Ethical and social aspects of neural prosthetics

Hearing loss is caused by destruction of sensory hair cells (Wilson and Dorman 2008, p 6). These cells 'no longer transduce acoustic energy into neural action potentials that are processed by the central nervous system and perceived as understandable speech' (Carlson 2020, p 1533). Children with congenital deafness and adolescents and adults with severe post-lingual hearing loss cannot benefit from hearing aids, Cochlear implants (CIs) allow the brain to process sound when this sensorineural function is severely impaired. This enables implant recipients to respond to auditory stimuli and communicate. CIs consist of electrodes and other components that bypass the inner ear and stimulate the auditory nerves to produce sound. By replacing or substituting for damaged of missing sensory hair cells in the cochlea, CIs provide an artificial form of hearing. Auditory training is necessary for CI recipients to learn to hear with them. Sound awareness and sound perception can improve when recipients control the device through their own brain signals.

Some researchers claim that restoration of function with CIs 'surpasses that of any other neural prosthesis' (Wilson et al 2011, p 117). Prosthetics that restore hearing have additional cognitive benefits. Among other forms of cognitive impairment, hearing loss in adults is a risk factor for dementia. "In the United States, approximately two-thirds of people 70 years or older have significant hearing loss" (Carlson 2020, p 1531). The ability of CIs to generate or restore some degree of hearing can improve cognitive functions and quality of life for people with severe hearing impairment.

A CI is not indicated when the auditory nerve is congenitally absent or has been destroyed. In people with these conditions, auditory brainstem implants (ABIs) or auditory midbrain implants (AMIs) may restore some auditory functions (Schierholz et al 2017). These implants deliver auditory stimulation to the inferior colliculus of the midbrain through surface and penetrating electrodes (Shannon 2012). Compared with CIs, ABIs sand AMIs have been implanted in a much smaller patient population. Outcomes of studies of the implants indicate that recipients of these implants do not develop hearing as well as CI recipients. Yet they appear to perform better on tasks involving visual and auditory functions (Schierholz et al 2017, p 2207).

Recent developments in auditory implant technology have improved sound processing. Implanted epidural recording electrodes using auditory evoked potentials in closed-loop systems that record and immediately respond to external sound in real time can improve hearing for auditory implant recipients (Haumann et al 2019). Yet sound distortion has been a persistent problem with these systems. Hearing generated from these devices may be degraded by ambient noise. This can disrupt music perception (O'Donoghue 2013, p 1191). Distortion may include sonic dimensions such as rhythm, pitch, tone, melody and timbre (Roy and Limb 2014). A new auditory prosthetic design consisting of a real-time audio source separation algorithm that can remix music may prevent sound distortion and improve music and other forms of sound perception in implant recipients (Tahmasebi et al 2020). This can improve the experience of artificially acquired or restored hearing.

The benefits of auditory implants for deaf children depend on the ability to use sign language before and after implantation. Implants alone may be limited in enabling children to hear and speak. One study has shown that learning to use and communicate with sign language from an early age can improve the ability to learn spoken language after cochlear implantation (Hassanzadeh 2012). Pre-implantation signing can enable young recipients to use the device for speech production. An implant cannot replace or substitute for the ability to use sign language. A CI implanted in a person who lacks the ability to sign can limit its positive effects on language and speech Other factors may account for variability of outcomes among CI recipients. The genetics of hearing loss, for example, can limit the ability of these devices to generate and maintain sound processing (Eppsteiner et al 2014).

There are social dimensions to auditory prosthetics that raise questions about fairness in access to them and how they affect recipients' sense of identity. These dimensions can influence whether or to what extent potential recipients of these devices can benefit from having them or be harmed by not having them. Implanting and monitoring an auditory prosthetic and speech therapy can cost approximately $10 000US. This can be prohibitively expensive for many deaf adults and parents of deaf children (Friedner et al 2019). Lack of access to such a prosthetic because of an inability to pay for it could prevent people from acquiring or recovering hearing. Those who could afford the prosthetic would have an unfair advantage over those who could not afford it because both groups would have the same type of impairment and medical need. Fairness is a measure of how a resource is distributed. A fair distribution gives equal weight to equal claims of need, and greater weight to greater claims of need (Rawls 1971, pp 111–114, Segall 2010, p 27ff). The need of the hearing impaired is something they have through no fault of their own but due to congenital or neurodegenerative conditions. This provides medical and moral grounds for a claim from them to receive a CI. Because economic inequality would influence who receives auditory prosthetics, factors other than need would determine access to them. This would make an unequal distribution of CIs an unfair distribution that would unjustifiably deny implants to those who could benefit from them (Daniels 2008, chapter 3). Unfairness and injustice in meeting medical needs are greater in developing countries with extreme economic inequality. Because a significant number of people globally are deaf, in principle health care systems have a moral obligation to make CIs or other auditory prosthetics available to as many deaf people as possible. This could raise them to the same or a similar level of functional capability and well-being as those with normal hearing. 'In principle' points to the practical reality that not all health care systems can afford to provide CIs or other devices for the hearing impaired. This point applies to all neural prosthetics and indicates that fairness cannot be divorced from economics. I return to this point in section 9.

