For about the past decade, we’ve been able to “read minds”…well, sort of. Researchers have been using large fMRI machines to detect which parts of the brain activate when people are shown a series of images and videos. Based on these observations, they’ve been able to construct a somewhat blurry image of what a person is looking at when shown various types of images.
This is incredibly cool stuff, but the research hit a roadblock because the resolution of fMRI scans just wasn’t improving significantly.
Then, generative imagery came into play, specifically the open-source Stable Diffusion library. When these blurry brain images were processed through Stable Diffusion, the results were high-resolution images. Essentially, the AI model could dramatically improve the quality of the blurry images.
This leads to an interesting question: when does the brain signal degrade to the point where the AI can’t generate a relevant image? For years, we’ve used brain-reading electrode caps for relatively simple tasks like selecting shapes or sensing concentration.
However, we might finally be on the path to performing more complex tasks using our minds. By training the system with a large fMRI machine, and then repeating the same training with an electrode cap, we might be able to determine where the signal degrades. This could help a new model predict brain-to-computer input more accurately, even with significantly less advanced hardware.
This AI model could enhance the signal, enabling more seamless computer control directly from the mind.
But why does this matter?
This leads us to the input problem. The input problem is the challenge of finding a seamless way to give commands to a piece of technology. Desktop computers faced this issue until the invention of the graphical interface and the mouse. Smartphones had the same issue until the advent of touch screen optimized operating systems and fully touch screen displays.
Augmented and virtual reality (AR/VR) are still dealing with this problem. AR/VR products aim to be more than just gaming devices, but they currently lack an effective input method to support rapid text entry. Meta has been working hard to solve this problem with their advanced pass-through functionality that recognizes keyboards for input, but this solution requires bulkier equipment, which is not ideal for a product intended to be mobile.
The ultimate solution to the input problem has always been the brain-to-computer interface. This is what Elon Musk’s company, Neuralink, has been trying to create for years with their somewhat invasive brain chips.
However, there might be a non-invasive way to achieve the same goal using AI to enhance and more accurately interpret brain signals.
Such an AI-enhanced, non-invasive brain-to-computer interface could revolutionize the way we interact with technology. Imagine being able to draft an email, navigate a virtual environment, or even create a piece of digital art just by thinking about it. That’s the kind of seamless interaction we’re talking about when we say we’re addressing the input problem.
AR/VR technology, in particular, stands to benefit immensely from this development. These platforms are on the cusp of breaking into the mainstream but are often held back by clunky controls and unintuitive interfaces. The promise of brain-to-computer interfaces could change all that, transforming these emerging technologies into powerful tools for work, play, and everything in between.
Elon Musk’s Neuralink has certainly made waves with its brain chip, but it’s not the only player in the game. The field is open, and there’s plenty of room for innovation. The real race is on to see who can develop an interface that’s not only effective but also accessible and user-friendly.
In conclusion, the intersection of AI and neuroscience could hold the key to solving the input problem. It’s an exciting time to be in tech, and we can’t wait to see what the future holds. Stay tuned for more updates as we continue to explore this fascinating frontier.