Five AI myths I meet at every talk

The questions after a talk are almost always worth more than the talk.

That's where you hear what people actually believe, rather than what they think they're supposed to believe. And after enough sessions the same misunderstandings keep coming back, whatever the industry.

One thing worth saying first. Most myth-busting pieces about AI only debunk the sceptics, because they're written by someone who wants to sell AI. But at least half of the most expensive misunderstandings I meet come from the enthusiasts. Two of the five below are theirs.

"It learns how we work over time"

This is the costliest of the lot, because it makes people wait for something that isn't coming.

An ordinary AI tool remembers nothing between conversations unless you've built something that makes it do so. Every new conversation starts from zero. It doesn't know your company's name, how you write, or what you concluded last week.

I've watched teams use the same tool for six months waiting for it to start understanding their tone. It never does on its own. What actually produces that effect is dull and takes an afternoon: one document with your examples, your words and your rules, that somebody pastes in at the top every time.

If you want it to happen automatically, that can be built. But it's a decision and a piece of work, not something that ripens.

"AI just makes things up"

This one comes from the sceptics, and it isn't wrong. It's just aimed in the wrong direction.

The model predicts the next word based on what's likely. Ask it for a figure, a date, a legal reference or a source, and it will produce something that looks right, because that's what it's built to do. Sometimes it is right. That's what makes it dangerous.

But look at what you were actually going to use it for. Summarise this meeting. Make this paragraph shorter. Pull the action items out of this email thread. Give me three headlines for what I've already written.

In every one of those it's working on material you handed it. There's nothing to invent.

The rule is simpler than the rumour: let it process what you give it, not retrieve what you don't have. And anything that's going to become a public claim, you check yourself, exactly as you would with an intern's draft.

"The young people on the team will just pick this up"

I hear this one from managers, and it's a comfortable assumption, because it means the problem solves itself.

It doesn't hold. Growing up with apps makes you quick at finding the button. It says nothing about your ability to spot which part of a workflow is worth handing over, which is the hard part and depends on experience of the work itself.

What I actually see is that the people who get furthest are the ones most annoyed by one specific recurring task. That irritation has no age. Someone who has spent twenty years on the same report knows exactly which parts of it are unnecessary.

So don't count on it spreading upward from below. It spreads from whoever has a concrete problem, at either end of the age range.

"We need an AI strategy first"

Enthusiast myth number two, and it sounds responsible, which is what makes it effective.

I've seen strategy work run for months while not one person saved a single hour. A document about how AI will transform the business, written by people who haven't used it on a real task themselves. It's an expensive kind of procrastination.

Turn the order around. Let three people use it for six weeks on work they actually do. Then write the strategy. Now it's about something you know rather than something you guessed, and it comes out shorter.

The only thing that needs to exist in writing before that is what must never be entered, and who to ask when you're unsure. That's one page, not a project.

"Is this taking my job?"

That question rarely arrives directly. It comes during the break, as a joke, and it deserves an honest answer rather than an upbeat one.

Here's mine. Some tasks do disappear, and some of them are tasks people are currently paid to do. Claiming otherwise would be dishonest to someone who is genuinely worried.

But what I see at clients isn't people being replaced. It's the same people doing the same work faster and spending the recovered time on what nobody had got to. That isn't a guarantee, and it depends on what the company chooses to do with the time.

What I am sure of is that whoever understands where the tool is good and where it isn't ends up better placed than whoever decided in advance, in either direction. That holds regardless of what happens to the industry.

If you'd like your team or your network to get this live, with the question session where it usually gets interesting, I give talks and run workshops. The formats are on the services page.