There has been considerable debate on whether deafness is a disability or a different state from hearing. Some researchers have argued that, instead of describing deaf people as disabled, we should describe them as one group among 'multiple normals' with different sensory capacities (Friedner et al 2019, p 2382). Auditory function is one sensory function among others along a neuropsychological spectrum. Despite the inability to hear, deaf people who can use sign language can have similar lifetime opportunities and fulfilling lives as those who can hear. Deafness does not limit a child's or adult's opportunities. Rather, discriminatory political and social policies and inadequate public resources limit them (Wasserman and Asch 2014; Bickenbach et al 2014).

For some, CIs and other devices may seem a threat to deaf culture and the identity of deaf people. But there is variability among deaf people in how they perceive their condition. Not all CI recipients perceive the implant as a threat to their identity. They do not identify themselves as deaf and do not claim that artificial hearing substantially changes who they are. Author Karina Cotran received a CI at age 7 without previously learning a sign language. At age 21, she expressed her attitudes about living with an auditory prosthetic:

'I have a hearing impairment, but I am not part of the deaf community. I do not identify as being part of the deaf culture. I do not know sign language. Wearing a hearing aid and, later, a cochlear, allowed me the opportunity to become part of the hearing community. I was able to attend regular school and became immersed in the hearing culture. My hearing impairment placed me between two worlds. I always felt like I had one foot in each but was never fully part of one. At one point in my life, I refused to accept my hearing impairment as part of me. I did not want it to define me. Yet I felt like I should not have to hide it, either. Writing has helped me fuse those two worlds together and has helped me accept who I am' (Cotran 2017, preface).

Deaf parents may want their children to be deaf as well. But parents may not always act in the best interests of their children. 'The view that deafness is a disability supports the view that parents who can give their child hearing should do so' (Glover 2006, p 26). Even if they have similar lifetime opportunities as hearing children, deaf children face environmental risks. These include motor vehicles or bicycles that they cannot hear. In this respect, deafness is a disability. Failing to allow a deaf child to have a CI, AMI or ABI that would enable them to hear could harm the child by exposing them to these risks and in some sense denying them their right to an open future (Feinberg 1992).

Even when auditory prosthetics can reproduce sound without distortion, adults with these implants may have psychological challenges in adjusting to artificial hearing. They must transition from normal hearing to hearing loss and then to artificial hearing. The experience of hearing with a prosthetic may fail to meet their expectations when they receive an implant. This could cause psychological harm when hearing with the implant is phenomenologically different from and inferior to their natural hearing before they lost it. Restoration of sound perception can improve their ability to safely navigate the environment. Still, the extent to which auditory implant recipients benefit from them may depend on their pre-implant expectations, the degree to which the implant restores hearing, and how they adjust to it. Adjustment may depend on their experience of artificial hearing and how it compares with their experience of natural hearing. These considerations indicate that there is variability among recipients in benefit from auditory prosthetics.

Visual prosthetics enable people who are congenitally blind or blind from injury or disease to generate or restore some degree of vision. They can significantly improve their spatial orientation and navigation. However, 'it is unlikely that a prosthetic device can be produced that can create 20/20 or even 20/100 visual acuity. But then, even lower levels of acuity would be quite helpful for the blind' (Schiller and Tehovnik 2015, p 340). Visual prosthetics enable the blind to perceive visual patterns and process spatial information in three dimensions. Without this ability, 'it would not be possible to move in the environment' (Schiller and Tehovnik 2015, p 352).

