Last month, AI researchers at computing hardware manufacturer Nvidia developed a neural network which was able to ‘imagine’ scenes or images it had never seen before. This, naturally, led to a great deal of nervous hand-wringing by neo-Luddites worried that artificial intelligence will somehow make human beings obsolete or begin to distort our perception of reality in strange, terrifying new ways. Now, a team of Japanese computer scientists has taken things one step further into the nightmare realm by developing an AI that can literally see your thoughts. As usual, I just have to wonder: what could go wrong?
The system works by attaching to individual’s heads fMRI scanners which can detect changes in blood flow within the brain. The signals from the scanners are analyzed in real time by an incredibly advanced neural network which is then able to actually visualize and recreate what individuals are seeing. Researchers ‘trained’ the system by showing participants and AI images of the natural world so that the network could get used to what blood flow in the brain looks while subjects perceive different images. After initial training, their neural network was then able to reconstruct images it had never seen before such as letters and numbers.
According to their open-source publication, the researchers claim that this could eventually lead to technologies which could allow us to ‘see’ into human thoughts, dreams, and imagination like never before:
Reconstruction of artificial shapes was also successful, even though the reconstruction models used were trained only on natural images. The same method was applied to imagery to reveal rudimentary reconstructions of mental content. Our approach could provide a unique window into our internal world by translating brain activity into images via hierarchical visual features.
Of all the AI developments recently, this one truly sounds like the stuff of science fiction. Imagine this technology falling into the hands of law enforcement or military interrogators (if it hasn’t already). No more questioning suspects trying to get them to confess – just hook them up to this machine, crawl inside their mind, and see for yourself what they did last Wednesday evening between 8:00 and 10:00.
Think of the advertising implications too. No more relying on search terms and browsing history for suggested advertisements. If brain scanning technology could be refined to the point where attaching cortical electrodes are no longer needed and brains could be scanned wirelessly at a distance, advertisers could be able to see inside consumers’ minds and try to sell them their wildest dreams in real time.
It can be easy to read headline after headline proclaiming some new development and accept them as merely some new novelty. Self-driving cars, poker robots, AI chess masters – all of these can seem like trivial uses of artificial intelligence compared to, say, a soulless AI hivemind which seizes control of the entire internet and brings human civilization to a crashing halt. But each of these new developments inches us one step closer to suddenly finding ourselves no longer the most advanced denizens of Earth. Sure, there will be many wondrous, life-saving and world-changing uses of AI, but will they be worth the cost? If AI is going to be the end of humanity – and that’s a big if – I wonder if it’s already too late to stop it.