Introduction

Brain–computer interface (BCI) technologies have been around for decades [1,2]. The underlying principles are simple: BCIs can be invasive or noninvasive (Fig 1). A typical invasive BCI consists of electrode probes inserted into the brain, which can record and/or stimulate neural activity in specific brain regions. The probes are connected to a computer (outside the brain and usually outside the body) that can process the recorded signals in various ways. These BCIs can record both local field potentials and—through application of spike sorting algorithms—spike trains from individual neurons. When BCIs are used in the “inside out” direction, they can control an external system such as a prosthetic limb or a speech synthesiser [3,4]. When they are used in an “outside in” direction, they can be used to drive neural activity to bring about changes in the brain, mind, and body. Bidirectional BCIs work in both directions, creating opportunities for closed-loop neurofeedback.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Varieties of BCI.

Noninvasive BCIs record from outside the skull using methods such as EEG. These methods are usually used only to record brain activity, but can be combined with brain stimulation methods such as (noninvasive) TMS. Invasive BCI methods include ECoG in which grids of electrodes sit on top of the brain surface, underneath the skull, and electrodes implanted inside the brain. ECoG and implanted electrodes can both record and stimulate brain activity. Brain icon by Servier https://smart.servier.com/ is licensed under CC-BY 3.0 Unported https://creativecommons.org/licenses/by/3.0/. BCI, brain–computer interface; EEG, electroencephalography; ECoG, electrocorticography; TMS transcranial magnetic stimulation.

https://doi.org/10.1371/journal.pbio.3002899.g001

Another variety of invasive BCI is based on electrocorticographic (ECoG) recordings, which involves surgical implantation of a grid of electrodes that lie on the surface of the cortex, beneath the skull, but which do not penetrate the neural tissue [5]. ECoG grids provide wider coverage than implanted electrodes, but do not allow the identification of individual neurons. While the majority of ECoG-based BCIs are used in recording (inside out) mode, they can also be used for stimulation [6].

Noninvasive BCIs, as the name suggests, do not require surgical implantation. Instead, they rely on wearable technologies such as electroencephalogram (EEG) and functional near-infrared imaging [7]. Noninvasive BCIs are easier to deploy, but will generally lack the precision of invasive BCIs and will also generally lack the ability to stimulate brain activity, exceptions may include coupling noninvasive recording methods with noninvasive stimulation methods such as transcranial magnetic stimulation (TMS) and transcranial electrical stimulation. In this Essay, we focus on invasive BCIs.

Despite their long history [8], BCIs have recently gained prominence in the neurotechnological landscape. In part, this has to do with the entry of new companies such as Neuralink, which have strong public profiles. In addition, technological advances in electrode design, and in the possibilities afforded by recent advances in artificial intelligence (AI) and machine learning for analysing recorded neural data. There is now widespread agreement that BCIs could quite soon develop into a market worth billions of dollars annually [9]. However, two application domains for BCIs need distinguishing. These domains partially overlap in practice, but they have largely distinct aims. The first domain is medical or clinical treatment, and the second domain concerns neural and/or cognitive enhancement.

Clinical applications motivated the first BCIs, with the first human clinical trials dating from the 1990s [10], and they remain the dominant goal for most BCI developers. Inside-out BCIs have been used to help decode movement and speech commands from paralysed patients [11,12]. Outside-in BCIs have been used to ameliorate symptoms of Parkinson’s disease, control epilepsy, and relieve severe depression [13]. Some of these applications come under the rubric of “deep brain stimulation.” Many other applications—such as restoring vision in blind people, treating chronic pain, and bypassing damaged parts of the spinal cord—are at various stages of development [14]. This progress includes restoring abilities to communicate. To give a sense of the current state of progress in line with BCI development in this case, in just the two year span between 2021 and 2023, paralysed patients have gone from being able to use BCIs to communicate at 15 words per minute to 78 words per minute, achieved by a volunteer collaborating with a team at the University of California San Francisco [15]. This and other therapeutic applications of BCIs evidently promise many benefits, though as with any medical intervention there are ethical issues to consider, including risk/benefit ratios, data privacy issues, and patient dependence on continued existence of the BCI company for support and maintenance.

In this Essay, we focus on the second application domain: BCIs for neurocognitive enhancement in people who are not (necessarily) suffering from any specific BCI-related medical condition [16]. Cognitive enhancements improve our cognitive functionality in ways that go beyond correcting pathology, making us “better than well” in some way. Such applications may involve both the enhancement of existing capacities (e.g., memory and attention) as well as the potential creation of new capacities [17,18].

