What should AI not do?

We spend every day digitising board work. Right now a lot of that time goes into artificial intelligence. We see a technology that can free people from administration and make large volumes of information comprehensible. That puts us in a better position to do our work.

We’re optimists.

But the better AI gets, the more another question pushes its way in. Just because something can be automated, does that mean it should be?

For a long time the tech industry has had a goal it has never really questioned: remove friction.

Fewer clicks. Shorter processes. Automated flows. Information before you’ve even started looking for it.

Often that’s wonderful.

Nobody learns anything from spending twenty minutes hunting for an old set of minutes. There’s no value in summarising eighty pages by hand, or typing the same information into three different systems. If AI can do that work in seconds, we should let it.

But not all friction is bad.

Having to think is friction.

Forming your own position is friction.

Listening to someone you disagree with is friction.

Weighing two poor options against each other is friction.

Making a hard decision and then standing behind it is definitely friction.

And perhaps, in our eagerness to make life easier, we have started treating that kind of friction as a technical problem to be solved.

From assistant to decision-maker

The difference becomes clear in our own field: boards and management teams.

Imagine a board about to make an important decision.

An AI can read hundreds of pages of material and summarise them. It can find what the board has said about the question before. It can point members to information they may have missed, and help draft the minutes afterwards.

All of this can give the board more time for what matters.

But the same AI can go one step further.

It can weigh the options against each other, analyse the organisation’s history and financial data, and finish like this:

”Based on the available information, I recommend option B.”

On the surface that’s just one more efficiency gain.

But something fundamental has changed.

The technology is no longer helping the person make the decision. It has started making the decision for them.

And the recommendation comes from a system that analysed more information in seconds than anyone could read. That makes it hard to argue back.

”The data shows.”

”The AI recommends.”

There is a risk here that our industry needs to talk about far more.

Not because AI is evil or conscious, or has an agenda of its own. But because handing over responsibility is very convenient for us.

AI doesn’t have to make us dumber

The debate about AI tends to collapse into two extremes.

On one side is the idea of AI as the solution to almost everything. On the other is the fear of a future where machines replace people.

We think a far more interesting question sits between them.

What happens to us when the machine gets good enough?

Satellite navigation didn’t make us unable to travel, but many of us have become much worse at finding our way.

Search engines gave us access to nearly all the world’s information, and changed what we need to remember.

Social media let us keep in touch with more people than ever, while we’re still working out what it’s doing to our attention.

Technology doesn’t only change what we can do. It changes us too.

With AI this is no longer only about memory, a sense of direction or attention. For the first time we have a tool that can genuinely take over parts of what we call thinking.

That makes the question bigger.

Responsible AI is also about what we choose not to build

When the tech industry talks about responsible AI, the conversation is usually about security, transparency, privacy, bias and accurate data.

All of it matters.

But we think one more principle needs more room.

AI should strengthen our ability to think and act, not give us a way out of doing either.

That means AI is welcome to summarise the material. But a person has to weigh it.

AI is welcome to find the arguments. But people have to discuss them.

AI can show what was decided before. But it shouldn’t determine what’s right today.

AI can draft a proposal. But someone still has to be able to say: this is what I think. This is the decision I stand behind.

Because responsibility cannot be automated.

Or at least it should not be.

We need to decide where the machine stops

For those of us who build digital products, this isn’t a theoretical discussion.

We make decisions like this every time we develop a new feature.

And we won’t always know where to draw the line.

When does a summary become an interpretation? When does a proposal become a recommendation? When does AI help someone think better? And when does someone start trusting the system so much that their own judgement weakens?

There are no easy answers.

But that may be exactly why we need to start asking the questions now.

AI will get better. It will be able to do more of what we think of today as skilled human work. So it’s no longer enough to ask what the technology can handle.

We also need to ask what we want to use it for.

Because the goal cannot be maximum automation.

The goal has to be better people and better organisations.