This month I decided to fill a few gaps in my record collection. I already own nearly complete sets of singles by The Jam, Dexys, The Undertones, The Who, Small Faces, The Kinks, and a few others. Just a few dozen records, each costing £2–£3, would be enough to complete the sets. The most expensive was £10.
The easiest way to pick up a haul like this is through Discogs—basically eBay for vinyl collectors. The site is full of multiple copies of most singles, from different vendors, sortable by condition, sleeve, and pressing. Prices vary hugely: from mint first pressings in picture sleeves to worn reissues in generic sleeves.
For the curious, these records are usually graded using the Goldmine scale.
But here’s the problem: while each single costs only a few pounds, postage and packing (P&P) also costs a few pounds. Buying from the cheapest vendor for each record would almost double the total bill.
What I really needed was a way to determine the best set of records, at the best price, using the fewest possible sellers.
So, I set this as a task for ChatGPT. I fed it a list of the records, with parameters for each: condition, whether it needed a picture sleeve, which label I wanted. Then I left it to work.
Days later it had nothing useful to show. Endless scanning of the Discogs database only left it more confused. Eventually I completed the exercise using a mix of AI, spreadsheets, and my own judgement.
It worked—although I now own three copies of Ready Steady Who, only two of which have picture sleeves.
This gets us to bounded rationality. I’ve written about this before: essentially, human beings aren’t perfectly rational in the way classical economics suggests. We don’t always make optimal decisions.
That’s not necessarily a bad thing. Humans are good at “satisficing”: finding a solution that’s good enough, with some compromises, rather than spending forever chasing perfection.
Most of the time in business, getting a really good answer right now is better than finding the perfect solution six months or a year later. The limits of rationality are often an advantage, not a weakness.
This is where AI struggles. It doesn’t know when “good enough” beats “perfect but incomprehensible.” It can tie itself in knots trying to calculate an ultimate solution when what’s needed is clarity and speed.
If this reminds you of The Hitchhiker’s Guide to the Galaxy you’re not wrong. Douglas Adams nailed the point: the “Answer to the Ultimate Question of Life, the Universe, and Everything,” computed over 7.5 million years, only made sense if you asked the right question in the first place.
AI is the same. Its answers only matter if we frame the problem sensibly and can understand its thought processes.
Or maybe completing a perfect record collection really is as impossible as finding the meaning of life.
‘AI is the same. Its answers only matter if we frame the problem sensibly and can understand its thought processes.’
When I was a student, my programming lecturer referred to this as GIGO (Garbage In Garbage Out).
If you wanted some software making, AI could in a lot cases handle the code, but don’t expect it to look anything like you want it to look like. Software engineers when getting the spec from a client will ask the right questions and often the right questions are not about whats been said, and more about what hasn’t been said.
For example you could ask an AI to write a simple program that allows you to enter data and export the results. AI does this and exports the data to a text file, it’s unorganised, looks crap and is totally unsuitable for this important presentation you need to attend today. A software engineer will ask “How do you want the results? A nicely formatted PDF, a word document, a spreadsheet? Maybe multiple options, a spreadsheet for your own records and smartly laid out PDF for the meetings.”
The example is very basic, but AI does not have the social intelligence to ask the right questions, to fill in the gaps, to see things in an holistic way.
Of course in your record example, you could feed in more data on your preferences to get better results. However, the AI is never going to be able to make that judgement call the way you, or even a partner or friend could. Its never going to spot a very rare record that maybe is on sale at £30, but is actually worth £100, unless you specifically program it to recognize and act on those specific opportunities. At least not yet anyway.
Just a quick thought after posting the above…
In order to program AI to make impulse purchases, we must first examine our own mind, our own subconscious as to why we make these purchases in the first place, things like emotional state, irrationalities, a desire to have rare things. Then we would feed them into the AI.
I’m not sure I like the idea of an AI having that much data on me, especially one that is not 100% controlled by me. Imagine a company selling this data to advertisers, or if a bad agent was to get hold of it.
We would be building an AI in the model of ourselves, and I see more negatives than positives.
It’s not just recording buying, in business and in life the ability to get to a quick solution that works and is good enough beats waiting for the perfect solution most of the time. Turns out that is an important skill that AI struggles with.
I use AI a lot, and some things it is great at, others it struggles with, and I am not sure if that will ever change
I do worry that a lot of low level jobs are lost to AI which means that there are fewer experienced people higher up in a few years. This will wipe out many of the productivity gains that AI brings
The low level jobs that AI can already do are in danger, paper pushing, data input, customer service jobs like live chat and email. Thats a lot of jobs. As AI improves it will consume more jobs. AI can do much of the analyse type of work managers do, but can’t motivate teams, show human empathy or resolve conflict – but does it need to, if all the lower jobs are gone?
The timeline is disputed, but when Artificial General Intelligence arrives, the gloves are off. In theory AGI can do all management jobs, and do them better than humans. As companies ‘right size’, ‘downsize’, ‘maintain competitiveness’, and become ‘leaner’, they either outright make staff redundant or do not backfill positions from leaving workers, profit being the main motivator. This has been going on for a long time, as long as technology has been able streamline processes and make things quicker. AGI will accelerate this.
That said, I don’t think we’ll see mass redundancies. It will carry on as it has been for the past few decades, more of a wither than an execution.
Of course, I’m speaking in a worst-case scenario. Some believe AGI may never arrive, while others, like Google, suggest 5–10 years. Either way, I would like to think measures will be put in place to prevent mass job losses. In a consumer-driven world, businesses need people with money to spend – and no jobs means no customers.