Although BCIs for enhancement are not generally immediate goals for BCI developers, prominent actors such as Neuralink and Synchron explicitly highlight this possibility as a medium- to long-term objective. For example, Neuralink’s mission statement is to “restore autonomy to those with unmet medical needs today, and to unlock human potential tomorrow.” The cyborg future envisaged by Neuralink and others raises ethical issues far more complex than those raised by current therapeutic medical applications (see, e.g., [19]). Here, we consider a range of these issues, as well as some potential responses that we contextualise within wider debates about the boundaries of the mind. We start by setting out some constraints on the development of enhancement BCIs (eBCIs), as well as the potential they hold. We then examine a range of ethical issues attending eBCIs, including privacy, inequality, standardisation of thought, inauthenticity, and cheapened achievements Altogether, we hope to provide a perspective on eBCIs that cautiously anticipates ways that embracing them would potentially improve, but perhaps also worsen, our lives.

Current potential and limitations of BCIs

Although the basic principles of all BCIs are straightforward, and although recent progress has been impressive, there are many constraints that will likely limit and shape the trajectory of this technology, especially regarding uptake for enhancement purposes. These constraints can be divided into two (overlapping) categories: engineering and scientific.

All forms of BCI face many engineering challenges. Current invasive BCI probes record from only a tiny fraction of neurons. The current state of the art lies at around 1,000 electrodes, each typically detecting only a few individual neurons after software-based spike sorting. This is a tiny sample of the total cortical neuronal population (about 16 billion in adult humans). Current ECoG grids are also limited to a few hundred surface electrodes [15]. Advances in engineering and materials science are promising major improvements in electrode number and density, potentially reaching many thousands for both depth electrode and ECoG approaches [20]. For depth electrodes, there is also the potential to implant multiple locations, widening coverage [21]. But even with these advances, only a tiny fraction of the brain’s neurons will be sampled.

Some other engineering challenges pertain specifically to implanted electrodes, which have a tendency to detach, and to move around after implantation, potentially damaging surrounding neural tissue. Preventing inflammation and other damaging immune responses to BCI implantation is also difficult, as is maintaining long-term power and functionality. Again, new approaches may alleviate some of these difficulties, for instance through the development of highly flexible probes and conformal ECoG grids [20].

Scientifically, the challenges are even greater. It barely needs saying that understanding how the brain works is still a scientific project in relative infancy. But doing so does serve to draw a contrast with other human accomplishments, such as space flight and particle accelerator design. These accomplishments posed enormous engineering challenges, but were based on solid, well understood, and sufficient scientific foundations.

It seems very likely that the enhancement potential of BCIs will depend on a deeper understanding of how to read from, and stimulate, the relevant parts of (and patterns of activity within) the brain. This knowledge will itself depend on a deeper understanding of the fiendishly complex neural circuitry involved in perceptual, cognitive, and motor processes. This constraint goes hand-in-hand with the engineering constraints having to do with limited access to this neural circuitry. It has become increasingly apparent that neurocognitive processes are difficult to localise to a single brain region or neuronal assembly. Also, most or all such processes involve not just neuronal activity, but encompass a broad range of neurophysiological factors, including chemical neurotransmitters and activity of non-neural brain cells such as glial cells. In the other direction, an influential literature in the philosophy of mind and cognitive science—the “extended mind”—emphasises that cognitive abilities extend beyond the brain to include the body and parts of the external environment, including tools such as notebooks and mobile phones [22–24]. This means that there is unlikely to be any simple correspondence linking a desired enhancement goal (e.g., improved memory) and modulation of some single neuronal target (e.g., neurons in the hippocampus).

Some mitigating factors are worth noting. First, BCIs themselves may help address the scientific challenges, by allowing greater precision when recording and stimulating neural populations, and through testing of specific causal hypotheses using neurofeedback. Some BCI technologies, such as those leveraging optogenetics in nonhuman animals, are particularly promising in this regard, but—thanks to their technical constraints—are currently out of range for human applications [25].

Second, perhaps the scientific challenges can be to some extent evaded. In the outside-in direction, machine learning algorithms can allow powerful classification and prediction given sufficient data without any (or very much) understanding of the processes that generated these data. (Sufficient data, in general, means a very large amount of data indeed.) In the outside-in direction, the human brain seems to be remarkably adaptable to the relatively uniformed and unnaturally structured inputs that current BCIs make use of. To some extent—especially for perceptual systems—the brain seems to be “plug and play” [26].

