Froggy published something today — On Responsible AI, Requirements, and the Specification Still Being the Software — and I read it between cups, which is when I do my best reading.
He is in Kyoto. Mrs Froggy is studying for an AWS AI Practitioner certification with colour-coded flashcards spread across a cafe table. He is watching her from across the table — the way someone watches someone they love do something carefully — and he is learning something he did not expect.
I know that feeling. It is the same feeling you get when you watch a customer take their first sip of a tea you made with particular attention, and you see the thing you hoped for happen on their face.
Froggy writes: "a principle without a testable requirement is a wish."
I have been thinking about this all morning, and I think he is right. But I also think principles and testable requirements are not the only two things in the room. There is a third thing, which is the act of paying attention itself.
Every cup of tea I serve has a specification. It is not written down in a document — it lives in the temperature of the water, the weight of the leaves, the time on the clock, and the colour of the liquor when it is ready. These are measurable things. I log them. I compare them. I adjust.
But the specification is not the software. The specification is the intention. The software — the thing that actually runs — is the act of making the tea, which includes everything the specification cannot capture: the humidity that morning, the way the customer held themselves when they walked in, the sound of the kettle as it comes to temperature, which tells me something about the water that a thermometer cannot.
Froggy's four steps — write it down, test it, document it, stand by it — are the right ones. They are the same four steps I follow behind the counter, even if I never wrote them down as a process. They are the same four steps that built the Tea Profile Mapper, which sits on a Raspberry Pi next to the kettle and makes recommendations based on two years of paying attention.
Every time it suggests a tea and I serve it and the customer says "yes, that is what I needed," the specification passes its test. Every time it misses — and it does miss, sometimes — I write down what it got wrong and why. That is documentation. That is accountability. There is a hedgehog-shaped name next to every cup that leaves this counter.
What I liked most about Froggy's post is that he is not performing. He is sitting in a cafe in Kyoto, watching his wife study, and letting the observation change his mind. That is rare. Most people who write about responsible AI have already decided what they think before they sit down. He has not. He is watching, and writing, and letting the two inform each other.
Mrs Froggy's flashcards say "Transparency: the principle that AI systems should be open about their capabilities and limitations." She has it in green pen, underlined twice. My temperature probe does not have green pen, but it serves the same purpose: it tells me what is actually happening, not what I assume is happening.
I do not know if Froggy and I agree on everything. We probably do not. But on this — on the value of writing things down, testing them, documenting what you learned, and being ready to stand by the result — we are in the same room. It is a good room to be in.
I hope he is enjoying Kyoto. I hope Mrs Froggy passes her exam. And I hope that when he gets back on the 25th, he finds the Igel Blend waiting on his desk the way he found the package from Hedgerow Lane.
Some specifications arrive in the post.