Don't Worry! AI Won't Be Reading Your Mind Anytime Soon, Experts Say

Researchers may have found a way to read images from your mind using artificial intelligence (AI). Even though the concept is far from reality, the idea raises privacy concerns. 

A neuroscientist from Japan claims to have discovered a new way that AI can decode brain activity into readable format. The team used machine learning to analyze the brain scans of subjects, showing up to 10,000 images inside an MRI machine.

"Neural interface technology is enabling recordings from a higher channel count than ever before so that we can 'see' brain activity in high resolution," George McConnell, a professor in the Department of Biomedical Engineering at Stevens Institute of Technology, who was not involved in the research, told Lifewire in an email interview. "In the case of 1000s of electrode recording sites in the brain, recording action potentials, or 'spikes,' each recording site samples extracellular voltages at more than 20,000 times per second. AI is suited for these kinds of huge datasets and has the potential to scan and classify neural activity patterns, i.e., 'thoughts,' that human eyes might miss."

Can AI Read Your Thoughts?

Using AI to decipher thoughts on brain scans has become relatively common. However, the new study from Japan is the first time an AI algorithm, called Stable Diffusion, has been used to elicit images from the brain. The researchers added training to the Stable Diffusion system, adding text descriptions of thousands of photos. 

"Reconstructing visual experiences from human brain activity offers a unique way to understand how the brain represents the world and to interpret the connection between computer vision models and our visual system," the authors wrote in their paper. 

Despite the study's promising results, experts say that computers are a long way from being able to read thoughts. Irina Raicu, director of internet ethics here at the Markkula Center for Applied Ethics, said via email that there are experiments in neuroscience that use AI tools in various ways, like the one described in this study. Still, researchers have gone to great lengths to push back against mind-reading claims. 

"All of us need to be careful so we don't contribute to the sea of misinformation swirling around AI developments, which makes our responses to the actual technology less effective," Raicu added. 

The study from Japan isn't breaking entirely new ground. Researchers in brain-machine interfaces (BMIs) have already demonstrated the feasibility of decoding' thoughts' from brain signals and using them to control things like a computer cursor, wheelchair, and a robotic arm, McConnell said. 

"It may be that miniaturization of the neural electrodes can minimize the immune response to allow for robust, long-term recordings, but this remains to be seen," he added. 

There might be good reasons to have AI reading your thoughts, Selmer Bringsjord, the director of the AI & Reasoning Lab and professor of cognitive science and computer science at Rensselaer Polytechnic Institute, said via email. He said that most people have experienced "Eureka!" moments when suddenly they figure something out, even if they haven't been actively pondering the problem in question just before the epiphany.  

"This tells us why, within 'guardrails,' we want AI reading our thoughts:  The artificial agents in question can, outside our own conscious thinking, be working to crack problems on our behalf, and then if they succeed, 'pop up' to us with the solution," Bringsjord added. "We don't want thought-reading AI to trample our mental lives, but in the manner that parallels our own unconscious problem-solving labor, they can help immeasurably."

AI Privacy Concerns

The privacy implications of AI poking into your thoughts, even if it's just deciphering images, are enormous. 

"I'm not sure I'd like to have any artificial agent get access to my mental life across the board; let's just say occasionally that life is not perfectly fit for public consumption, and whatever the AI gets hold of... well, no matter what we are closer to such consumption," Bringsjord said. "But if I can engage, disengage, surgically direct... the efforts of mind-reading AI under my employ, with verified protection, the prospect seems palatable."

Raicu said that when AI is trained on datasets including personal data, one concern is that the data might be included in specific AI-generated outputs and lead to unintended disclosures of information. 

"As everywhere in the digital realm, there is also the broader issue of a lack of consent: people whose data is used to train AI were not anticipating such a use and never consented to it," Raicu added.