According to an article by Tim Urban, Neuralink is only the first step, with the final goal being to connect humans with AI to prevent human extinction. To that end, the company is trying to invent invasive devices that have long-term biocompatibility, can be easily embedded like LASIK. Such a device should also have wireless capabilities and be able to read and write the activities of millions of neurons simultaneously. The company raised 27 million dollars recently.

Meanwhile, Bryan Johnson, a Silicon Valley entrepreneur, has invested 100 million dollars of his own funds to launch the neural link company Kernel, which ultimate goal is to enhance human intelligence.

The British company ARM has also started to develop a brain embedded chip for bi-directional BMI utilizing the ultra small and energy efficient microprocessors the company is famous for. The work is in collaboration with the Center for Sensorimotor Neural Engineering at the University of Washington. The goal of this partnership is to allow patients with Parkinson’s disease, Alzheimer’s disease, and paralysis to regain diminished motor abilities such as movement and the sensation of touch.

On another front, DAPRA has been promoting the development of BMI for more than a decade. One recent project, the Neural Engineering System Design (NESD) program, has granted the company Paradromics with 18 million dollars of research funds. Paradromics is trying to develop a micro-implant (smaller than two 5 cent coins stuck together) for the concurrent connection >65,000 neurons, with the goal of decoding and understanding the “chatter” of brain activity.

Higher Resolution Implantable BMI are Technologically Feasible

“Neural dust”, millimeter-sized implantable sensors, have been reduced to as small as 3×1×1mm by Dongjin Seo (the co-founder of Neuralink), and there is development underway to miniaturize it down to 50μm. As an engineering issue without hard theoretical constraints, it is only a matter of time until this goal is realized.



(Image from UC Berkeley)

Moreover, at the Center for Information and Neural Networks (CiNet) in Japan, research of chips that read neural signals wirelessly from an embedded device in the brain has already begun; currently, it is able to read up to 4,096 channels of ECoG signals at speeds of 128Mbps [1].

Kernel’s Bryan Johnson former interest was memory function in particular, tried to enhance it with Professor Theodore Berger from the University of Southern California. In Theodore Berger’s work, a model that describes the stream of the neural activity inside the hippocampal regions associated with memory had been developed [2]. The model was verified using mice, demonstrating that it could improve performance in a memory task.



(Image from: J Neural Eng. 2011 Aug; 8(4): 046017.

There is another research done by Pittsburgh University, which tested the effectiveness of electrode arrays embedded into the somatosensory cortex of a subject suffering from long-term spinal cord injury [3]. Microstimulation by the arrays could induce the sensation of touch in the subject, with stimulus strength and location affecting the intensity and location of sensation. This study provides an important precedent in bi-directional BMI that ARM and the Washington University are going to do.



(Image from: Sci Transl Med. 2016 Oct 19;8(361):361ra141.

What Will be Possible if Implantable BMI Can Achieve High Accuracy?

What will we be able to do if the technology to read and write the activities of one million neurons at the same time is developed, which Neuralink, Kernel, and Paradromics are trying to do?

The brain is said to be functionally localized roughly like this:



(Image from: The Dana Foundation

Currently, the most focused on the area of BMI is the motor area, which is being directly targeted to do things like moving robotic armsand simulate keyboard input into a computer. Attempts to do this are being done by reading neural activities through embedded ECoG or electrode arrays, but if we can read the activities of tens or hundreds of thousands of neurons, much more sophisticated and robust BMI applications can be realized.

The somatosensory area has been validated for decades as associated with tactile sensation. As mentioned earlier [3], research at Pittsburgh University found that it would be possible to induce tactile sensation by inserting a 2×4mm sized 6×10 electrodes array and stimulating it. If such methods are refined to provide tactile feedback at μm spatial resolutions, it would be possible to induce fine-tuned variations of strong and weak pressure, hot and cold, and pain.

Research in visual cortex has had strong advances, for example, Kyoto University professor Yukiyasu Kamitani found that it is possible to decode what type of object is seen by the subject from the activities of the visual area using fMRI [4]. If measuring over the tens of thousands of neural activities of the visual area with a μm spatial resolution and several hundreds Hz temporal resolution, it will be possible to decode what is seen from the visual area more accurately, or to go a step further, we will be able to stimulate visual areas and control what is seen.