Research in visual prosthetics has not advanced to the same level as research in auditory prosthetics because the visual system is larger and more complex than the auditory system. But advances in implant design have increased the functionality of visual prosthetics and useful vision for recipients. The most important of these advances has been an increase in the number of implanted microelectrode arrays generating phosphenes in the visual cortex and retina (Lewis and Rosenfeld 2016). There are two main types of visual prosthetics, cortical and retinal. In visual cortical prosthetics, a microelectrode array is implanted in the striate cortex, or area V1 of the primary visual cortex (Brant Fernandes et al 2012). Alternatively, it may be implanted in area V2 of the secondary visual cortex, or in the thalamus (Niketeghad and Pouratian 2019; Mills et al 2017). These are distinct from but project to areas V3 and V4. In retinal prosthetics, a microelectrode array is implanted in the subretinal or epiretinal surface. The array causes visual inputs to activate cells in the inner retina (Schiller and Tehovnik 2015, p 341). Retinal implants are designed to restore vision when photoreceptors have been lost from neurodegenerative diseases such as retinitis pigmentosa (RP) and age-related macular degeneration (AMD). Another type of implant stimulates the optic nerve, but it produces lower visual acuity than other visual prosthetics (p 341). The US Food and Drug Administration (FDA) approved the Argus II Retinal Prosthetic System for adults with RP in 2013. Results from a three-year study of this 'bionic eye' for people with this condition showed that it improved their visual function and quality of life (Ho et al 2015). The FDA granted approval for the first human feasibility study of the Orion Cortical Visual Prosthetic in 2017. These advanced designs are capable of generating or restoring a greater degree of vision in affected individuals.

Visual prosthetics do not completely replace normal vision and cannot generate and process visual information when there is extensive neural degeneration in the visual system (Fernandez 2018). Electrical stimulation of intact neurons in the visual cortex or retina from implanted arrays can enable the blind or severely visually impaired to perceive phosphenes. Stimulation enables recipients to map out a visual scene from them. Because phosphenes are not natural, prosthetic recipients must be trained to interpret and use them. What they see is not generated automatically by the visual system but artificially by electrical stimulation of preserved neurons within this system (Gabel 2017). Some recipients are more successful than others in learning to see with these prosthetics. This is one factor accounting for variable outcomes among recipients.

The benefit from visual cortical and retinal implants depends on neurobiological and cognitive factors. A recipient must have a certain number of preserved cortical or retinal neurons for the implants to generate phosphenes. One reason why electrical stimulation of area V1 may generate phosphenes and visual images is that neurons in this area 'remain largely unaffected for many years after sight is lost' (Schiller and Tehovnik 2015, p 347). This could facilitate activating these neurons from stimulation and be more likely to generate images than in retinal neurons, which tend to degenerate to a greater degree. In addition, implanted microelectrode arrays must be biocompatible with surrounding neural issue to maintain vision. Even if they are initially biocompatible, 'implanted electrodes can affect brain tissue adversely over time' (Schiller and Tehovnik 2015, p 353). This would cancel any restorative effect of a prosthetic.

When a visual prosthetic enables a congenitally blind person to see or restores vision in a person who became blind, it enables safer and more effective navigation of the natural and social environment. It can reduce the risk of self-harm from an inability to see. The psychology of receiving an implant may be very different depending on the recipient's stage of life and how well they transition neurologically and psychologically from blindness to sight. Those who are congenitally blind have to make one transition: from blindness to acquired artificial vision. Those who become blind later in life have to make two transitions: from normal vision to blindness, and then from blindness to restored artificial vision. Two transitions may be more psychologically challenging for adults receiving and learning to use the implant than for children or adolescents. It can be especially challenging if the experience of seeing with a visual prosthetic does not compare favorably with their memory of what it was like to see before they became blind. Depending on their cognitive and emotional capacities and their experiential memory, there may be significant differences in how adults with visual prosthetics respond and adjust to them.

The case of a patient described by Richard Gregory illustrates how restoring vision with a prosthetic may not always result in a net benefit for every recipient. Although this case involved corneal transplants performed years ago, a recipient of a visual prosthetic today could have a similar experience. The patient, who had been blind from a very early age, had a successful corneal transplant in 1959 when he was 52. The procedure enabled him to identify and distinguish objects, and his mobility improved. Initially, he seemed to adjust well to restored vision. However, 'soon he became curiously dispirited' (Gregory 1997, p 154). 'He "found the world drab"; and was aware of "the imperfections in things ... expecting a more perfect world"' (p 154). He 'gradually gave up active living, and three years later he died' (p 154). Gregory notes that, 'though the operation was a success, his story ended tragically' (p 154).

