Chilling AI creates pictures by reading people’s MINDS – and it could be used to record your dreams

THESE weird fuzzy images look like paintings – but they actually have a much more sinister story behind them.

With a smudged teddy bear, a peaceful forest scene and a strange cloud formation, there is something uncanny about the images.

bioRxiv

The strange images have an uncanny quality[/caption]

bioRxiv

They could have be pieces of art hanging in a gallery[/caption]

bioRxiv

But not all is as it seems with the images – such as this cubic clock tower[/caption]

bioRxiv

This strange could scape was scanned from a human’s brain[/caption]

And that is because the pictures have not been drawn or painted by humans.

But images have actually been ripped from a person’s mind by an AI-driven brain scanner.

The machine studies the brain patterns in your head and then creates an image based on what you are thinking.

So yes, it means that scientists are now one step closer to developing technology that can read your mind.

And they boast it is 80 per cent accurate.

Researchers at Osaka University used the “Stable Diffusion” AI tool – which use text images to create inputs.

But instead of using words, this team the scientists used brain scans from fMRI (functional magnetic resonance imaging).

The fMRI could pick up blood flow changes in brain areas linked to image perception.

And from this the AI managed to construct a roughly accurate image of the picture the subject is thinking of.

Four participants were used in the study – each viewing a set of 10,000 images.

The AI at first developed images which looked like TV static, but as the process went on – it learned how to work out the images.

Scientists found the AI was getting an approximately 80 per cent match on the images.

As seen in the pictures above, the AI has most of the basics of the image and the composition correct.

But there are little inaccuracies – such as the colour of the teddy bear’s tie.

And at times it gets confused, as it turned the shape of a plane into a plane shaped cloud.

“We show that our method can reconstruct high-resolution images with high semantic fidelity from human brain activity,”the team said, in the study published in bioRxiv.

“Unlike previous studies of image reconstruction, our method does not require training or fine-tuning of complex deep-learning models.”