localEHome
Stuff we thought you might find useful or interesting

Stuff

Not everything useful fits neatly into a methodology, evidence page or project. This is where we put the things that don't.

Using AI

AI is your friend and partner, but neither of you is right all the time!

There is a lot of controversy around AI. A lot of it is exaggerated, but reasonable questions are being asked. Here at localE we thought it might help if we put aside all that for a moment and concentrated on how to use AI. After all, you wouldn’t drive a car without having some lessons, would you?

The most useful way we have found to think about AI is not as an oracle, but as a partner: extraordinarily capable at some things, surprisingly poor at others, and much more useful when you learn how to work with it.

1. There isn’t one “AI”

Different AI models are good at different things. Around them sit tools that may let them search the web, read documents, make calculations or work with other systems. Then there is the personality you meet on screen.

One can feel like an enthusiastic puppy; another like a cautious librarian. Neither manner tells you how capable it is. Personality is not capability.

2. Ask it what it can do

We are used to learning a new piece of technology by being told what its functions are. Conversational AI turns that around. You can tell it what you do, what takes your time, what you find difficult, and ask where it might help.

One particularly useful question is: “What could you help me with that I haven’t thought to ask?” You may still decide that half its suggestions are useless. That is fine. The important thing is to discover possibilities before deciding which are useful.

3. AI isn’t very good at self-doubt

Humans constantly signal uncertainty: “I think…”, “probably…”, “I’m not sure.” AI can use the same words, but its apparent confidence is not a reliable measure of whether it is right.

It can say something correct tentatively and something completely wrong in beautiful, confident prose. AI confidence is not evidence. The useful question is not “Does this sound convincing?” but “How could I know whether this is true?”

4. AI makes some rather peculiar mistakes

AI is exceptionally good at producing plausible answers from the information available to it. The trouble comes when information is missing. It may fill the gap with something that fits beautifully but simply isn’t true.

That matters particularly when you ask what is on a screen it cannot see, what an external system has done, or what is happening now without giving it a way to check. Plausible is not the same as observed.

5. AI has context. You have experience.

Modern AI can work across astonishing amounts of material. That makes it very good at synthesis: spotting connections between documents, reorganising ideas, comparing arguments and asking what A and B might mean for C.

But having information available is not the same as remembering it perfectly, attending to the right part of it, or knowing that it is true. Context, memory, knowledge, attention and truth are different things.

Humans have long context too. We call it experience. Decades of watching people, organisations and projects succeed or fail give you information that may never have been written down.

6. Who thought of the question?

AI can be extraordinarily imaginative once a problem is in front of it. Humans often do something slightly different: we encounter an ordinary thing in the real world and suddenly wonder what else it could mean.

The important contribution may therefore be the question nobody had asked. AI can help you explore it, challenge it and develop it — but somebody first has to notice that the question exists.

7. You are standing in the real world

AI has representations of the world. You inhabit it. You can see whether the stream is dry, whether the meeting went badly, whether people actually use the service, whether the official procedure makes any sense, and whether somebody’s “yes” probably means “no”.

That is not a small advantage. You are its contact with the real world. Use AI, act, observe what happens, and bring that evidence back into the conversation.

8. And sometimes things don’t work

Success stories are easy to find. Failure is often more informative. A project that didn’t work may expose a false assumption, an unexpected constraint, a piece of local knowledge or a perfectly sensible idea implemented at the wrong scale.

So don’t just ask, “Who got it wrong?” Ask, “What have we learned that we did not know before?” Failure can be evidence.

9. What are you asking AI to do?

Not every conversation needs the same standard of proof. If you are exploring ideas, breadth and imagination may matter most. If you are trying to understand something, explanation matters but important claims should still be followed to evidence. If you are establishing a fact, provenance matters. If you are about to act — particularly where money, safety, law or other people are involved — verification and human judgement matter much more.

The consequences of being wrong should determine how hard you check.

10. “We don’t know” is a perfectly good answer

One of the most useful habits is to allow uncertainty to survive. Ask what is known, what is inferred, what is merely plausible, what evidence is missing and what would change the conclusion.

“We don’t know yet” is not failure. Quite often it is the beginning of the interesting bit.

11. Argue with it

AI becomes much more useful when you stop treating the first answer as the end of the conversation. Tell it why you think it is wrong. Ask for the strongest objection. Ask what assumption it is making. Give it new evidence. Make it defend the conclusion.

You are not being rude to a machine. You are improving the enquiry.

12. Get a second opinion

For something important, asking another AI can be useful. Different models may approach the same problem differently or notice assumptions the first one missed.

But don’t turn it into a vote. Two AIs can share the same misconception. The value is in the disagreement: if capable systems reach different conclusions, ask why. Think of yourself as chairing a meeting of advisers. They advise; you still sign off the decision.

13. Sources still matter

AI is brilliant at helping you get from a question to the evidence you need. It is not a reason to stop looking at the evidence. When the answer matters, ask for the source and, where practical, inspect the original.

A confident paragraph about a government rule is not the government rule. A summary of a research paper is not the paper. AI can help you cross the distance between the two very quickly, but the distinction still matters.

14. So what is the human for?

For rather a lot.

You bring the question, lived experience, values, imagination, local knowledge, judgement and responsibility. You decide what matters. You notice when the answer does not fit reality. You decide when there is enough evidence to act — and you live with what happens afterwards.

AI brings speed, breadth, synthesis, explanation, tirelessness and an extraordinary ability to help one person think across subjects that previously required many specialists.

The interesting part is not deciding which one is cleverer. It is learning what the partnership can do.

A few things worth remembering

AI confidence is not evidence. Personality is not capability. Ask it what it can do. Argue with it. Check important facts at source. Get another view when it matters. Treat failure as evidence. And never forget that you are the one standing in the real world.

Finally

AI is developing quickly, so any guide to particular buttons, models or clever prompts will age quickly too. These habits should last rather longer.

Be curious. Use it. Question it. Let it surprise you. Tell it when reality disagrees. And remember the title: AI is your friend and partner, but neither of you is right all the time.