While it may be the exception rather than the rule, this case shows that personal expectations about one's visual experience with a prosthetic can affect the extent to which a recipient benefits from, or is harmed by, it. Variability in outcomes may depend on psychological factors as much as neurobiological ones. Advances in prosthetic design will continue to improve visual acuity, color and depth perception. But restored vision may not always result in a significant improvement in a prosthetic recipient's quality of life.

Novel retinal prosthetics could significantly improve useful vision. One group of researchers has commented that 'a combined approach that includes new biomaterials and novel neurostimulation approaches may ultimately lead to devices with greater integration with the remaining functional pathways of the visual system that can allow new stimulation paradigms to better replicate the natural neural code of the retina' (Barriga-Rivera et al 2017, p 4). These comments also apply to novel prosthetics in V1 or V2 areas of the visual cortex to replicate the neural code in them. A new retinal prosthetic uses nanoparticles injected subretinally in a rat model of retinal dystrophy (Maya-Vetencourt et al 2020). Light-evoked stimulation of retinal neurons resulted in improved visual function in this model. In addition, viral vector-based optogenetics has been tested in human trials for RP. In 2021, optogenetic therapy resulted in partial recovery of vision in a 58 year-old male with this condition diagnosed 40 years earlier. Intraocular injection of an adeno-associated viral vector encoding ChrimsonR combined with light stimulation through engineered goggles activated retinal ganglion cells. The patient was able to perceive, locate, count and touch different objects (Sahel et al 2021).

The extent to which this and other patients benefit from this therapy will depend on how they adjust to the change from blindness to partial vision. This may be influenced by psychological and social factors as much as the therapy itself. Optogenetics currently lacks the level of light sensitivity for the degree of neural activation necessary for complete retinal restoration (Schiller and Tehovnik 2015, pp 15–18, Yun and Kwok 2017, Gaub et al 2018). Opsin gene therapy has considerable therapeutic potential for humans with RP and may have this same potential for individuals with blindness associated with damage or degeneration in the visual cortex. However, it may be some time before this potential is fully realized.

In both auditory and visual prosthetics, the extent to which they benefit recipients and improve their well-being depends on bioengineering and its effects on sensorineural function. It also depends on their experience of artificial hearing and vision and how they respond and adjust to this experience. These factors in turn are influenced by the natural and social environment in which they hear and see.

DBS involves electrical stimulation of dysfunctional neural circuits implicated in disorders of movement, mood and cognition. It modulates these circuits to restore varying degrees of motor, affective and cognitive functions. DBS is approved therapy for Parkinson's disease, essential tremor and dystonia (Hariz et al 2013, Lozano et al 2019, Krauss et al 2021). This technique has also been tested in clinical trials for treatment-refractory obsessive-compulsive disorder (OCD), major depressive disorder (MDD), anorexia nervosa, schizophrenia, posttraumatic stress disorder (PTSD) and other psychiatric disorders. Although the US Food and Drug Administration (FDA) granted a Humanitarian Device Exemption (HDE) for the use of DBS to treat OCD in 2009, the technique remains an experimental and investigational intervention in psychiatry (Hamani et al 2016).

In a DBS system, electrodes are implanted unilaterally or bilaterally in a targeted brain region. The electrodes are at the end of leads connected through an extension to an implantable pulse generator located under the collarbone or in the abdomen. The pulse generator delivers current to the electrodes at different frequencies and amplitudes, which can be modified (Wang et al 2018). It can be turned on and off with a programmable hand-held device like a smartphone. Higher or lower frequency is delivered to the brain depending on the type and degree of circuit dysregulation.

Implanting electrodes in the brain entails a 1%–5% risk of intracranial hemorrhage, as well as the risk of stroke and infection (Krauss et al 2021). A technical limitation of these systems has been battery depletion in the pulse generator, which requires replacement. This limitation has been largely resolved through the use of rechargeable implants (Hitti et al 2019, Jakobs et al 2020). Automatic recharging could also decrease the likelihood of lead impedance and fracture and ensure efficient and sustained operation of the system. This technical improvement has resulted in greater functional improvement and satisfaction in patients with DBS systems implanted in them.