These mitigating factors remain just that. They may soften but they do not get close to eliminating the relevant scientific challenges in understanding precisely how, where, and when to intervene in, and record from, specific neuronal populations in order to enhance existing cognitive capabilities or install new ones. This is why near-term eBCI applications may be limited to controlling apps on phones or other similarly prosaic activities. This may well be game-changing for people with medical conditions such as paralysis, but wider uptake may be limited.

In spite of these considerable challenges, BCIs still hold promise for enhancement applications. To better understand what might be possible, we now turn to the enhancement potential of BCIs, exploring how they could in principle, and with further developments, augment human capabilities.

Enhancement potential of BCIs

How might eBCIs, aimed not at alleviating neurological disease or pathology, but rather human enhancement, “unlock human potential”? While this question occupies its own emerging research area in bioethics [19], a useful framing for this question comes from the “extended mind” setting introduced earlier [23]. eBCIs provide many new ways in which extended cognitive processes can be implemented. They offer an immediacy and directness that seems—at least in principle—substantially different from that offered by other extended mind situations, such as those involving notebooks and phones.

The enhancement potential of eBCIs applies both to individuals, and to society at large. Consider what may be possible if—for the moment—we set practicality and feasibility to one side and give imagination free reign. One can imagine eBCIs that enhance memory, by providing instant thought-based access to both autobiographical and semantic information stores. eBCIs might improve attentional focus by directly modulating neural circuits involved in sustaining attention (see [27] for a clinical example in children with ADHD). Relatedly, eBCIs might also be able to modulate mood, in order to allow people to function more effectively in various demanding situations [28,29]. Perhaps eBCIs could allow people a wider bandwidth of control and communication by transcending the limits of having only 1 mouth and 2 hands. More generally, eBCIs might bring about substantial epistemic benefits, by dramatically changing capacities for knowledge access and discovery. Language barriers might become a thing of the past, once eBCIs can implement direct thought-based translation. Even more provocatively, could eBCIs increase the “bit rate” of human cognition, as the Neuralink founder has suggested [30]?

Other examples arise when we widen the lens. eBCIs might be able enhance our physical prowess as well as our cognitive capability, perhaps by improving reaction times or sensory sensitivity. And why not use eBCIs to endow us with entirely new sensory and perceptual modalities, by stimulating the brain with signals derived from things like sonar, compass direction, or even current stock prices [31]? Consider the potential enhancement benefit of being able, as technology advances (though for challenges here see [32]), to control additional robotic limbs, when two hands are not enough. Or consider, for example, being able to decode a stream of thought about one topic, while simultaneously holding a conversation about another. What about being able to dramatically accelerate your rate of thinking, much as you might speed up the playback rate for an audiobook?

Then there are wider benefits that may accrue to society at large. If many people augment their individual cognitive prowess, perhaps this could lead to more efficient and effective coordination and synthesis of information, leading in turn to enhanced capability to address societal challenges such as curing disease, mitigating climate change, and the like. eBCIs might even allow direct brain-to-brain interactions, so that the extended mind now becomes an extended collective mind, with additional capabilities that would be difficult to anticipate (see [33]). However, these more ambitious applications have their own unique challenges. Such scenarios are not just implausible by the scientific and engineering criteria discussed earlier; they are psychologically and philosophically implausible too.

We have already noted that the neurocognitive architecture of a human being is deeply intertwined with its embodied and embedded existence. Whereas implementing a novel perceptual modality may be relatively “plug and play,” implementing the ability to hold two simultaneous conversations, for example, would likely require a wholesale reorganisation of our entire neurocognitive architecture, and likely the body as well. And the pacing of cognitive activity—our “speed of thought”—may have difficult-to-overcome upper bounds, no matter what the bandwidth limitations are imposed on brain–body–world interactions [34]. What’s more, embodied aspects of cognition are not necessarily limitations but can also usefully considered as part of the generative process itself. How do I know what I think until I see what I say?

These feasibility constraints substantially undermine the likelihood of a future in which we all pop along to our high-street neurosurgeon, install the latest eBCI, and emerge newly superintelligent—and with a new hole in the head. It is true that if one allows imagination free reign, there seems to be little limit on how eBCIs could transform our lives and our society. And if our imagination were also enhanced by eBCIs, then what is unimaginable now might become imaginable in the future. But this kind of untrammelled neuro-utopianism is useful only to sketch the space of possible futures. Our imagination ought not be given entirely free reign, but—to be useful—should be reined in by what is, and what is likely to be, practical, feasible, and ethical. With this in mind, we now turn to the ethical considerations.