In Theodore Berger’s research, 16 channel electrodes were inserted into CA3 and CA1 of the hippocampus to create a communication model of the neural pulse between CA3 and CA1 [2]. If we can expand this to observe the activities of neurons in the hippocampus in the units of tens of thousands, at the very least, we may better understand the mechanisms of memory formation, with the eventual goal of building models that allow the accurate implantation of artificial memories.

Feelings and emotions are thought to be mainly controlled by the limbic system, with the amygdala, ventral tegmental area, and anterior cingulate cortex, etc. are related to emotion [5]. If it is possible to read the activities of neurons in these areas in high definition, it can open the possibility of decoding emotions or understanding what emotions are. Furthermore, like mice that can be conditioned to push a lever through VTA self-stimulation (activating reward pathways of the brain), it should be possible to induce feelings and emotions, thus producing behavioral changes from it [6].

Research on reading thoughts using fMRI have been published [7]. Through reading which part of the brain is active and how active it is, it is possible to estimate whether it is related to people, places, or something else. However, a caveat is that fMRI has unsatisfactory spatial resolution and worse temporal resolution. So, efforts to produce BMI devices with higher spatial and temporal resolutions are invaluable and will make it possible to better analyze human thought patterns.

Contributions to Neuroscience and Humanity

In research on the brain and neuroscience at present, fMRI, EEG, and MEG are widely used. With each of these methods, the subject has to stay laid or sat down and are limited to restrictive tasks [8]; an unusual condition far from daily life.

If a BMI device that can connect wirelessly and make ultra-high precision measurements is made, we can absolve these conditional restraints, allowing the monitoring of mass quantities of neurons during the subject’s daily life. Therefore, while some forms of research will continue to require rigid experimental conditions, doors will open for research that aims to be as unintrusive as possible for the subject. This will result in huge amounts of data, defining an era of Big Data analysis in neuroscience.

Knowledge of how we feel, think, and behave will thus develop rapidly, and at the same time, technology to change how we feel, think, and behave will inevitably develop as well.


  1. Ando H, Takizawa K, Yoshida T, Matsushita K, Hirata M, Suzuki T, Wireless multichannel neural recording with a 128-Mbps UWB transmitter for an implantable brain-machine interfaces, IEEE Trans Biomed Circuits Syst, 10(6):1068-1078 (2016).
  2. Theodore W Berger, Robert E Hampson, Dong Song, Anushka Goonawardena, Vasilis Z Marmarelis, and Sam A Deadwyler. A cortical neural prosthesis for restoring and enhancing memory. Journal of Neural Engineering. 2011 Aug; 8(4): 046017.
  3. N Flesher, Sharlene & Collinger, Jennifer & Foldes, Stephen & M Weiss, Jeffrey & Downey, John & Tyler-Kabara, Elizabeth & Bensmaia, Sliman & Schwartz, Andrew & L Boninger, Michael & Gaunt, Robert. (2016). Intracortical microstimulation of human somatosensory cortex. Science Translational Medicine. 8. . 10.1126/scitranslmed.aaf8083.
  4. Horikawa T, Kamitani Y, Generic decoding of seen and imagined objects using hierarchical visual features. Nature Communications. 2017 May 22;8:15037.
  5. Rajmohan V, Mohandas E. The limbic system. Indian Journal of Psychiatry. 2007;49(2):132-139. doi:10.4103/0019-5545.33264.
  6. Vincent Pascoli, Jean Terrier, Agnès Hiver, Christian Lüscher, Sufficiency of Mesolimbic Dopamine Neuron Stimulation for the Progression to Addiction, In Neuron, Volume 88, Issue 5, 2015, Pages 1054-1066, ISSN 0896-6273,
  7. Marcel Adam Just et al. Predicting the Brain Activation Pattern Associated With the Propositional Content of a Sentence: Modeling Neural Representations of Events and States. Human Brain Mapping, June 2017.
  8. Carter M, Shieh JC. Guide to Research Techniques in Neuroscience. Burlington, MA: Academic Press; 2015.