Redefining our relationship between our brains and computers.
“I know kung-fu”
What it is
Brain-Computer Interfaces (BCIs) enable communication and control without movement, by means of direct input of a person’s neuronal signals into computer systems.1 BCIs can also transform brain signals into available sensory inputs that can in turn modify behaviour, i.e. neurofeedback.
How it works
The interface can either be invasive, e.g. surgically implanted; or non-invasive e.g. a headset.2
BCIs can assess brain activity in various ways, including by measuring signals such as electrical impulses via electroencephalogram (EEG) and Electrocorticography (ECoG); or blood flow via functional near-infrared spectroscopy (fNIRS).3
BCIs can compute various inputs,4 but nevertheless require time-consuming training from users. The extent depends on the adaptability of the patient’s brain.5 Users learn to associate particular patterns of thinking with an output, such as generating a letter on a screen or playing the piano.6
BCI applications are generally still in the experimental phase.7 So far, they have focused largely on assisting disabled users, replacing, restoring, enhancing, or improving their natural central nervous system (CNS) output. Future applications of BCIs may well serve the able-bodied too. Examples:
- Neuro-prostheses.8 A tetraplegic patient was able to walk using a BCI-controlled exoskeleton,9 and some BCI-controlled upper-limb prostheses enable fine-motion.10
- BCIs have allowed communication with locked-in and comatose individuals,11 as well as brain- to-speech communication and brain-to-brain communication, i.e. telepathy.12
- Gaming: several companies have released commercial BCIs as gaming devices.13
- Fashion: BCI-controlled ears that perk up when the excited brainwaves are detected.14
- AI-brain integration, suggested as an extra layer of processing for the human brain.15
Implications and issues
BCIs began the move from lab to commercial application in the past decade but face practical obstacles before adoption becomes widespread. While focussed on rehabilitation, the ethical issues that BCIs raise are similar to those of other therapies: much more controversial is the augmentation of the capabilities of healthy individuals.16 Further issues include:
- Accuracy. Heterogeneity of different people’s brain structures and “noisy” brain signals require that individuals undertake long-term training.17
- Privacy. ‘Brain data’ are highly personal and intimate, raising the fundamental ethical question
of who owns them,18 particularly in the case of devices implanted into the brain.
- Longevity. Current electronics lack the longevity required for implanted BCIs.19
- Liability. Long-term availability of products and product support is a key issue.20
- Bandwidth. Current BCIs do not allow for large transfers of information in either direction: this limits their application.21
- Standardisation of components and systems is required for BCIs to become mass-market products.22
- Security. BCIs could be the riskiest manifestation23 of all the ever-growing number of unsecured‘Internet of Things’ devices.24
- Humanity. Many people feel that it is the way people think that makes humans human. If we improve upon this, it could well – presumably will – have unintended consequences.25
For now, BCIs are a research tool, and a bespoke way of (partially) restoring human function. Some regard BCIs as our best chance at integrating with AIs; others regard BCIs as having the potential to exacerbate existing social disparities, and thereby lose our humanity.◼
1 A fuller definition: “BCIs are systems which measure central nervous system activity and convert it into an artificial output that replaces, restores, enhances, or improves the natural CNS output”. This definition is provided with further context of the history and common terminology of BCI, in Wolpaw, J., & Wolpaw, E. W., 2012. Brain–Computer Interfaces: Something New under the Sun. Brain–Computer Interfaces: Principles and Practice, [e-journal] https://doi.org/10.1093/acprof:oso/9780195388855.001.0001 (paywall).
2 For further details on the applications and implications of worn computer devices see, Cameron, J., and Llewellyn, P., 2018. Wearables. Llewellyn Consulting.
3 For articles describing the techniques used by BCIs to obtain signals from inside and outside of the brain see, Otto, K. J., et al., 2012. Acquiring Brain Signals from within the Brain. Brain–Computer Interfaces: Principles and Practice, [e-journal] https://doi.org/10.1093/acprof:oso/9780195388855.003.0005 (paywall), and Srinivasan, R., 2012. Acquiring Brain Signals from Outside the Brain. Brain–Computer Interfaces: Principles and Practice, [e-journal] https://doi.org/10.1093/acprof:oso/9780195388855.003.0006 (paywall).
