Brain implants today: what's real and what's still speculation

The Short Version

The people actually building brain implants describe something real but narrow. Neuralink's head neurosurgeon, Dr. Matthew MacDougall, says the near-term goal is letting people with paralysis move a cursor and type by imagining movement, using electrodes that can reach only the brain's surface [1]; Stanford's Dr. E.J. Chichilnisky says retinal implants already let some blind people see blobs and streaks of light, but nothing like natural vision, because they ignore most of what science knows about the retina [2]; and Salk's Dr. Terry Sejnowski stresses how much about the brain, and even about "understanding" itself, is still unknown [3]. Joe Rogan, talking with psychiatrist Rick Strassman, goes much further, imagining implants that create a universal language, telepathy and the end of lying, while Strassman doubts it would work without changing human nature [4]. MacDougall and Chichilnisky do share some long-range hopes, such as thought-based communication and augmented vision, but they frame them as engineering problems for decades ahead, not products [1][2]. The honest line runs between decoding movement intentions and crude vision today, and everything else later, if at all.

Table of contents

What They Said

Matthew MacDougall: a blank canvas for hopes and fears

MacDougall, who trained at UC San Diego and Stanford and leads neurosurgery at Neuralink, opens by lowering the temperature. He says Neuralink, like Tesla and SpaceX, becomes a blank canvas for people's hopes and fears, and both extremes are mostly wrong [1]. What the company is actually building is a neural implant plus a robot that inserts electrode threads thinner than a human hair into a small region of the brain [1].

The first target is narrow. The initial patients are people with severe spinal cord injuries who are quadriplegic: their motor cortex is intact but disconnected from their limbs [1]. The implant reads their movement intentions so they can control a computer mouse and keyboard, regaining what MacDougall calls digital freedom [1]. At this stage, he stresses, they will not move their own limbs; imagining a movement moves a cursor, opens a word processor and types on an on-screen keyboard [1]. The company's broad near-term mission is to reduce human suffering, with using AI as a tool embedded in the brain a longer-term hope [1].

Why a robot does the placement. The electrodes are so tiny and the surface blood vessels so dense that no human hand is steady or fast enough to place them accurately at the right depth [1]. A neurosurgeon still opens the skin and skull; the robot handles only electrode placement [1].

Only the surface, for now. Current technology can reach only the brain's surface, MacDougall says, so deep functions like mood, appetite, addiction, pain and sleep will come later [1]. Huberman notes that deep regions handle more stereotyped housekeeping functions shared across species, such as temperature, blood pressure, sex drive and hunger, and quotes Elon Musk's description of the human brain as "a monkey brain with a super computer placed on the outside." [1]

Most of the work is software. Much of the difficulty is moved from the patient's brain into software that decodes motor intentions [1]. Neuralink has monkeys playing video games for smoothie rewards and, MacDougall says, holds the world record for the bit rate of information decoded from a monkey's brain to control a cursor [1]. He credits the late Stanford researcher Krishna Shenoy, a Neuralink advisor, whose work was foundational, and notes that a Stanford group including Shenoy and Jaimie Henderson decoded speech from the hand area of motor cortex, showing each brain region carries many useful signals [1]. He and Huberman credit a century of academic research, including Apostolos Georgopoulos's early recordings from motor cortex, describing basic scientists who worked without stock options for papers read by a few dozen people [1].

The AI in the loop is adaptive, not magical. At first, firing patterns mean nothing to the chip; by correlating them with what the person is asked to do in games (move the cursor up and right, down and left), the software builds a model the person then learns to use, even as signals drift from day to day [1]. MacDougall compares it to learning a piano whose keys shift over time [1]. Human participants will speed progress, he says, because unlike monkeys they can tell researchers what they're trying and adjust [1].

Would he get one? Absolutely, he says, but it wouldn't be useful yet, because early devices do things he can already do with his hands [1]. Once a device lets someone type faster than speech-to-text or their fingers, it would make sense for someone like him [1]. From hundreds of surgeries he has performed with the robot, he vouches that the procedure is already safer than many standard FDA-approved operations [1].

Next steps and a surprising limit. The longer-term goal is to reconnect brain and body by pairing the brain implant with a spinal cord implant below the injury, and Neuralink has done considerable work on spinal cord stimulation toward that [1]. But on boosting adult neuroplasticity, MacDougall gives an answer that cuts against the hype: he thinks drugs, including the psychedelics Huberman mentions, will be more fruitful than implants [1]. Changing plasticity means altering something like a trillion synapses across the whole brain, which a drug can reach and an electrode cannot, and even steerable electrodes will never, in his view, cover the brain the way a drug does within our lifetimes [1]. Neuralink isn't currently combining drugs with stimulation, he adds, because decoding the brain electrically is hard enough [1].

