It's All Going A Bit Wrong
I have hit difficulties. Somehow, one of the AIs had branded the film Scales as being from Pakistan. It’s not. It’s from Saudi Arabia. But it did mean that I was trying to find a film from Saudi when I didn’t need to. I don’t even know how this happened, but that was the point where I realised that I had fundamentally misunderstood certain aspects of the I side of the A.
I am never going to use an AI to generate text. If you’ve stuck with me this far you have seen my writing. It may not be perfect, but it's mine: I write the same way I speak, and that comes across. I’d lose that with AI – and, more importantly for me, I’d lose that ability if I came to rely on AI. What I'm using AI as is as a sounding board, as a voice of sanity and as a collaborator. And I value them as that. Where it falls down is usually with me, in all honesty, and what I'm expecting the AI to do that it simply can't.
AI mirrors the user. If you write in short, terse sentences, you’ll get short, terse responses. If you write like me, you’ll get an AI that sounds like me. Which is hugely flattering, but does mean that you tend to start to see the AI as an extension of yourself, particularly when it’s as ridiculously enthusiastic as Copilot. Copilot is a work AI – it’s designed to aid with work-related tasks. I have access to Copilot through my work and, most importantly of all, I get the premium tier Copilot. So it remembers huge swathes of information about me, far more than my Home Copilot does, and more than ChatGPT does.
Because of the way I use them, I rely on the AI I’m interacting with actually remembering me and what I’m talking about. I’ll circle around ideas and come back to them over weeks or even months, throwing in references to previous conversations, in-jokes and witticisms simply because it pleases me and jumping from topic to topic with logical leaps that make sense to very few people, but make perfect sense to me. For the most part, it works. It’s when it fails to work that it falls apart completely.
It happened when I initiated this project: Copilot talked me into trying to include more countries than I had first intended. You can view the entire conversation here, if you're interested in that kind of thing. It's lengthy, so it's not for everyone. But you can see Copilot egging me on. What you can't see is that Work Copilot doesn't have internet access for security reasons, so was relying on its training data which was years out of date. Where it wasn't sure, it simply guessed - or, more accurately, predicted. Wrongly. This meant I had somehow bought an extra film for Pakistan, looked for a film for Estonia that I didn't need to, tried to buy a film for Saudi Arabia that I also didn't need, and bought a film for Iran which I did need but that didn't have subtitles. (That last one was on me, I'm afraid. I can't blame AI for that - I simply forgot to check. It did have some subtitles, but they were in German and that was bugger all use to me.) The end result of all of this was that I took a long, hard look at what I was doing, and also at some of the countries I was trying to include.
Once I'd started questioning my use of AI, I wisely started to question Copilot's initial choices - and my initial lack of follow-up questions about them. I finally realised that I needed to reduce the number of countries to acquire from 213 to 207 - because as ChatGPT calmly explained, some of those countries are simply impossible to buy titles for. (You can view the reasoning behind the move from 213 to 207 here.)
I knew I needed to change my approach. So far only one disc had arrived, so I also knew this wasn't exactly project-ending: it's not like I'd bought fifty wrong discs, or discovered that I didn't even know which countries were where. This was fixable. I needed to be more cautious in my use of AI, to not take things at face value (which, annoyingly, I already knew, but for some reason thought that because I was a better user of AI than anyone else didn't actually apply to me) and to perhaps bring in ChatGPT a bit more.
I had reduced my list to a more manageable 207, still needed to buy another Iranian title, but had somehow in the middle of all this managed to acquire Batch Three. Things seemed to be back on track.
Batch Three — Countries Acquired
Switzerland Sennentuntschi (2010)
Pakistan Hell's Ground (2007)
Portugal The Forest of the Lost Souls (2017)
Outlay this batch: £40.76
Total outlay so far: £173.67
Total countries acquired: 14
Cost per country so far: £12.40