Augmented Brains Are Not The Future of Humanity

Fundamental cognitive barriers will prevent humans from merging with digitial intelligence

A recently reconstructed wiring diagram for a fruit flies’ approximately 140,000 neurons

Even in a benign AI scenario we will be left behind,” Elon Musk said. “With a high-bandwidth brain-machine interface, we can actually go along for the ride. We can have the option of merging with AI.”, Wired

Connecting neurons in the brain to a computer through so-called brain-computer interfaces (BCIs) has achieved some remarkable recent successes and has been used to restore the ability to walk, move a cursor, play chess, type and even feel shapes and textures.

While Musk’s Neuralink is best known, similar research is conducted by various research institutions and companies including one, called Synchron, that is backed by Bill Gates and Jeff Bezos. All are currently focused on neurotherapeutic applications, but some companies have as the ultimate goal to enhance human capabilities beyond the limitations of our biological brain and body. Precision Neuroscience’s tagline is “Connecting human and artificial intelligence”, while Synchron’s vision reads: “Neurotechnology to address the limitations of the human body”. Elon Musk believes merging with AI through BCIs is required to keep up with the machines.

Physically, we can already create superhumans. Exoskeletons enhance our strength and night vision goggles extend our perceptual range. Even fighter jet pilots and forklift operators are, in a way, augmented superhumans. But these technologies build on, and are thus constrained by, our existing ways of interacting with the external world.

Physically, exoskeletons and fighter jets already create superhuman capabilities but these technologies build on, and are thus constrained by, our existing ways of interacting with the external world.

BCIs promise a more fundamental expansion of abilities by circumvention of our cognitive bottlenecks. A 2020 RAND report (a rare example of an article that aims to be more specific than the general promise of ‘merging with AI’ and creating ‘superhuman’ capabilities) lists among the potential augmentation possibilities more rapid and accurate decisions, controlling machines with your thoughts, and cognitive improvement. In the longer term, the idea is that high-bandwidth connections would facilitate more efficient exchange or even integration of thoughts, intentions, memories, and other information. Your thoughts could control instruments and machines, while you can quickly exchange vast amounts of information with the external world, or with a “digital version of yourself”, and further into the future even acquire complex skills such as playing tennis or fighting, reminiscent of the kung fu skills uploaded into Neo’s brain in The Matrix.

Establishing physical connections to control and read the brain is highly challenging both technically and because our current understanding of the brain’s neural architecture is incomplete. Last year, scientists reconstructed a complete wiring diagram (or connectome) for fruit flies (shown in the picture at the start) whose brain contains around 140,000 neurons. The human brain has 86 billion neurons and trillions of connections and is thus not only far larger but freezing a brain seconds after death to prevent degradation also has obvious ethical and practical challenges. Furthermore, mapping the brain’s intricate wiring is not the same as understanding it. Even in artificial neural networks, where the internal parameters are accessible, interpretation remains very difficult. Other challenges include vast brain differences between individuals, and the damage to other neurons while connecting to those deep within the brain. It is far from clear these obstacles can ever be overcome sufficiently to make the risk acceptable for healthy individuals. But given the recent impressive successes with patients, optimism about continued progress is understandable. Besides, much research focuses on non-invasive methods, like stents, infrared, and ultrasound, that reduce or eliminate brain damage although such technologies would sacrifice spatial resolution, accuracy, and control compared to making direct connections.

Conceptual fallacies

Largely overlooked is that the challenges for brain augmentation of healthy individuals face are not just technical but also conceptual. The promise of superhuman skills through a merger with artificial intelligence relies on a view of the brain that misconstrues how thoughts arise and how personalities are sustained.

Speeding up?

Let’s start with the benefits from direct control of instruments and machines. Some of the specific potential military BCI applications described in the RAND report include ‘transfer commands to systems’, and ‘reduce reaction time’. But which inefficiencies are BCIs addressing that would allow for improvements in command and control? Typing with your thoughts has achieved a speed of approximately 18 words per minute, a major benefit for the speech-impaired, but around half the speed of an average adult, and less than one tenth the speed of the fastest typists. Existing speech-recognition software and regular physical controllers are highly efficient, leaving little room for improvement for BCIs that bypass parts of the prefrontal cortex (responsible for executive function) and the downstream cognitive-motor pathways. In high-stakes scenarios, such as intercepting an incoming rocket or drone warfare, a few milliseconds can indeed make all the difference. But if humans are part of the decision process, the time needed for briefing, communication, or human-controlled safety protocols will far exceed the time saved from bypassing some of the usual neural processes required to push a button. When success depends on milliseconds, humans are removed from final execution with pre-determined decision parameters that allow for instantaneous and autonomous action. Moreover, direct brain connections that circumvent conscious consideration would introduce error rates that are likely unacceptable for high-stakes decisions (more on that below). While beyond the military, minimally faster decision-making would matter very little except for fields like high-frequency trading where human brains have also been largely removed from the execution process.

Higher bandwidth information exchange?