4 EEG and fNIRS are relatively portable forms of measuring brain activity, but they are not the most accurate, and they do not allow access to the deeper areas of the brain. Real-time Functional Magnetic Resonance Imaging (rtfMRI) provides far greater detail on regions of the brain that are active during specific actions, such as particular movements or emotional states. However, current MRI machines are physically large, hugely expensive and, for the foreseeable future, will remain reserved for specific functions such as functional brain mapping prior to surgery, and for research purposes. See, Ruiz, S., et al., 2014. Real-time fMRI brain computer interfaces: self-regulation of single brain regions to networks. Biological Psychology, [e-journal] https://doi.org/10.1016/j.biopsycho.2013.04.010 (paywall). For details of a company attempting to replace conventional MRI scanners with more portable devices based on ultrasound and NIR, with corresponding application in BCIs, see Openwater, 2019, Technology. [online] Available at: <https://www.openwater.cc/technology> [Accessed 15 January 2020].
5 This is termed “neuroplasticity” and refers to the physical changes in the brain and central nervous system (CNS) which accompany the learning of new skills or information. Extensive neuroplasticity can also occur following a stroke, when the brain re- wires to compensate for the lost activity. Neuroplasticity has also been detected after 1 hour of BCI use. See, The Free Dictionary by Farlex, 2019, Medical Dictionary. [online], USA: Farlex, Inc. Available at: <https://medical- dictionary.thefreedictionary.com/neuroplasticity> [Accessed 15 January 2020], and Nierhaus, T., et al., 2019. Immediate brain plasticity after one hour of brain-computer interface (BCI). The Journal of Physiology, [e-journal] https://doi.org/10.1113/jp278118.
6 These examples outline two output methods; goal-selection and process-control respectively. In goal selection, the user specifies the desired goal via the BCI, and the output device then produces an action that achieves the goal. In process control, the user controls all the details, producing an action via the BCI. To achieve a high-functioning BCI, a user might require both goal-selection and process-control. Distribution of the control between the user and the output would be beneficial. An example would be a robotic hand able to grasp a glass and play the piano. To grab the glass the user specifies holding the glass as a goal, the hand would carry out the actions required and, using pressure sensors and processors in the hand, would use that feedback to maintain a firm grasp. To play the piano, the user needs to specify each individual finger movement via the BCI, controlling the complete process. See Wolpaw, J., & Wolpaw, E. W., 2012. Brain–Computer Interfaces: Something New under the Sun. Brain–Computer Interfaces: Principles and Practice, [e-journal] https://doi.org/10.1093/acprof:oso/9780195388855.001.0001 (paywall).
7 For a broad look at BCI applications see the review article Chaudhary, U., et al., 2016. Brain–computer interfaces for communication and rehabilitation. Nature Reviews Neurology, [e-journal] https://doi.org/10.1038/nrneurol.2016.113 (paywall).
8 In addition to the larger movements of walking and holding objects, neuro-prostheses focussing on speech rehabilitation have used the small movements of the vocal tract to produce speech, with markedly less distortion than from recording brain activity directly. For reporting on this research see Pandarinath, C., and Ali, Y.H., 2018. Brain implants that let you speak your mind. Nature, [e-journal] https://doi.org/10.1038/d41586-019-01181-y. For the original research article see Anumanchipalli, G.K., et al., 2019.
Speech synthesis from neural decoding of spoken sentences. Nature, [e-journal] https://doi.org/10.1038/s41586-019-1119-1 (paywall). Work to develop artificial neurons has also proved fruitful, with recent developments providing low-power consuming silicon neurons which use the biological feedback of ions rather than electricity. Abu-Hassan, K., et al., 2019. Optimal solid-state neurons. Nature Communications, [e-journal] https://doi.org/10.1038/s41467-019-13177-3.
