Futurist space exploring car manufacturing entrepreneur Elon Musk made headlines at the end of March when The Wall Street Journal ran a story about his new company Neuralink—a Californian company registered in July 2016 as a medical research company. Musk confirmed the newspaper article with a tweet saying “Long Neuralink piece coming out on @waitbutwhy in about a week. Difficult to dedicate the time, but existential risk is too high not to,” quickly followed by a flurry of tech blogs speculating about the venture. [you can learn about the follow up article here]
Musk has spoken in the past about developing a “neural lace” which is a brain-computer interface that can monitor neural activity. This device, already in research development, is a thin mesh that is injected, via a syringe, onto an area of the brain where it can conform to the surface of that brain area and pick up a fine grained neural activity—it is taking electrode monitoring of brain activity to the next level. Such technology has been used experimentally on mice, into various brain regions including the hippocampus to record brain activity (Liu et al, 2015).
Invasive technologies like this have been used to treat neurological disorders like Parkinson’s Disease. Known as deep brain stimulation, this surgical implanting of electrodes to deliver electrical stimulation into targeted areas of the brain—typically the thalamus, subthalamic nucleus, and globus pallidus—that block signals causing Parkinson’s symptoms. NeuroPace is a Silicon Valley company that develops and manufactures implantable devices to treat epilepsy. Their RNS System is rather like a pacemaker that monitors brain activity in specific areas via implanted electrodes and a monitoring device delivers stimulation whenever seizure activity is detected. The monitor, called the neurostimulator, is placed in the skull, hidden under the scalp, and can wirelessly transmit information to a remote computer.
The future for neural lace type of interfaces according to Musk and other visionaries like Bryan Johnson, founder and CEO of Kernel, is much grander than simply blocking neural pathology. After selling his payments company Braintree to PayPal for $800 million, Johnson founded Kernel to “build advanced neural interfaces to treat disease and dysfunction, illuminate the mechanisms of intelligence, and extend cognition.” Johnson’s company wants to bio-hack the brain to enhance cognition—that we might “co-evolve with machines.” The vision is to create neural prosthetics for human intelligence enhancement.
On October 20, 2016, Johnson announced that he was committing $100 million to Kernel to enhance human intelligence. “We’re starting to identify the mechanisms underlying neural code and make them programmable. Our biology and genetics have become increasingly programmable; our neural code is next in line,” he wrote in his announcement. “Programming our neural code will enable us to author ourselves and our existence in ways that were previously unimaginable.”
What Neuralink and Kernel bring to the research and development of brain-computer interface is money and the nimbleness of Silicon Valley start-ups. Bringing cleaver people together, with the resources they need, to go boldly where no venture has gone before, is what entrepreneurs like Musk and Johnson do so well. But it is unlikely the technology side of things that will determine the speed of advancement for intelligence-enhancing brain prosthetics. Understanding how the brain works, how we are intelligent, is just in it’s infancy. Consciousness and how neural activity relates to our subjective experience is still largely a mystery—neural architecture modulates the experience of the mind and yet the mind can seemingly modulate neural architecture. Yes we have increasingly better imaging tools, we can see the brain in action like never before, but there is still so many dots we haven’t connected yet and dots we haven’t seen. We are still a long way off from the Eternal Sunshine of the Spotless Mind type of encoding and recoding of neural data.
Although there may be near-term goals of combating diseases and advancing neuroscience research, both Musk and Johnson want this technology to keep humans ahead of the intelligence race. The fear is that artificial intelligence (AI) will hit an exponential that will quickly overtake our own, leaving us in an apocalyptic scenario rather like The Matrix. To enhance our cognitive capacity by interfacing with machines is, for these futurists, is the answer to staying ahead in the race.
How advanced is this AI that some fear will one day may be our master? We have some very sophisticated algorithms that can simulate the output of human intelligence—like driving a car from one side of the city to the other. But these algorithms, even the deep neural networks that are modelled after non-linear connections of neural architecture, are still crude approximations of a few basic elements of neural function. Nevertheless, for specific tasks the performance of such AI is impressive. Take for example Google’s AlphaGo developed by DeepMind, the Go-playing AI that beat the world Go master Lee Sedol. Go is a highly sophisticated game with an enormous number of possible moves and masters of the game rely on an intuitive knowing to gain an edge against worthy opponents. AlphaGo utilised deep learning, using networks of ‘nodes’ that mimic layers of neural connections, and it’s creators fed it massive amounts of data related to the game Go so it could learn. It played against itself, learning from it’s mistakes and improving it’s game—it gained it’s own type of intuition for the game. After AlphaGo beat Lee, Demis Hassabis, co-founder of DeepMind, said that DeepMind was “very pleased that AlphaGo played some quite surprising and beautiful moves.”
The fact that Hassabis was surprised at some of AlphaGo’s moves is telling—even though the machine is not thinking in the same way that we are, it is, in it’s own way, demonstrating some autonomy of “thought” apart from the control of it’s creators. It does not take much stretching of the imagination to end up in The Matrix. But how far away is true autonomy of AI cognition and what we would consider general AI? Typically we underestimate how rapidly technological advances are happening (with the exception of Hollywood that has us doing interplanetary space travel in 2001 while Blade Runners take down rogue bioengineered replicants in 2019), but who knows? Maybe truly autonomous thinking machines are in our near future—a new species of synthetic, yet cognitively aware, entities—and maybe they will be driven by a quest for dominance, maybe not.
In our imagination of an apocalyptic future we may be projecting our own, distinctly human, lust for domination and control onto neither good nor evil machines. Of course this takes us down the path of philosophising about the nature of evil and if indeed AI could or would evolve with evil intent. Would our own digital Adam and Eve rebel against us in a desire to be just like us?
Whatever the ultimate motivations are for Musk and Johnson, there will no doubt be some very positive outcomes for neuroscience and our understanding of how the brain works, quite apart from intelligence-enhancing brain prosthetics. We hope to see therapeutic breakthroughs, like those developed by NeuroPace, as a natural part of this journey toward enhancing our own cognition and input/output capacity.
Liu, J., Fu, T., Cheng, Z., Hong, G., Zhou, T., Jin, L., Duvvuri, M., Jiang, Z., Kruskal, P., Xie, C., Sou, Z., Fang, Y., & Lieber, C. M. (2015). Syringe-injectable electronics. Nature Nanotechnology, 10, 629-637. doi: 10.1038/NNANO.2015.115