One of the biggest problems with having people in charge of Celtic who have no grasp of strategy, no sense of due diligence, and no understanding of systems outside their own narrow knowledge space is that they are prone to mistakes.
Not just ordinary errors, but major miscalculations with long-term consequences across the club. The more complex the thing you’re trying to adapt to your environment, the greater the capacity for those mistakes.
Right now, I have a serious concern about where Celtic are heading.
Most Read on The Celtic Blog
We know that part of the plan for next year involves so-called “data-driven signings.” That phrase alone sets alarm bells ringing. It doesn’t just mean analysing numbers — which can be useful when done properly — but, I strongly suspect, it means leaning heavily on AI systems to do large parts of the scouting work.
If that’s true, then we may be in even bigger trouble than I first feared.
It’s one thing to have analysts look at data, weigh it against human scouting reports, and make balanced recommendations. It’s another thing entirely to have a computer crunch numbers and spit out a list of names. I wouldn’t be remotely surprised if Celtic have already dabbled in this — and if that’s the real reason behind some of the disastrous transfer windows we’ve suffered through.
AI is one of those subjects that has fascinated me for a long time.
People often ask if I’m worried about AI making my own job redundant, and my answer is simple: no. Not because I’m complacent, but because I know what AI can and can’t do. It’s not very good at replicating what I do. In learning how to use it — and where it breaks down — I’ve seen first-hand just how many limitations it has.
I use it in editing and for image generation. I work by dictating my articles into Word, and because Word is not great at understanding the Glasgow accent (or indeed any accent) I then wash the piece through ChatGPT to untangle it. That produces text that is close to what I was trying to say, but it comes with problems.
It often changes my meaning. It often omits key words. It often excises entire paragraphs that were central to my argument.
And sometimes, as it did yesterday, when it doesn’t understand a name or something it just substitutes it for another one, which is why my report on the game referenced Dembele (no longer at the club) and Yang who didn’t start.
I don’t like how ChatGPT writes either, and I often have to instruct it quite specifically not to attempt to rewrite my prose in any way, shape or form. It does not always listen.
I also use ChatGPT at times for image generation and that’s an area where it is meant to excel. I have to be honest; I’ve had more than a few problems with it. I’ve got fairly good at using AI tools for pictures, but only because I have a creative mind, generally know the mood I am going for and because I know how to ask for very specific things.
I know when something looks wrong, I can tweak prompts, and I can iterate until I get something usable. But even then, it makes mistakes. Every image it produces has small errors — a hand wrong here, a shadow out of place there. The “best” version of an image I wanted to create recently took me ten attempts, and still wasn’t perfect.
Now imagine that in a football context.
You can’t afford to “nearly” sign the right player after ten attempts. You can’t afford a striker who looks like he fits the profile, but whose weaknesses are invisible to the algorithm. Football isn’t about approximate results. It’s about margins.
The same issue shows up in other fields. A Harvard study on AI coding was celebrated because the machine only made errors 10% of the time. Sounds impressive, until you remember that if a human coder made errors in 10% of their work, they’d be fired immediately. When it comes to code, one mistake breaks the whole program. That isn’t “good enough.” That’s useless.
Autonomous AI is another great myth of our time.
Leave it unsupervised, and it will make catastrophic mistakes. Driverless cars are the perfect example. On paper, the task is simple: map routes with GPS, use visual recognition software to read signs and traffic lights, program rules for when to stop and start. And yet, years of testing show they make errors all the time.
Why? Because the real world is messy. Sooner or later the car encounters something outside its frame of reference — and it doesn’t know what to do. That’s when you get an accident. The system wasn’t built to handle the unexpected.
Football scouting is the same. Numbers can tell you certain things, but what happens when a player doesn’t fit the model? What happens when his “data profile” looks good, but he’s in a weak league, or has a temperament issue, or can’t adapt to the culture? A human being would spot some of those things. A computer can’t.
This is why I don’t believe we’ll ever get to “artificial general intelligence,” much less the fantasy of “superintelligence.” The whole rollout of AI into public policy and corporate decision-making will, I believe, end in embarrassment. AI isn’t magic. It’s not a brain. It still depends on human beings to feed it information, to design its code, and to shape its outputs. It will never be free of that.
And the gap between what’s theoretically possible and what we can actually do is vast. Fusion power is a good analogy here. We know it works in the Sun. The physics is sound. It’s theoretically possible to build reactors that give us limitless clean energy. But we can’t do it yet. The technical requirements are far beyond us.
The same with space elevators. The physics says you could build one: a tether running from Earth to orbit, carrying payloads cheaply into space. But we don’t have materials strong enough to make it work. In theory it’s achievable. In practice, it’s laughably out of reach.
Or take room-temperature superconductors. If we could make them stable and reliable at normal pressures, they would revolutionise energy grids, transport, and computing. Physics doesn’t forbid them. But despite tantalising lab results, we can’t yet build one that works outside experimental conditions.
AI sits in that same basket. There’s no law of physics stopping us building smarter systems. But the gap between today’s brittle, mistake-prone models and the hype of “machines that think like humans” is enormous.
Anyone who treats the hype as reality is walking blindfold into disaster.
There’s an entire industry built around AI now, and it’s fuelled by hype. That’s why it’s everywhere; it’s a technology in search of a use case, and a lot of the claims for what it can do are tenuous at best. Yet an industry is practically built on it.
How does that happen, if the tech itself isn’t sound?
Well, the best analogy is the California Gold Rush. The dream was gold. The reality was that most prospectors went broke. So where was the industry? Well, you know who did make money, and a lot of it? The people selling picks and shovels.
In AI, those “picks and shovels” are GPUs and data centres. NVIDIA sells the hardware. Microsoft and Google build the infrastructure. They’re the ones making billions. The rest is speculation and storytelling.
And here’s the thing: even in the fields where AI works best — specialist tasks in narrow domains — it’s unreliable without heavy human supervision.
The idea of using it to design something as complex and unpredictable as a football transfer strategy is worse than naïve. It’s dangerous.
Celtic’s transfer record in recent years is already patchy, to put it kindly.
We’ve had windows where millions were wasted on players who weren’t up to standard, and where obvious squad gaps went unfilled. Imagine layering AI into that system. Imagine letting algorithms, not scouts and football men, drive recruitment.
The result wouldn’t be clever Moneyball. It would be driverless-car scouting — a system that works fine until it encounters something it wasn’t programmed for. And in football, that’s every window. Different leagues, different personalities, different hidden variables. Things a computer can’t measure.
The irony is that this board, so risk-averse in some ways, may convince themselves that AI scouting reduces risk. In fact, it multiplies it. It hands control over to systems that can’t be trusted, then pretends the results are scientific. It would be a disaster — a greater disaster than hiring ten Mark Lawwell’s and telling them to go wild.
Celtic need strategy. They need vision. They need people who understand the club, the game, and the market.
What they don’t need is to outsource decision-making to machines that don’t understand football and never will. AI has its uses, but football recruitment is not one of them. A transfer system built on AI recommendations would be madness.
It wouldn’t be progress. It would be another act of self-sabotage, on a grand scale.