Schedules, mostly bang up to date, and links to train companies. Useful recommends on accommodation. The ability to plan journeys within Europe from one destination to another.
I did a bit of reverse engineering for this piece and fed in some of the parameters for the holiday into an A.I. One hotel was recommended that had been booked, but then that was directly beside a station, a no-brainer as it were (and given I’d booked months ago, not sure it’d be possible to afford it). Otherwise train times, the complexity of crossing a frontier or two where multiple train operators run, none of that was there. It’d be a start, sure, but again, no more useful than having a calendar and just going to Seat Sixty-One. It offered a list of dates that I have the paper and pen counter part – and even that wasn’t quite right.
What’s troubling is that people are waving in a supposed tech into their decision making that is not fit for purpose. As the Macalope column on Macworld noted earlier this year: Every attempt to make AI a product has failed or is having a lot of trouble getting off the runway.
Too few people in government, the private sector, education, wherever appear to understand this.