The mechanisms of DBS are not completely understood. Four hypotheses have been proposed for its neuromodulating effects (Lee et al 2019, Krauss et al 2021). An inhibitory hypothesis is that stimulation inhibits hyperactive neural processes. It inhibits oscillatory activity in a given circuit, such as the dysfunctional frontostriatal circuit implicated in OCD. An excitatory hypothesis is that DBS causes direct excitation of neural activity in a circuit, such as the one involving a hypoactive nucleus accumbens (NAc) in anhedonia. A disrupting hypothesis is that electrical stimulation disrupts the flow of information in a particular region of the brain, A filtering hypothesis is that stimulation filters low-frequency signals across synapses and axons. The basic idea behind these four hypotheses is that electrical stimulation of the brain can modulate neural processes so that they are neither overactive nor underactive and do not result in different pathologies. Specificity in targeting dysregulated circuits at the right frequency, amplitude and duration is necessary for DBS to have neuromodulating effects.

Closed-loop systems have improved outcomes in neuromodulation. Most DBS systems have used open-loop devices (OLDs) (Grahn et al 2014, Potter et al 2014). They deliver constant electrical stimulation without adjustments in voltage and frequency. They do not respond to changes in the brain because there is no feedback to the device from brain activity. Stimulation parameters are programmed into the device and remain unchanged until the next programming session. Among other limitations, any adverse neurological effects from stimulation may not be known until the patient reports them.

Closed loop devices (CLDs) respond in real time to changes in the brain, 'Closed loop DBS systems simultaneously record and stimulate neural activity, allowing the stimulation to be adjusted according to disease-specific neural biomarkers' (Aggarwal and Chugh 2020; Krauss et al 2021, p 75). CLDs are thus more likely to restore and maintain normal neural oscillation and signalling patterns in dysregulated neural circuits than OLDs. In closed-loop systems, stimulation is activated when they detect a dysregulated physiological signal or oscillation pattern in the brain. In this regard, they are a form of adaptive neuromodulation. Their ability to both record and respond to brain dysfunction makes CLDs superior to OLDs in their neuromodulating ability. The signals may include 'the occurrence of an electroencephalographic abnormality in patients with epilepsy or the increase in oscillatory activity in patients with PD (Parkinson's disease)' (Lee et al 2019, p 339). In addition, DBS with directional leads allows more precise stimulation based on neural activation patterns in individual patients (Schuepbach et al 2017, Krauss et al 2021).

The Percept PC DBS system developed by Medtronic is another advance in personalized neuromodulation. In addition to the electrodes implanted in the brain, this system consists of a pacemaker-like device implanted subcutaneously in the chest. Connected to the electrodes through thin wires, the device continuously records signals from the targeted brain region. The Percept system can allow patients with PD or epilepsy to control tremors or seizures by adjusting stimulation settings in response to brain changes detected by the system. The ability to adjust the settings could not only allow patients to have more control of symptoms associated with brain diseases but also prevent sequelae from imprecise stimulation. These systems also allow physicians to remotely program them. This could obviate the need for patients to travel to clinics to change stimulation parameters. Such remote control could limit patients' ability to operate the device on their own, however. Accordingly, patients and physicians would have to agree on stimulation parameters to ensure optimal neuromodulation. There are more serious ethical concerns about remote control of brain implants, specifically brainjacking by unauthorized third parties. These concerns will increase as wired neuromodulating systems become wireless and the potential for remote control increases. I discuss them in section 8.

In the remainder of this section, I focus on the use of neurostimulation to restore motor and cognitive functions in patients with prolonged DOCs. Experimental use of this technique for these disorders has occurred on a smaller scale than its use for movement and mood disorders. But its therapeutic potential may be greater for patients who are neurologically compromised from traumatic or anoxic brain injury and have no other treatment options.

Prolonged DOCs are disorders of the level, degree or state of awareness (Giacino et al 2002, 2014). They result from traumatic brain injury (TBI) or anoxia of the brain and often progress from coma as the immediate effect of brain injury. The two main types of DOCs are the vegetative state (VS) and the minimally conscious state (MCS). In the VS, patients show arousal and have sleep-wake cycles but are unaware of themselves and their surroundings. The VS has also been described as unresponsive wakefulness syndrome (Laureys et al 2010). The general view among neurologists is that a persistent VS becomes a permanent VS 3 months after an anoxic or hypoxic brain injury or 12 months after a TBI. A permanent VS means that the patient has no chance of recovery to awareness.

