Food for thought

The year is 2043. You get into your car, which scans your brain and detects that you want to go home. When you arrive at your front door, your house immediately senses what you want to eat and puts in an order at a local takeaway.

That is the future that some are predicting after a recent study found AIA computer programme that has been designed to think.  may be capable of reading our thoughts.

In the study, a group of people was presented with a series of simple images, like a teddy bear, an aeroplane, and a clock tower. An fMRIFunctional magnetic resonance imaging measures brain activity by detecting changes associated with blood flow. Blood flow is linked to the activity of neurons.  machine was used to scan their brains, and an AI programme was able to use these scans to construct an image that resembled the one they were thinking of.

Being able to read people's brains could have huge benefits. For example, thought-reading AI could detect depressive thoughts in someone who struggles to talk about their mental health.

It also offers a way of talking to people in a "vegetative state", where a patient cannot speak or move. It is estimated that around one in five people in a vegetative state still have fully functioning brains.¹ Programmes like this one would give them a way of communicating.

But others think the risks outweigh the advantages. They say we have the right to keep our thoughts private.

Still, it might be too soon to worry. Some experts think this is just another example of AI smoke and mirrorsA classic technique in magic illusions that makes something look like it is hovering..

Many magicians use a technique known as "cold reading". They usually start with general statements that could apply to most people, and then narrow their questions down based on the responses they get from their audiences.

With this technique, they can persuade people that they know far more about their lives than they actually do, and make it seem as if they are reading their minds.

The key to their success is that people want to believe in it. This is called the "Forer effect": an innate tendency to relate everything back to our own lives. And some say the AI is just doing the same thing. 

The researchers used a programme called Stable DiffusionA deep-learning, text-to-image tool released in 2022. It can create images based on text descriptions. , which is based on a vast database of images pulled from the internet. Because of this, the programme almost certainly already had access to the pictures that the subjects of this study were shown.

That means it does not really have to read our thoughts: it just has to figure out broadly what we are thinking about, and then match this to the image it already has.

Because it has many photos of teddy bears, aeroplanes and clock towers, it will not necessarily pick exactly the right one. Instead, it will create a mix of them. That means the image comes out quite different from the original, which helps to make it seem as if it has reconstructed it from our brains.

This is still an impressive feat, but it means the AI is still far from being able to read our thoughts. It would not be able to detect an original thought or image that we had generated within our own minds. Yet we believe it can actually read our minds because that is what we want to believe.

Yes: Being able to detect unhappy, negative thoughts could save people's lives. And if all our thoughts were known to others, we would have no choice to be honest with them. 

No: There are all kinds of good reasons why we might choose to hide our true thoughts. We should always have the choice as to whether or not we let other people know what we are thinking.

Or... It is impossible to know what someone is "really" thinking anyway. We do not even know all of our own thoughts. No programme will ever be capable of making our thoughts known to others.