The long horizon. Asked about his hopes, MacDougall gives two timeframes. Over 10 to 20 years, he wants to give people control over the ways the brain goes wrong, such as addiction, depression, suicide and obesity [1]. For his children's lifetimes, he imagines full expansion of human cognition into AI, instant access to information, and communication freed from what he calls the narrow straw of speech, perhaps even opt-in linked minds [1]. Brain-to-brain messaging, he says, is mostly an engineering problem, since two people texting already communicate silently; what matters is bit rate [1]. He contrasts implants with peripheral devices like Neosensory, a wrist device that translated sound into touch for a deaf woman: peripheral devices have lower barriers to adoption but low ceilings on bit rate, while implants could potentially interface with millions of neurons [1]. At the time of the conversation, human trials were still being finalized with the FDA, and he mentions a patient registry for people with quadriplegia [1].

On animals. MacDougall says the FDA requires proof of safety in animals first, and that Neuralink monkeys are never water- or food-deprived; they have full access to food and water and play games only for extra treats [1]. Pigs are used mainly as a safety platform because their skulls resemble humans' in size and shape, while monkeys provide functional data [1].

E.J. Chichilnisky: retinal implants work, crudely, and ignore the science

Chichilnisky, a Stanford professor of neurosurgery, ophthalmology and neuroscience, offers the most detailed account of what an existing class of brain implants does and why it falls short.

What works today. Much blindness in the Western world, including macular degeneration and retinitis pigmentosa, comes from the death of photoreceptors [2]. The long-standing idea, which he stresses is not his invention, is to bypass the damaged layers with a camera, process the image as the retina would, and electrically stimulate the retinal ganglion cells that send signals to the brain [2]. Pioneering engineers have made this work: people blind for decades can see blobs and streaks of light, enough to find a bright doorway or avoid a coffee table [2]. That's the glass half full [2].

What doesn't. The glass half empty is that this is nothing like natural vision [2]. The retina has about 20 types of ganglion cells, each covering the whole scene and extracting different features, such as fine detail, motion and color. "You can sort of think of them as Photoshop filters," he says [2]. Existing devices treat the retina like a camera's grid of pixels and stimulate whatever cells happen to be near each electrode [2]. He compares the retina's output to an orchestra playing a symphony, each cell type with its own score, and says current implants are like a conductor who has scattered the sheet music so everyone plays whatever [2]. Asked how decades of retinal knowledge show up in vision-restoring devices, his answer is blunt: "Well, currently it's not." [2] Essentially none of what's been learned since the National Eye Institute was founded in 1968 has been built into existing implants, he says [2].

How much is still unknown. Chichilnisky calls the retina arguably the best-understood part of the brain, but he qualifies what that means. About seven ganglion cell types are well understood and make up perhaps 70% of the neurons sending visual information to the brain [2]. A senior researcher in his lab, Alexandra Kling, recently found about 15 more types in their recordings with "crazy properties," some responding to three or four separate blobs of light, others with spidery response profiles; their function remains unknown [2]. He also concedes that his lab's standard stimulus, random flickering "TV snow," doesn't capture everything cells do in natural vision, and that work with naturalistic stimuli is still early [2].

The smart implant he wants to build. His goal for the coming decade is a device that respects the circuitry [2]. It would record, using a chip roughly two millimeters in size to identify which cells and types are present in that person; calibrate, by stimulating and recording to learn how each electrode activates each cell; and then stimulate, using knowledge of the neural code to drive each cell in the right sequence [2]. As he puts it in the transcript, "You have to build a device that you can implant in a human that can recognize the distinct cell types," stimulate them separately, and "conduct this orchestra to create a musical score that reasonably closely resembles the natural one." [2] He calls it "a technological feat that's really challenging" [2]. AI helps, but he stresses it is an engineering tool that executes what the retina does, not a source of understanding [2]. His metaphor: an American yelling in English on a street in China might eventually get directions, but it's far better to speak the language. "Make our instruments smart," he says, rather than expecting the brain to wrap itself around a dumb device [2]. He and Stanford engineers have spent about 15 years on the basic science and are now working to build a device for living humans [2].