“Over time, I think we will probably see a closer merger of biological intelligence and digital intelligence….It’s mostly about the bandwidth, the speed of the connection between your brain and the digital version of yourself, particularly output”, Elon Musk

Beyond decision making speed, the broader BCI promise is to exchange vastly more information between our brains and the computer, or a digital version of yourself. For this to be true you need to believe your brain is already processing many thoughts in parallel but is constrained in acting on them by the bottleneck of our consciousness, speech, and — in the case of actions — by our limited number of limbs. But while our brain contains many terabytes of information, and billions of processes, it is designed to control and maintain a single body. Sometimes a shower, sleep, or another break from active consideration can trigger new ideas in a process that psychologists call incubation, but such insights are limited in scope and complexity until shaped by continued conscious thinking. A thought becomes an idea when you consider it, and an urge becomes a decision only when you make it. Real-time brain imaging can reveal brain activity and indicate mental alertness, and could be used to indicate appetite or stress before you consciously become aware of such feelings. A well-calibrated BCI could read your brain and trigger a water dispenser or order food before you become aware of those needs. But those predictions will be far from perfect, because forming a specific decision, and turning a thought into language, changes it. Only when you are about to decide you are craving beef, do you remember your goal to eat less meat, or the salmonella that was just found, or the fact that you have leftovers at home. After you decide to get a tea do you realize you want to hydrate with water first, and only after that does the smell of coffee convince you to have an espresso after all. BCIs that short-cut these processes result in mistaken decisions. The difficulties of reading a person’s emotional state to trigger well-defined actions compound when the goal is to read more complex thoughts or ideas. No matter how integrated or advanced BCIs become, broadband connections to our unconscious brain will not extract well-formed opinions, nor simultaneous instructions for many drones — another example mentioned in the RAND report — or any other specific commands. The BCIs’ successes were achieved with people consciously playing chess, and consciously trying to write, not by tapping into the patients’ unconscious minds while they were watching YouTube. Complex decisions and thoughts arise as the culmination of a process that integrates a broad range of inputs from across the brain and body, including memories, emotional states, explicit and implicit sensory processing, the instructions you just received, and the decisions and thoughts being actively formed in that very moment. Improving reliability and precision is central to current BCI research, but the futurists’ belief that better technology can bridge the fundamental barrier between general brain signals and specific actions is flawed. You cannot remove the conscious element from decision-making and still expect instructions precise and reliable enough to be useful. There is no well-formed strategy or speech in the deeper unconscious layers of your brain that direct connections can unlock. Not even a well-formed sentence or conceptual thought. And as far as you can use correlations between neuron groups and thoughts or preferences, they will lack accuracy because you rely on correlations between brain states and decisions, and the moment you become conscious of it yourself, you might change it. Allowing brain signals to take action before we have consciously made any decision will be like putting Microsoft’s Clippy in charge of your life.

A thought becomes an idea when you consider it, and an urge becomes a decision only when you make it. Broadband connections to our unconscious brain will not extract specifc opinions or complex commands, nor even a well-formed sentence.

Skill and knowledge uploads?

“We’re going to be a combination of our natural intelligence and our cybernetic intelligence,” “and it’s all going to be rolled into one. We are going to expand intelligence a millionfold by 2045” Ray Kurzweil author of “The Singularity Is Nearer”

In the other direction, from the outside world into the brain, the more ambitious promises also dissipate on closer inspection. I can’t upload Federer’s tennis skills because my brain does not contain a ‘tennis skills’ chamber that can be filled or replaced. Our brains are physically very different and motor skills are distributed across the brain making them inseparable from our memories, emotions, and personality. Replacing brain sections involved in movement might enhance certain skills but might also erase memories I have from playing tennis, while altering other behaviors, and thus changing my personality. While a full brain redesign to replicate most of Federer’s brain processes would not only kill one’s original self but also be far inferior to the original due to physical differences beyond the brain, like muscles, reaction speed, and hormonal activity that regulates stress, energy levels, and metabolism. The same applies to seemingly simpler things such as facts or sensory inputs. You cannot download (or merge with) the Encyclopedia Brittania or the ability to speak Chinese without making changes throughout your brain, with uncertain results for both the skills acquired and your personality. A history professor’s knowledge of history is related to her feelings of enjoyment, to her goals and aspirations, to her identity, and to her teaching. These connections enable extraction, association, and higher-level reasoning based on this information. Standalone facts that don’t affect any of your thoughts or behaviors and can’t be recalled, are meaningless. This is not to say that BCIs might one day accompany external brain stimulation and drugs to stimulate brain plasticity, or to enhance access to lost memories, and possibly nudge a brain into more specific skill directions. Non-invasive neurostimulation through electric current has shown some limited results in improving learning. Skill acquisition could thus be accelerated with more advanced BCIs, when combined with actual training and other activities that ensure consistency and establish the right connections. But you won’t upload Biles’ gymnastics skills if you spend your days on Instagram, nor will you ever make substantial progress learning Italian without studying it. For people with serious learning difficulties this could become a viable therapy, and perhaps some risk-tolerant healthy people will choose similar BCI assisted therapy. But even if all technical challenges are solved, the outcome for the individual will remain uncertain and far from the visions of merged or superhuman intelligence imagined by Kurzweil and Musk.

Superhuman already

The promises for developing superskills for healthy humans from direct, high bandwidth brain computer connections thus dissipate upon closer inspection. Impressive success in aiding disabled people does not extrapolate beyond natural human limitations. Because restoring brain functions differs fundamentally from creating super-human capabilities that disregard the processes that govern our thoughts and shape our personality.

Modern humans are cyborgs already. Though mediated by our legacy limbs, senses, and brains, we have instant access to nearly all of the world’s knowledge and we already command a multitude of supercomputers and machines in real time.

That is, if we manage to avoid wasting our time on X telling people to fuck themselves in their faces.