Ethical concerns raised by eBCIs

We have now unpacked some of the main ways in which eBCIs have the potential to enhance us, and we have highlighted various hurdles—engineering, scientific, and psychological. The present section takes us into philosophical territory: on the assumption that we can create eBCIs that satisfactorily address practicality/feasibly issues, should we? Do any considerations count against a “full steam ahead” approach to eBCI development, if this development became practically feasible? In this section, we canvass six key philosophical challenges that any affirmative answers will need to grapple with (see Box 1). In addressing these challenges, we do not give imagination entirely free reign, but instead focus on what might be possible given conceivable practical advances. We do this because the ethical issues can be made more vivid by focusing on eBCIs more advanced than are presently feasible. That said, the issues we identify are all at least in principle applicable to all eBCIs.

Conclusions and future perspectives

The balance between benefit and risk in eBCIs suggests the need for a middle way, somewhere between accelerationist/techno-optimist unrestricted development and outlawing nonmedical applications entirely. There is also the pragmatic issue that separating medical from enhancement applications will likely be very difficult in practice. Relatedly, the more obviously justifiable benefits of medical applications may put pressure on any attempt to introduce any form of regulation for eBCIs; people being offered relief from paralysis may quite understandably think that pontificating about potential mental monocultures to be irrelevant and obstructive. But what might this middle ground between techno-optimism and bioconservatism look like? A full discussion on this point is beyond the scope of what we can do here. However, it is worth emphasising that some efforts to define this landscape are already underway (Box 2).

In this Essay, we have focused on invasive BCI technologies. While potentially transformational, these technologies are less likely to see near-term widespread deployment than wearable BCIs based on EEG and other noninvasive imaging methods. There is already a vibrant consumer market for wearable BCI systems. As noted earlier, these technologies are typically limited to recording (inside-out) and have lower precision than invasive methods. Nonetheless, some of the ethical concerns we have raised—including erosion of privacy, uniformity of thought, cheapened achievements, and loss/gain of autonomy—apply to noninvasive eBCIs too. The noninvasive combination of ease of access, but lower precision and emphasis on recording, means that the overall ethical picture will look rather different [38].

One further issue to raise in closing is the apparent significance of the boundary of the skull. This boundary is crossed by all BCIs, though for invasive systems it is crossed in a more profound and permanent way. It is the crossing of this boundary that seems to be the most distinguishing features of BCIs—whether for enhancement or otherwise. After all, modern (non-BCI) technologies have led to all kinds of vulnerabilities in both of the directions we have identified: privacy (inside-out) and manipulation (outside-in). Social media, for example, both reveals our personal data and exposes us to manipulative pressures. BCIs take these same concerns and import them more directly into the brain. But why is this boundary significant? It does seem to have a prima facie lay significance. People may think that this boundary is important in the same way they might think that the integrity of one’s genome is important. A certain degree of squeamishness seems undeniable. But is there anything more substantive to say?

One possible line of thinking is that the boundary of the skull makes a difference because, broadly, it demarcates what we in principle have agency over. Recalling Schopenhauer [44], we can decide what to do, or say, but we cannot decide what to think. However, this line of thinking rapidly runs aground on the rocks of the debate over free will, and of the plentiful empirical evidence that we have less control over many over our actions than we might think [76,77]. The boundary is eroded in both directions. In the case of free will, the idea that the skull marks a meaningful divide behind which some sacrosanct conscious agent orchestrates behaviour is extremely difficult to maintain in the face of philosophical, theoretical, and empirical objections [64,76–78]. In the case of action control, many apparently voluntary actions unfold without accompanying experiences of intention or agency, and so have more in common with reflexive action than one might immediately think [79].

Nonetheless, a case can be made for the importance of the skull. This is simply that once one breaches this boundary, no further boundaries remain. The importance of the skull may therefore lie in preserving the idea of autonomy, rather than in preserving anything particular about autonomy or agency itself.

What’s clear from this discussion, and from other discussions of issues surrounding eBCI development, is that the questions raised are not limited to concerns about positive and negative individual and societal consequences. The merging of our neurobiology with technology raises fundamental questions about who, and what, we are—and who and what we can, and ought, to be.