Where augmentation comes in. Chichilnisky is open about the path beyond restoration, and careful about it. He offers a thought experiment: vision and hearing use separate systems, which is why hands-free calls are safer than texting while driving; if two retinal pathways (the "midget" and "parasol cells) carry independent streams, a device might one day send text to one and moving traffic to the other [2]. He says explicitly this is not a research goal, only an illustration [2]. He also argues there's no bright line between restoring and augmenting: even the crudest vision device must filter out infrared light, which many cameras detect, or users would gain infrared vision [2]. It's not sci-fi anymore. It's sci," he says, adding that because humanity will clearly build such technology, scientists have a responsibility to do it thoughtfully [2]. On whether an adult brain could handle, say, twice the visual detail, he draws on Eric Knudsen's finding that adult animals show plasticity when sensory changes are introduced gradually, and expects any new signal would need to be dialed up slowly [2].

He also points to progress beyond the eye: devices that read signals from motor or language cortex so paralyzed people can move cursors or communicate, citing Shenoy and Henderson at Stanford, Eddie Chang, and Neuralink building on that research, as well as spinal cord stimulation for rhythmic movement [2].

Why specificity matters. Huberman frames a spectrum of brain interventions: the "crudest would be drugs," which act on receptors throughout the brain and bring side effects; electroconvulsive therapy is even less specific; cell-type-specific stimulation is the other extreme [2]. Chichilnisky likens shock therapy to rebooting a malfunctioning computer: it might work, but you'd rather diagnose and fix the specific problem [2]. Huberman contrasts this grounded approach with hype about chips for super memory; the hippocampus, he says, has decent knowledge of cell types but nothing close to understanding what each does, so stimulating it would produce unpredictable effects [2].

Terry Sejnowski: how much we don't know

Sejnowski, a computational neuroscientist at the Salk Institute, isn't building implants, but his conversation shows both the tools and the gaps behind them. Some of his group's key discoveries came from epilepsy patients at Massachusetts General Hospital who have electrodes implanted for about two weeks to locate seizure origins; recordings revealed circular traveling waves across the cortex during non-REM sleep, sleep spindles that help consolidate memories [3]. That's a reminder that much of what's known about the living human brain comes from electrodes placed for medical reasons.

He describes the field as having moved from bottom-up parts lists and top-down behavior studies toward a middle level of algorithms, recipes the brain follows [3]. His flagship example is simple: the basal ganglia's learning rule, "It's simply to predict the next reward you're going to get." [3] But he also stresses the limits. New optical recording methods show that even simple tasks engage many brain areas at once; in mice, the visual cortex gets more input from the motor system than from the eyes [3]. In the transcript he says that neuroscientists "have studied each one of these parts of the brain independently. And now we have to start thinking about putting the pieces of the puzzle together" [3].

On the AI side of brain-computer interfaces, Sejnowski is neither doom-laden nor naive. He calls free will, consciousness, intelligence and understanding "weasel words because you can't pin them down," and says the debate over whether language models understand language reveals that "we don't understand what understanding is." [3] He compares AI today to the Wright brothers' first flight: off the ground, but with control still the hard problem [3]. And he sees AI as complementing human experts rather than replacing them, citing a dermatology study in which doctors and an AI each classified skin lesions with about 90 percent accuracy, but doctors using the AI reached 98 percent: "It's not going to be adversarial. It's going to be a partnership." [3] His own frontier project, on how the brain does self-attention, is explicitly a hypothesis about the basal ganglia that he says "we'll see" about [3].

Joe Rogan and Rick Strassman: the speculative end

In his conversation with psychiatrist Rick Strassman, Rogan lays out the most ambitious vision in these sources. He notes that Neuralink's first patient already plays video games with eye-driven aim, and that Elon Musk has said the planned Blindsight device could let people see in infrared, ultraviolet or radar wavelengths [4]. From there he imagines brain interfaces producing a universal language and telepathic communication that makes deception impossible [4]. In his words: "So if you have real communication with people then you have universal language and then here's the big one the ability to detect deception." [4] He pictures people "experiencing each other's consciousness in a way that eliminates all possibility of deception," which he believes would force humanity to communicate at a higher level and could eventually upend money and capitalism in a practical, not Marxist, way [4].