9 For reporting on the tetraplegic patient who was able to walk using a BCI-controlled exoskeleton see, Jee, C., MIT Technology Review, 2019. A brain-controlled exoskeleton has let a paralyzed man walk in the lab. [online] (October 4, 2019) Available at:
<https://www.technologyreview.com/f/614476/a-brain-controlled-exoskeleton-has-let-a-paralyzed-man-walk-in-the-lab>, [Accessed 15 January 2020]. For the original research article, see Benabid, A. L., et al., 2019. An exoskeleton controlled by an epidural wireless brain–machine interface in a tetraplegic patient: a proof-of-concept demonstration. The Lancet, [e-journal] https://doi.org/10.1016/S1474-4422(19)30321-7 (paywall).
10 For details of the upper body (shoulder to finger) prothesis capable of fine motion, see the DEKA/LUKE arm produced by Mobius Bionics, LUKE arm, [online] Available at: <http://www.mobiusbionics.com/luke-arm/> [Accessed 15 January 2020]. The company CTRL Labs are also active in this area, in which a BCI detects motor neuron signals rather than interfacing directly with the brain. For further details see CTRL Labs. [online] Available at: <https://www.ctrl-labs.com/> [Accessed15 January 2020].
11 Some individuals with ‘disorders of consciousness’, i.e. clinical conditions that follow a severe head injury, such as patients in
vegetative or minimally conscious states, are able to communicate through BCIs. For a vivid account of how fMRI allowed
communication with a comatose individual, see Adrian Owen, 2017. How science found a way to help coma patients communicate. The Guardian, [online] 5 September. Available at: <https://www.theguardian.com/news/2017/sep/05/how-science-found-a-way- to-help-coma-patients-communicate> [Accessed 15 January 2020]. For an original research article describing the first use of a BCI to communicate with a “locked-in” patient with amyotrophic lateral sclerosis (ALS) see Vansteensel, M. J., et al., 2016. Fully Implanted Brain–Computer Interface in a Locked-In Patient with ALS. The New England Journal of Medicine, [e-journal] https://doi.org/ 10.1056/NEJMoa1608085. For a detailed review on the various applications of BCIs in individuals with disorders of consciousness, see Gibson, R. M., et al., 2016., Brain-computer interfaces for patients with disorders of consciousness. Progress in Brain Research, [e-journal] https://doi.org/10.1016/bs.pbr.2016.04.003 (paywall).
12 The University of California San Francisco (UCSF) have developed, in collaboration with Facebook, a BCI capable of decoding individual words and phrases purely from brain activity in real-time. To achieve this, they made a contextually aware BCI, able to consider both the (pre-defined) question asked and the likely answers to increase the accuracy in predicted words to ~75%. The Defense Advanced Research Projects Agency (DARPA), have also expressed interest in the field, aiming for direct brain-to-brain communication. For the press release on the UCSF work see, Weiler, N., 2019, Team IDs Spoken Words and Phrases in Real Time from Brain’s Speech Signals. University of California San Francisco, [online] Available at:
<https://www.ucsf.edu/news/2019/07/415046/team-ids-spoken-words-and-phrases-real-time-brains-speech-signals> [Accessed 15 January 2020], and for the original research article see, Moses, D. A., et al., 2019, Real-time decoding of question-and-answer speech dialogue using human cortical activity, Nature Communications, [e-journal] https://doi.org/10.1038/s41467-019-10994-4. For information on DARPA’s N3 Programme, see DARPA Outreach, 2019. Six Paths to the Nonsurgical Future of Brain-Machine Interfaces, DARPA, [online] 20 May. Available at: <https://www.darpa.mil/news-events/2019-05-20,> [Accessed 15 January 2020].
13 The most commercially successful of these has been the Star Wars Science Force Trainer, which uses an EEG headset manufactured by NeuroSky. Users assume the role of a trainee Jedi and, using brain activity, manoeuvre various holograms. Computer games using BCIs are in production but have yet to reach the market. For details on Nehttps://store.neurosky.com/#other-productsuroSky’s Star Wars themed toy see, NeuroSky, The Force Trainer II: Hologram Experience, [online] Available at: https://store.neurosky.com/#other-products [Accessed 15 January 2020].
