Don't trust the pilot
Jan. 6th, 2026 10:17 am![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
baroque_mongooseThis morning I got an e-mail from a company asking me to review it on Trustpilot. I like this company a good deal, and I would be delighted to review it anywhere else. I am not, however, touching Trustpilot with the proverbial barge pole any more; and this is why.
One day last year I got an e-mail out of the blue from Trustpilot to say that they had taken down one of my reviews (which had been up there for several months) on the grounds that it might possibly be fraudulent. I didn't recall having reviewed that company, so initially I was just puzzled. I e-mailed them back asking for a copy of the text of the review so that I could confirm whether or not I had written it, and then we could take it from there. To this I got a canned response saying that they understood that I was disappointed but they couldn't change their decision.
Well, that was nothing at all to do with what I'd said; so I tried again, this time making what I was actually asking even clearer by putting it in bold type in a separate paragraph. My initial assumption was that my e-mail had been skim-read by some overworked grunt who was probably getting paid for answering e-mails as fast as possible. I got the identical canned response.
This is where I realised I was dealing with not so much AI as artificial stupidity. So I looked to see if there were other ways of contacting them that would necessarily involve a human in the loop; well, there was snail mail. I wrote them a letter explaining the situation, and requested them to reply by e-mail. Meanwhile, there was always a chance that if I kept the e-mail ticket open someone would eventually spot what was happening and respond, so I decided to keep replying to the canned response, and while I was waiting I might as well amuse myself by training their AI. In unexpected ways.
Trustpilot therefore got a whole string of e-mails from me; every time they sent me their identical canned response, I'd reply with an e-mail that started "this isn't relevant and I need to speak to a human" (or words to that effect), followed by something pretty much entirely random. I told their AI exactly how to fit a bra. I explained to it my philosophy of maths education. I gave it a lecture about arsenic poisoning during the Victorian period. Sometimes I just gave up and fed it a couple of paragraphs of lorem-ipsum. You get the idea.
I can't remember how I eventually got through to a human except that I do recall the letter didn't work; but in the end I did. It took more than two months. The human was actually very nice, and she did, at last, send me the text of the review, which enabled me to confirm that I had indeed written it. So I said, right, then, what exactly is the problem with it? I don't see how you can consider it in any way fraudulent. I've just told you I wrote it. What more do you need?
Ah. Well. Apparently the AI had picked up some kind of patterns in the text that were often associated with fraudulent reviews, and obviously they couldn't tell me what those were because they had to be secret so that fraudsters couldn't work round them, so for the protection of the general public they couldn't reinstate my review.
By this point I was really furious with them. I told them in no uncertain terms that they knew very well it wasn't fraudulent; I'd written it myself and I'd told them so. Being humans with brains, they could override the AI and it was their duty in this case to do so. But they were too scared to do that, so I said, right, in that case I am having no more to do with you, please tell me how to delete my account.
There was a pause for about a week, and then I did actually get an e-mail telling me how to delete my account.
It was in Italian.
No, really. The entire correspondence up to this point had been in English.
So I e-mailed them back in the same language asking them how on earth (come mai) they knew I spoke Italian, and thanking them for the information; I was quite seriously tempted to write "how the heck" (come cavolo), but decided that was perhaps a little lacking in dignity. Then I went and deleted my account, in English.
And they can all, as one says in Italian, go away and get themselves blessed.
One day last year I got an e-mail out of the blue from Trustpilot to say that they had taken down one of my reviews (which had been up there for several months) on the grounds that it might possibly be fraudulent. I didn't recall having reviewed that company, so initially I was just puzzled. I e-mailed them back asking for a copy of the text of the review so that I could confirm whether or not I had written it, and then we could take it from there. To this I got a canned response saying that they understood that I was disappointed but they couldn't change their decision.
Well, that was nothing at all to do with what I'd said; so I tried again, this time making what I was actually asking even clearer by putting it in bold type in a separate paragraph. My initial assumption was that my e-mail had been skim-read by some overworked grunt who was probably getting paid for answering e-mails as fast as possible. I got the identical canned response.
This is where I realised I was dealing with not so much AI as artificial stupidity. So I looked to see if there were other ways of contacting them that would necessarily involve a human in the loop; well, there was snail mail. I wrote them a letter explaining the situation, and requested them to reply by e-mail. Meanwhile, there was always a chance that if I kept the e-mail ticket open someone would eventually spot what was happening and respond, so I decided to keep replying to the canned response, and while I was waiting I might as well amuse myself by training their AI. In unexpected ways.
Trustpilot therefore got a whole string of e-mails from me; every time they sent me their identical canned response, I'd reply with an e-mail that started "this isn't relevant and I need to speak to a human" (or words to that effect), followed by something pretty much entirely random. I told their AI exactly how to fit a bra. I explained to it my philosophy of maths education. I gave it a lecture about arsenic poisoning during the Victorian period. Sometimes I just gave up and fed it a couple of paragraphs of lorem-ipsum. You get the idea.
I can't remember how I eventually got through to a human except that I do recall the letter didn't work; but in the end I did. It took more than two months. The human was actually very nice, and she did, at last, send me the text of the review, which enabled me to confirm that I had indeed written it. So I said, right, then, what exactly is the problem with it? I don't see how you can consider it in any way fraudulent. I've just told you I wrote it. What more do you need?
Ah. Well. Apparently the AI had picked up some kind of patterns in the text that were often associated with fraudulent reviews, and obviously they couldn't tell me what those were because they had to be secret so that fraudsters couldn't work round them, so for the protection of the general public they couldn't reinstate my review.
By this point I was really furious with them. I told them in no uncertain terms that they knew very well it wasn't fraudulent; I'd written it myself and I'd told them so. Being humans with brains, they could override the AI and it was their duty in this case to do so. But they were too scared to do that, so I said, right, in that case I am having no more to do with you, please tell me how to delete my account.
There was a pause for about a week, and then I did actually get an e-mail telling me how to delete my account.
It was in Italian.
No, really. The entire correspondence up to this point had been in English.
So I e-mailed them back in the same language asking them how on earth (come mai) they knew I spoke Italian, and thanking them for the information; I was quite seriously tempted to write "how the heck" (come cavolo), but decided that was perhaps a little lacking in dignity. Then I went and deleted my account, in English.
And they can all, as one says in Italian, go away and get themselves blessed.