Some patients in a persistent VS progress to a MCS. 'MCS is a condition of severely altered consciousness characterized by minimal but definite behavioral evidence of self or environmental awareness' (Giacino et al 2002, 2014, p 100). The MCS involves disruption of thalamocortical and corticocortical connections mediating awareness. Yet the fact that some of the connections are preserved despite brain injury explains why these patients have some level of awareness and capacity for cognitive processing. They have been described as being 'covertly aware' (Owen 2019). Emergence from the MCS is defined as 'the re-emergence of a functional communication system or restoration of the ability to use objects in a function manner' (Giacino et al 2014, p 101).

DBS was used in an FDA-approved clinical trial conducted in 2006–2007 on a patient who had been in an MCS for 6 years after an assault. Stimulation of the central thalamus resulted in recovery of some motor and cognitive functions and behavioral improvement (Schiff et al 2007). There have been a number of clinical trials testing DBS for the VS and MCS in the last 14 years. The results have not shown significant functional recovery in most patients. For example, investigators in a 2016 trial concluded that thalamic stimulation 'did not induce persistent, clinically evident conscious behavior in the patients' (Magrassi et al 2016). In a more recent trial, 14 patients in the VS or MCS from TBI or hypoxic encephalopathy received stimulation of the left centromedian-parafasicular complex of the thalamus. Two of the fourteen patients regained consciousness and the ability to live independently. One patient regained consciousness but remains in a wheelchair. Another patient is able to follow simple commands. Seven patients remained unchanged, and three died from causes unrelated to the TBI (Chudy et al 2018).

A crucial factor in potential recovery from the MCS is the extent of axonal preservation or degeneration from brain injury. Patients with more preserved axonal connections would be more likely to respond to neurostimulation. The technique could induce neuroplastic and neurogenerative mechanisms to promote restoration of cognitive and motor functions. Diffusion tensor imaging (DTI) has been used in one study to accurately predict the extent of chronic degeneration from TBI (Graham et al 2020), This would allow researchers to exclude patients with diffuse axonal injury from clinical trials testing the safely and efficacy of DBS for these disorders. DTI and algorithmic EEG could identify biomarkers of intact axons or axonal injury and other signs of neural pathology in selecting patients for these trials (Shin et al 2014, p 1226). These biomarkers could identify patients more likely to recover motor and cognitive functions from neurostimulation and regain functional independence. One challenge in assessing outcomes of these trials is determining which patients would have recovered spontaneously and which would have recovered because of DBS (Eskandar 2018). Biomarkers alone would not decisively answer the question of why one or more patients experienced functional recovery. This is just one of many 'logistical and methodological difficulties of conducting placebo-controlled trials in this population' (Giacino et al 2014, p 107).

There is one recent noteworthy positive outcome of DBS for a patient with a DOC. In 2019, researchers reported that an implant delivering electrical stimulation to an area of the brain of a woman with a TBI from an automobile accident 18 years earlier restored near-normal levels of neural function. The technique restored many of her motor and cognitive functions, and she is now able to live independently (Thibaut et al 2019). More careful selection of patients with specific neural signatures associated with axonal, thalamocortical and corticocortical connectivity and more advanced neurostimulation tailored to these signatures may restore close to normal levels of cognitive and motor capacities and functional independence in more patients with DOCs. To date, though, the number of these patients who have significantly improved from DBS is small. The benefits of this technique to this group of patients do not appear to outweigh the risks. Nevertheless, neurostimulation offers these patients more hope than living indefinitely with little or no improvement in rehabilitation centers.

Placebo-controlled trials are the most scientifically robust and reliable way of assessing the safety and efficacy of DBS for prolonged DOCs. But they have ethical implications. Patients in the MCS who might be considered as research subjects based on imaging showing preserved neural function likely would not have the cognitive capacity to give informed consent to participate in a clinical trial (Beauchamp and Childress 2019, chapter 4, Emanuel et al 2008, part 8). Families or others could give proxy consent for an MCS patient to be included in a trial when this is in the patient's best interest and is what the patient would want if she were competent to make her own informed decisions. Yet some of these substitute decision-makers may be motivated by a therapeutic misconception (Mathews et al 2018). Rather than acknowledging that the primary purpose of DBS research for the VS and MCS is to generate scientific knowledge about the therapeutic potential of the technique for patients in these states, they may believe that the trial will directly benefit the patient. They may want and allow the patient to be included in research because of the lack of any therapeutic alternatives.

