Mind-reading AI isn’t sci-fi anymore… and it’s just getting started

Despite its overwhelming success, the human brain peaked about two million years ago. Lucky for us, computers are helping us understand our brains better, but there may be some consequences to giving AI a skeleton key to our mind.

A team of Japanese researchers recently conducted a series of experiments in creating an end-to-end solution for training a neural network to interpret fMRI scans. Where previous work achieved similar results, the difference in the new method involves how the AI is trained.

An fMRI is a non-invasive and safe brain scan similar to a normal MRI. What differs is the fMRI merely shows changes in blood flow. The images from these scans can be interpreted by an AI system and ‘translated’ into a visual representation of what the person being scanned was thinking about.

This isn’t totally novel; we reported on the team’s previous efforts a couple months ago. What’s new is how the machine gets its training data.

In the earlier research, the group used a neural network that’d been pre-trained on regular images. The results it produced were interpretations of brain scans based on other images it’d seen.

The above images show what a human saw and then three different ways an AI interpreted fMRI scans from a person viewing that image. Each image was created by a neural network trained on image recognition using a large data set of regular images. Now it’s been trained solely on images of brain-scans.

Basically the old way was like showing someone a bunch of pictures and then asking them to interpret an inkblot as one of them. Now, the researchers are just using the inkblots and the computers have to try and guess what they represent.

The fMRI scans represent brain activity as a human subject looks at a specific image. Researchers know the input, the computer doesn’t, so humans judge the machine’s output and provide feedback.

Perhaps most amazing: this system was trained on about 6,000 images – a drop in the bucket compared to the millions some neural networks use. The scarcity of brain scans makes it a difficult process, but as you can see even a small sample data-set produces exciting results.