14 Created by neurowear, a project team based in Tokyo focused on creating “communication for the near future”, it involved a headband with cat ears and NeuroSky’s MindWave EEG reader, which would quantify the mood of the wearer. The ears would perk up upon detection of “concentration”. For details and videos of the necomimi ears in action, see Neurowear, 2011. Projects / nekomimi. Available at: <http://www.neurowear.com/projects_detail/necomimi.html> [Accessed 15 January 2020].
15 The acceleration of technological progress since the 20th Century has led many to speculate that, at some point, the learning abilities and complexity of AIs will surpass that of humans, leading to an ‘intelligence explosion’. At this point – the so-called ‘Singularity’ – that humans will supposedly be superfluous. Some therefore see effective BCIs constructively, in that they might integrate with AIs, thereby ensuring the continuing relevance of humans. For the first description of ‘The Singularity’ see, Vinge, V., 1993, The Singularity. [online] Available at: <https://mindstalk.net/vinge/vinge-sing.html> [Accessed 15 January 2020]. For an interview with Bryan Johnson, an advocate of integrating humans with AI via BCIs, and founder of the company Kernel to attempt just that, see Levy, S., 2017. Why You Will One Day Have a Chip in Your Brain. Wired, [online] Available at: <https://www.wired.com/story/why-you-will-one-day-have-a-chip-in-your-brain> [Accessed 15 January 2020].
16 For a discussion of the ethical issues facing BCIs and associated ‘neurotechnologies’, with a focus on AI, see Yuste, R., et al., 2017. Four ethical priorities for neurotechnologies and AI. Nature, [e-journal] https://doi.org/10.1038/551159a . For a discussion of the ethics surrounding BCIs specifically see, Drew, L., 2019, The ethics of brain–computer interfaces. Nature, [e-journal] https://doi.org/10.1038/d41586-019-02214-2.
17 Interpreting the noisy data of brain signals is difficult, and there are many different algorithmic approaches to extract particular features. For a useful overview of the basic processes involved, see Krusienski, D. J., et al., 2012. BCI Signal Processing: Feature Extraction. Brain–Computer Interfaces: Principles and Practice, [e-journal] https://doi.org/10.1093/acprof:oso/9780195388855.003.0007 (paywall).
18 These issues are discussed extensively in Greenburg, A., 2018. Inside the Mind’s Eye: An International Perspective on Data Privacy
Law in the Age of Brain-Machine Interfaces. Social Science Research Network [e-journal], https://dx.doi.org/10.2139/ssrn.3180941.
19 The problem of integrating hard metallic electrodes that conduct electricity with the soft organic tissue of the brain in which the signals are transmitted via the movement of ions has been a long-standing challenge in the development of BCIs. Producing long- lasting, reliable probes has been difficult. While silicon probes have been used, more recent approaches include using hydrogels and thin, flexible polymer electrode arrays. Details of the body’s reaction to electronics implanted in the brain can be found in the review article, Wellman, S. M., and Kozai T. D. Y., 2017. Understanding the Inflammatory Tissue Reaction to Brain Implants To Improve Neurochemical Sensing Performance. American Chemical Society Chemical Neuroscience [e-journal] https://doi.org/10.1021/acschemneuro.7b00403. For original research articles detailing advances in these approaches see Chung, J.E., et al., 2019. High-density, long-lasting, and multi-region electrophysiological recordings using polymer electrode arrays.
Neuron, [e-journal] https://doi.org/10.1016/j.neuron.2018.11.002, ; Huang, WC., et al., 2018. Ultracompliant Hydrogel-Based Neural Interfaces Fabricated by Aqueous-Phase Microtransfer Printing. Advanced Functional Materials, [e-journal] https://doi.org/10.1002/adfm.201801059, [Accessed 11 November 2019]; Musk, E., NeuraLink, 2019. An integrated brain-machine interface platform with thousands of channels. biorxiv, [e-journal] https://doi.org/10.1101/703801.
