I Taught An AI to innovate.

And it can teach you too.

Not long ago I was sure that an AI could never offer a creative, innovative solution to a complex problem that contains many contradictory demands. I was so sure that I even made content about it.

Despite being so sure, I was completely wrong.

But I’m wrong all the time, so what’s the harm? This is an occupational hazard when you innovate for a living. So, what changed?

Perhaps AI itself changed, but my error was more basic. The AI couldn’t innovate because I had never taught it to innovate. So, I attempted to do precisely that.

However, it is often said that to build a strong foundation to a skill one must first learn one then do one and finally teach one. Could I offer an AI the ability to innovate by teaching it how to offer this to you? Could the AI be taught how to teach? Could it give a presentation? A lesson? An interactive tutorial?

When I was growing up, there was a postcard stuck to the door of the cupboard in which my mother stored her archery equipment. The postcard had some advice upon it, which has stuck with me. Aim high, and then whatever you hit, claim that was the intended target.

So, I not only attempted to teach an AI how to innovate, I have attempted to teach it to teach you.

Aim high.

But I needed a problem for you and the AI to solve together as you learned. The Amazon Prime series The Expanse offers just such a problem. I won’t spoil it for you here, as the dawning realisation that a problem exists at all is all part of the innovation process.

This is a feature of innovation that is often overlooked in engineering education. The ability to not only solve a problem, but to detect a problem exists at all. I have written about this previously.

---

Engineers are not taught to innovate.

---

To rob you of this moment of inspiration would be to deprive you of one of the best moments to be found in innovation — the detection of the quarry to be pursued. That moment when you realise that the game is afoot. That the chase is on.

Your tutor will be Joe Miller himself. The erstwhile detective of Ceres station, out on the belt. He will break this tutorial into two lessons.

The first lesson is all about Problem Discovery. You can’t innovate if you don’t know why you are innovating. Most innovation exercises jump straight to efforts to find a solution. But if you don’t properly understand the nature of the problem, your solution will be junk. No-one wants to stand before a room full of sceptical stakeholders and present junk.

Go into a room like that too fast, and the room eats you.

I keep warning you. Doors and corners, kid. That’s where they get you.

You first have to crack the case. Lay it all out before you. Understand its moving parts, its contradictions and its paradoxes. Miller will try to help you understand how this is done in the following ChatGPT-4 session.

You can find the first workshop here.

Unfortunately, you will need an OpenAI account and be paying for the Plus plan to run this effort. Sorry. I tried to find another way that does not require this, but failed. If anyone has any suggestions, I’m all ears.

In the meantime, I recorded a session. You can find that here.

To drive this mechanism to stick to a formal schedule is fairly tricky. The AI is almost too good at embarking upon tangents in the discussion. So, on occasion it can be a little flaky. However, if it does this just ask it to restart and it may hopefully recover.

This is innovation. The reliable functioning of a prototype balanced on the edge of a razor presented to a sceptical audience. I have been here many times.

The second tutorial offered by Miller starts with the assumption that we do properly understand the problem to be solved. From here, Miller will take you on a journey to discover as many innovative solutions as you can.

You can find the second workshop here.

If you are inclined to follow Miller on this journey, thank you for taking the time and I would very much appreciate any feedback that you have to offer. This can be appended to the comments of this article.

But it’s all an experiment, so where Miller will actually take you is somewhat out of my hands. Good luck.

For those of you who cannot embark upon this journey, I can offer a consolation in the problem to be solved. Its a puzzle that has followed me around for years, the details of which can be found in the following content.

---