Strassman, the more measured of the two, is skeptical. He asks whether a universal language would produce something good for everyone or whether "would we just you know build a tower of you know Babel all over again" [4]. He thinks white lies serve peace, that such a change would require a change in human nature, and that it amounts to a new species [4]. He recommends Olaf Stapledon's Last and First Men, which imagines 19 successive human species over two billion years, the last communicating telepathically only in rare global events [4]. Strassman offers speculation of his own, too, suggesting the rapture could have a biological basis, a worldwide switching-on of DMT synthesis [4].

Where They Agree, and Where They Don't

What's real. The builders agree on a short list. Implants can decode movement intentions from motor cortex well enough to move a cursor and type [1][2], and speech signals have been decoded from motor cortex by Stanford researchers [1]. Retinal implants can restore crude light perception to some blind people [2]. Both MacDougall and Chichilnisky credit decades of academic work, including Shenoy and Henderson's, for this progress [1][2]. Rogan's example of a patient playing video games is consistent with what MacDougall describes as the first goal [1][4].

What's missing. They also agree on the gaps. MacDougall says current devices reach only the surface, so deep functions like mood, pain and addiction are later [1]. Chichilnisky says existing retinal implants ignore most of what's known and that 15 of roughly 20 retinal cell types remain mysterious [2]. Sejnowski says brain areas have been studied separately and must now be put together, and that we don't even understand understanding [3]. Huberman argues that stimulating memory regions would be unpredictable because we don't know what each cell type does [2].

Two philosophies of the machine-brain handoff. There's a real difference in emphasis between the surgeon and the retina scientist. MacDougall describes adaptive software and a brain that learn each other, with the patient learning to use a model built from their own firing patterns, like a piano whose keys shift [1]. Chichilnisky argues for the opposite burden: "Make our instruments smart," so the device speaks the tissue's language instead of expecting the brain to wrap itself around a dumb device [2]. These aren't flatly contradictory, since the motor system and the retina pose different problems (reading intentions versus writing signals into a sensory system), but they reflect different bets about where the intelligence should live.

Implants versus drugs. MacDougall, of all people, says drugs rather than electrodes are the better tool for boosting brain-wide plasticity, because a drug reaches a trillion synapses and an electrode can't [1]. Chichilnisky and Huberman rank drugs as the crudest intervention precisely because they act everywhere [2]. Both positions are consistent: drugs are broad, implants are precise, and which is better depends on whether the goal is a brain-wide change or a specific signal.

Timelines and telepathy. The sharpest split is over how far this goes. MacDougall shares some of Rogan's vision, imagining thought-based communication and even opt-in linked minds in his children's lifetimes and calling brain-to-brain messaging mostly an engineering problem [1]. But he places deep-brain treatments 10 to 20 years out and the cognitive expansion further still [1]. Rogan goes beyond that to implants that eliminate deception and reshape economics [4]. Strassman thinks that would require changing human nature, not just hardware [4]. Chichilnisky's augmentation examples are offered as thought experiments, not research goals [2]. Rogan's Blindsight example of infrared and radar vision is attributed to Musk [4]; Chichilnisky's point that a retinal device would give infrared vision unless it's filtered out suggests some augmentation could be a side effect of restoration rather than a leap [2].

Why It Works

The sources explain the mechanisms behind what works, and why the rest is hard.

Motor cortex keeps working after spinal injury. In quadriplegia, the motor cortex is intact but disconnected from the limbs [1], so the intention signals are still there to be read. Research going back to Georgopoulos showed those signals can be recorded [1], and Shenoy's work showed they can be decoded fast enough to control a cursor [1].

Decoding is a learning loop. The software correlates firing patterns with instructed movements, builds a model, and the person learns to use it, adjusting as signals drift [1]. Huberman likens it to biofeedback [1]. That explains why progress depends on software and why human participants, who can describe what they're trying, speed it up [1].

Precise placement requires robots. Hair-thin electrodes among dense surface blood vessels need speed and steadiness no hand can match [1].

Retinal implants bypass dead photoreceptors. Because blindness in macular degeneration and retinitis pigmentosa often comes from photoreceptor death while the ganglion cells survive, stimulating those cells can still send signals to the brain [2]. The crude results follow from stimulating cells without regard to type: 20 different "filters" are driven indiscriminately, so the brain gets a scrambled message [2]. The smart-implant plan of record, calibrate, stimulate is designed to fix exactly that [2].

Why the retina is the test bed. It's physically accessible at the surface, and "when you're looking at somebody's eyes, you are looking at two pieces of their brain," as Huberman says [2]. Chichilnisky's lab can keep donated human retinas alive and record from 512 electrodes at once, recording the exact signals ganglion cells would have sent to the brain [2]. That level of understanding doesn't exist for most other brain regions [2].