20 As devices become more personal and invasive, the long-term support and future of a company becomes far more important. The sensitivity of data in this context is a key concern, a comparison being the genetic data owned by direct-to-consumer genome sequencing companies. In practice there are no legal obstructions to selling this genetic data once it has been stripped of personal information, despite the risk of re-identification. In 2009, there was considerable concern that when deCODE went bust it would do this, although it guaranteed that no data would be sold as part of its bankruptcy. Another pertinent example is that of Nest, a
‘Smart Home’ product company that sold smart thermostats. Upon their purchase by Alphabet (Google’s holding company) they inactivated the existing devices. This proved an inconvenience for many; but the potential inconvenience of a BCI lacking support is likely far greater. For a legal discussion of the risks to genomic data upon bankruptcy of DTC genome sequencing companies, see Vorhaus, D., and Moore, L., 2009. What Happens if a DTC Genomics Company Goes Belly Up?. The Privacy Report [online] Available at: <https://theprivacyreport.com/2009/09/18/what-happens-if-a-dtc-genomics-company-goes-belly-up> [Accessed 15 January 2020] and the follow-up, Moore, L., and Sherlock, E., 2009. Federal Privacy Regulation and the Financially Troubled DTC Genomics Company. The Privacy Report, [online] <https://theprivacyreport.com/2009/10/27/federal-privacy-regulation-and-the-financially- troubled-dtc-genomics-company> [Accessed 15 January 2020]. For details of Nest and the inactivation of their devices, see Price, R., 2016. Google’s parent company is deliberately disabling some of its customers’ old smart-home devices, Business Insider, [online] Available at: <https://www.businessinsider.de/googles-nest-closing-smart-home-company-revolv-bricking-devices-2016-4> [Accessed 15 January 2020].
21 Most current BCIs lack the bandwidth to be able to communicate with more than a few million neurons, thereby falling far short of the 86 billion neurons in the human brain. Even with this limited communication currently available, extraordinary feats have been achieved, but future applications would greatly benefit from the increased bandwidth. For a recent description of a BCI with increased bandwidth see, Musk, E., NeuraLink, 2019. An integrated brain-machine interface platform with thousands of channels. biorxiv, [e-journal] https://doi.org/10.1101/703801.
22 Some regard BCIs as an orphan technology that lacks the market demand to sustain an industry. As long as the value proposition for healthy users is lacking, that may well remain the case. The outcome will depend on standardisation of products. This is suggested as a concern in the Horizon 2020 Roadmap by the Brain/Neural-Computer Interaction (BCNI) consortium. See, BCNI, 2015. Graz University of Technology, [online] Available at: <http://bnci-horizon-2020.eu/roadmap> [Accessed 15 January 2020].
23 Should a hacker be capable of accessing a BCI, one potential risk is that of brain malware. One of the primary issues facing these devices is the failure to implement security by design, instead leaving this to be considered separately, if at all. The following two articles outline these issues, and suggest countermeasures, with a focus on current generation BCIs. See, Li, QQ., et al., 2015. Brain- Computer Interface Applications: Security and Privacy Challenges. 1st Workshop on Security and Privacy in Cybermatics — 2015 IEEE Conference on Communications and Network Security, [online] https://www.doi.org/10.1109/CNS.2015.7346884, and Bonaci, T., Calo, R., & Chizeck, H. J., 2015. App Stores for the Brain: Privacy and Security in Brain-Computer Interfaces. IEEE Technology and Society Magazine, 34(2), 32–39 [online] https://www.doi.org/10.1109/mts.2015.2425551.
24 For further details on the applications and implications of the Internet of things (IoT) see Cameron, J., and Llewellyn, P., 2018, Internet of Things. Llewellyn Consulting.
25 The use of BCIs depends on the plasticity of the brain, rewiring individual’s thought-processes. If BCIs are to grow in complexity it is only to be expected that they will also produce associated deeper changes to our brains, potentially altering fundamentally the way individuals think.