AI in the Pinball Sphere: What It Can Genuinely Do, and What It Can't
Mention artificial intelligence in a pinball forum and you'll get two reactions within the hour. One half sees a tool that could save a hundred hours of troubleshooting. The other half sees the thin end of a wedge that ends with a screen where the playfield used to be.
Both reactions are reasonable. We'd like to work through them properly.
We'll say our bias up front: we think AI is, on balance, good news for this hobby. But we've spent enough time inside these machines to know that "on balance" is doing real work in that sentence, and the objections deserve better than being waved away.
Where AI genuinely helps
1. Diagnostics — the obvious one
Fault-finding on a forty-year-old machine is pattern recognition under uncertainty. A display goes blank. That could be the display itself, the interface, a single bad bit from the MPU, a ground fault, a connector that looks fine and isn't, or a power supply that sags only under load.
An experienced technician narrows that list quickly because they've seen the pattern before. That's precisely the kind of task machines are now good at: given these symptoms, in this system, with this history, here's the ranked list of likely causes.
This doesn't replace a technician. It replaces the bottleneck — the fact that there aren't enough experienced technicians to go around, and the ones who exist are busy.
2. Preserving knowledge that's disappearing
This one worries us more than people realise.
The engineers who designed these systems are retiring or gone. The people who serviced them commercially in the 1980s are largely out of the trade. A vast amount of what's known about Gottlieb System 80 exists as forum posts, scanned manuals, and undocumented experience in a few hundred heads worldwide.
The ground mod problem is a perfect example: a documented design flaw from 1987, still catching people out today, still mostly transmitted by word of mouth. Tools that can absorb scattered documentation and make it answerable in plain language are, quite simply, a preservation technology.
3. Lowering the barrier for newcomers
The hobby has an entry problem. A first-timer buys a project machine, opens the backbox, sees four boards and a wiring loom, and quietly gives up. That machine goes back on the market or into a barn.
Anything that helps a beginner get past the first terrifying week is a machine saved. We're not sentimental about this — the alternative to a newcomer with an AI assistant is often not a newcomer with a mentor. It's no newcomer at all.
4. Documentation and translation
Less glamorous, genuinely useful. Our own installation guides ship in French and English. Making technical documentation clear, accurate and available in more languages is work that AI does well and that nobody was doing before.
Where the scepticism is right
Now the other side, without softening it.
1. Confident wrong answers
This is the serious one. An AI system that says "check the driver board" with total confidence when the actual fault is a ground connection will cost you an afternoon and possibly a working board.
In a hobby where a wiring mistake can cook a transistor, confidently-delivered bad advice is worse than no advice. It's precisely why we build our boards with socket-mounted fuse components — because we assume mistakes will happen, including ours. Any diagnostic tool worth using has to be honest about its own uncertainty.
2. The erosion of hard-won skill
There's a real argument that if the machine tells you the answer, you never develop the intuition to find it yourself. And that intuition is the thing that makes an experienced restorer valuable.
We take this seriously. Our honest view is that it depends entirely on design: a tool that says "replace this board" teaches nothing, while a tool that says "these three symptoms together usually indicate a floating logic ground — here's why, and here's how to confirm it" teaches something every time it's used. The technology doesn't determine which of those you build. The designer does.
3. Authenticity and the slippery slope
The fear that AI is a step toward machines that aren't really pinball machines any more is not irrational. Every era has produced modifications that purists found appalling and that later became normal — and some that were genuinely appalling and stayed that way.
Our position: there is an enormous difference between helping you keep an original machine running and changing what the machine is. We're firmly on the first side of that line, and we've written a separate article about exactly where we think the line sits.
4. Dependency and longevity
A System 80 built in 1985 still runs today because it doesn't depend on anything outside itself. Any tool requiring a server, a subscription, or a company's continued existence is a tool that could stop working. That's a legitimate objection to a great deal of modern technology, and it applies here.
Our honest conclusion
The drawbacks are mostly design problems, not inherent properties. Confident wrong answers can be mitigated by building in uncertainty. Skill erosion can be mitigated by explaining rather than instructing. Dependency can be mitigated by not requiring a server for basic function.
The benefits, meanwhile, address problems the hobby genuinely has: not enough expertise to go around, knowledge disappearing with the people who hold it, and a barrier to entry that turns away people who would have loved this.
So: cautiously, conditionally, and with our eyes open — yes, we think this is good for pinball. The condition is that it's built by people who actually care about the machines rather than people who see a market.
We intend to be the first kind. Hold us to it.