Why deep and brain-wide changes are harder. Electrodes can currently reach only the surface [1]; plasticity involves roughly a trillion synapses across the brain [1]; and adult brains appear to need gradual, dialed-up changes to learn a new signal [2]. Sejnowski's observation that even simple tasks engage many areas at once [3] suggests why a small patch of electrodes can't simply rewrite complex functions.

Why bit rate is the key metric. MacDougall frames the future of brain communication in terms of how much useful information flows in and out [1]. Peripheral devices are easier to adopt but capped; implants could interface with millions of neurons [1]. That's the technical basis for his long-range optimism, and the same metric shows how far today's cursor control is from telepathy.

Where It Applies, and Where It Doesn't

What's described as working applies to specific groups: people with quadriplegia whose motor cortex is intact, in the cursor-and-keyboard sense [1], and some people blinded by photoreceptor loss, who can regain crude light perception [2]. For healthy people, MacDougall himself says an implant wouldn't yet be useful, since it does what hands already do [1].

It does not yet apply to moving one's own limbs (that requires a spinal implant still in development) [1], to mood, appetite, addiction, pain or sleep (deep regions the technology can't reach yet) [1], to natural-quality restored vision [2], or to memory enhancement, which Huberman calls hype given how little is known about hippocampal cell types [2]. Brain-wide plasticity, per MacDougall, is better targeted with drugs [1].

The telepathy, universal-language and end-of-lying scenarios are, in these sources, speculation: Rogan's vision [4], challenged by Strassman [4], and related to long-range hopes MacDougall places in his children's lifetimes [1]. The infrared, ultraviolet and radar vision of Blindsight is something Musk has said, per Rogan, not a result described by any of the scientists here [4]. And at the time MacDougall spoke, human trials were still being finalized with the FDA, so the human results he anticipated were forthcoming [1].

One more caution cuts across all of it: Sejnowski's reminder that neuroscience doesn't know what understanding is, and that AI is at its Wright-brothers stage, with control still the hard problem [3]. Claims that implants will soon merge minds with AI rest on two unfinished sciences at once.

For anyone with paralysis or vision loss considering a trial, the right move is to talk with a neurologist, ophthalmologist or trial team about what a specific device can realistically do.

What To Do With It

1. Sort claims by what the device reads or writes. Decoding movement intentions from motor cortex into a cursor is real [1]; writing crude light into the retina is real [2]; anything involving deep brain functions, memory or thought-sharing is future work [1][2].

2. Ask surface or deep? MacDougall says current electrodes reach only the brain's surface, so a claim about treating mood, addiction or pain with today's implants should raise a flag [1].

3. Check for cell-type specificity. Chichilnisky's test for a serious neural device is whether it can recognize distinct cell types and stimulate them separately; devices that treat tissue like a pixel grid produce crude results [2].

4. Look for bit rate, not buzzwords. MacDougall's metric for any brain interface is how much useful information flows in and out [1]; compare new announcements against the cursor-and-typing baseline he describes.

5. Separate engineering tools from understanding. When AI is part of a pitch, remember Chichilnisky's point that it executes a transformation rather than explaining the brain [2], and Sejnowski's that we don't yet understand understanding [3].

6. Don't assume implants beat drugs, or vice versa. For brain-wide change like plasticity, MacDougall favors drugs [1]; for precise signals, specificity favors devices [2].

7. Treat telepathy talk as a question, not a forecast. Rogan's vision of deception-free universal language [4] is worth weighing against Strassman's objection that it would take a change in human nature [4] and MacDougall's framing of it as a long-range engineering problem [1].

8. Credit, and follow, the basic science. The progress described rests on decades of academic research, from Georgopoulos to Shenoy and Henderson [1][2]; announcements that build on published work deserve more trust than ones that don't.

9. If it's relevant to you, get on the right list. MacDougall mentions a patient registry for people with quadriplegia interested in future trials [1]; people with photoreceptor-related blindness can follow work on next-generation retinal implants like Chichilnisky's [2].

Sources

1. Dr Matthew Macdougall Neuralink Technologies To Enhance Human Brains · Huberman Lab
2. Dr. E.J. Chichilnisky: How the Brain Works, Curing Blindness & How to Navigate a Career Path · Huberman Lab
3. Dr Terry Sejnowski How To Improve At Learning Using Neuroscience Ai · Huberman Lab
4. Rick Strassman · The Joe Rogan Experience