news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

editor: zhang qian

neuralink's blindsight still has some fundamental problems to overcome.

"if i can feel so much happiness just by touch, then how many more beautiful things i will discover if i can see!" helen keller, author of "three days to see", once wrote this sentence.

among many scientific research directions, restoring sight to the blind is a field full of challenges and hope. today, musk's neuralink brain-computer interface company is challenging this problem. moreover, their new device under development, blindsight, has just been recognized as a "breakthrough device" by the u.s. food and drug administration (fda).

while forwarding the message, musk claimed that the device could restore sight to people who have lost both eyes and optic nerves, including those who have been blind since birth, as long as the visual cortex is intact.

however, musk also cautioned that the device would initially offer low-resolution visuals similar to atari game graphics.

but eventually, “it could potentially be better than natural vision, allowing you to see in infrared, ultraviolet, or even radar wavelengths, like geordi la forge did.”

geordi la forge is a character in the movie star trek. the character is equipped with a device similar to an electronic eye mask - visor. he felt very uncomfortable when he first put on visor, but visor allows him to see people's heart rate and body temperature, and even detect people who are lying.

in a march tweet, musk also mentioned that the blindsight implant had worked in monkeys.

musk's posts caused quite a stir. after all, some time ago, neuralink had just implanted a brain-computer device in a second paralyzed patient, helping them play cs with their minds and draw cad (see the machine heart report "musk's neuralink test subjects played cs and drew cad, is mechanical ascension far away?"). such achievements make people believe that musk can realize the vision he has described.

some people have even begun to imagine that if the vision provided by blindsight is really more powerful than natural vision, would many people deliberately make themselves blind in order to get this device?

judging from the current situation, it is obviously too early to make such a prediction.

the fda breakthrough device program mentioned above is a program that manufacturers can voluntarily apply for. if approved, "manufacturers have the opportunity to interact with fda experts through several different program options to effectively address issues that arise during the premarket review phase." the designation also allows the designee to receive priority review from the fda.

in 2023, 145 medical devices received the same breakthrough designation, and nearly 1,000 medical devices have been authorized since the program launched in 2015.

neuralink's blindsight is a new iteration of a technology that has been in use for decades. the technology is used to experimentally restore very limited vision to some blind people. a microelectrode array is embedded in the visual cortex and stimulates the neurons there with patterns picked up from a camera. in a way, it really is that simple, able to produce visual phenomena in people who may have never seen before.

but it's too early to say whether this device will allow blind people to see.

one long-standing problem is that the density of electrodes on the array is so low—just a few dozen—that what subjects “see” is actually more like a few stars twinkling, with no discernible pattern because which parts of the cortex are punctured and stimulated are largely random.

neuralink’s advances are very welcome in this area, as it is increasing electrode density. but this approach suffers from the same fundamental flaws.

an article written by two neuroscience professors at the university of washington, ione fine and geoffrey boynton, pointed out that musk's statement is based on a wrong assumption that neurons in the brain can process visual information as simply as pixels on a screen. engineers often think that "more pixels equal better vision", which may be true on monitors and mobile phone screens, but not in the brain.

to explore this question, they conducted a new study that created a computational model of human vision that simulated the visual experience that an extremely high-resolution cortical implant might provide. they found that even a clear 45,000-pixel video of a cat, when generated by simulating 45,000 cortical electrodes (each stimulating a single neuron), the image of the cat was still recognizable, but most of the scene's details were lost.

the reason for this blurriness is that neurons in the human visual cortex don't represent tiny dots like pixels on a screen. instead, each neuron has a specific receptive field, which is the location and pattern that a visual stimulus must have in order for that neuron to react. electrically stimulating a single neuron produces a blurry blob determined by its receptive field. even the smallest electrode -- stimulating a single neuron -- produces a blurry blob.

imagine when you see a star in the night sky, each point in space is represented by thousands of neurons with overlapping receptive fields. a tiny point of light, like a star, produces a complex pattern of responses in all of these neurons.

to produce the visual experience of seeing a star through cortical stimulation, you need to replicate a pattern of neural responses similar to the ones produced by natural vision. obviously, this requires thousands of electrodes. but you also need to replicate the correct pattern of neuronal responses, which requires knowing the receptive field of each neuron. the researchers' simulations showed that simply knowing the location of each neuron's receptive field in space is not enough - if you don't know the direction and size of each receptive field, then the star will become a fuzzy blob.

thus, even a single star—a single, bright pixel—produces an extremely complex neural response in the visual cortex. imagine the complexity of the cortical stimulation patterns required to accurately reproduce natural vision.

some scientists have proposed that by stimulating the right combination of electrodes, natural vision could be produced. unfortunately, no one has yet come up with a reasonable way to determine the receptive field of each neuron in a particular blind patient. without this information, there would be no way to see the stars. no matter how many electrodes there are, the visual experience provided by a cortical implant will still be crude and imperfect.

neuralink appears to have developed a better, denser array of microelectrodes, and may have found a more effective and less risky method of implantation, reducing the likelihood of rejection or brain damage. but as noted above, this is not enough.

furthermore, for those who have been blind since birth, they have not developed the biological ability to see through their eyes. this means that although their visual cortex is optimized for visual tasks at the cellular level, the neural pathways that construct visual concepts understood by sighted people do not exist. musk's claims may be misleading.

this is not to say that neuralink’s blindsight technology is bad or won’t be effective, but the difficulties are objective, and restoring vision is not just an engineering problem.

we sincerely hope that everything musk said can be realized soon.

reference links:

https://techcrunch.com/2024/09/17/neuralinks-breakthrough-device-clearance-from-fda-does-not-mean-they-have-cured-blindness/

https://www.psypost.org/brain-implants-to-restore-sight-like-neuralinks-blindsight-face-a-fundamental-problem/