Knowing that the patient may be assigned to the placebo arm of the trial may make some families reluctant to give proxy consent for the patient to be enrolled as a research subject (Buchanan and Brock 1989). Also, understanding the design and goal of a clinical trial of DBS for DOCs may be difficult for families adjusting to having a loved one with a severe brain injury. Proxy consent after a patient has been neurologically compromised for many years may avoid this problem. But substitute decision makers may be motivated to give consent because of physical and emotional fatigue from caring for the patient for an extended period. It is unclear whether this would be in the patient's best interests. These ethical issues compound the challenges of conducting the research necessary to determine whether or to what extent DBS or other forms of neuromodulation can restore awareness and cognitive and motor functions in a significant number of patients with prolonged DOCs (Young et al 2021).

A BCI, or brain-machine interface (BMI), is 'a system that measures CNS activity and converts it into artificial output that replaces, restores, enhances, supplements or improves natural CNS output and thereby changes the ongoing interactions between the CNS and its external of internal environment' (Wolpaw and Wolpaw 2012a, p 3, Nicolelis 2003, Wolpaw et al 2020). BCIs bypass damaged neural circuity through real-time direct connections between the brain and a signal processing algorithm in a computer (Lebedev and Nicolelis 2006, Wolpaw and Wolpaw 2012a, Lebedev 2014).

BCIs consist of wired or wireless systems that record electrical signals in motor and adjacent area of the cerebral cortex. They transmit these signals as sensorimotor input to an external device in producing motor output. This includes such actions as moving a prosthetic limb, robotic arm, computer cursor, and selecting letters from a wheelchair-mounted computer tablet to communicate. BCIs enable people to overcome motor impairment from traumatic brain and spinal cord injury, stroke, limb loss and neurodegenerative disease. They can restore varying degrees of sensorimotor capacity and functional independence (Lebedev and Nicolelis 2017). Closed-loop BCIs are more likely to enable users to complete action plans and have greater control over motor output than open-loop systems because they allow sensory feedback not only from the computer to the brain but also from the brain and computer to the user.

There are three general types of BCIs. Non-invasive systems consist of scalp-based electrodes attached to a cap as part of the equipment required to record EEG signals. Because these systems do not involve intracranial surgery, they do not entail risks of hemorrhage, infection or inflammation in the brain. But their ability to directly record electrical signals from sensorimotor areas may be limited by cranial smearing. Newer high-resolution EEG systems may be able to access signals deep in the brain. In one study, subjects using this type of BCI displayed a significant degree of control over a robotic arm This technique can improve neural recording and motor and cognitive control over prosthetics or external devices such as wheelchairs (Edelman et al 2019).

Electrocorticography (ECoG) is an invasive form of BCI. In ECoG, electrode arrays are implanted epidurally or subdurally to record and monitor electrophysiological signals in the cerebral cortex (Leuthardt et al 2004, Schalk and Leuthardt 2011). They have higher spatial and temporal resolution than EEG and can decode motor cortical signals more directly than scalp-based electrodes because they are not susceptible to cranial smearing. But they entail the risk of hemorrhage, infection and inflammation from implantation. Like the EEG scalp-based cap system, epidural and subdural ECoG BCIs impose constraints on subjects' movements because of the wires running from the electrodes to the machine Newer wireless versions of ECoG could avoid these constraints.

The most invasive BCIs are fully implantable intracortical wireless systems consisting of microelectrode arrays in brain regions mediating sensorimotor functions (Wolpaw and Wolpaw 2012a, Lebedev 2014). They are less burdensome to users than EEG or ECoG systems. More importantly, they may be more effective in recording and transmitting electrical signals to the computer because the arrays are in the critical brain regions and more directly connected to the signals. Implanted arrays are thus more likely to facilitate the execution of the subject's intention to move a cursor, arm or limb in a desired action. These BCIs enable a subject to regain a greater degree of motor control and, in turn, functional independence and autonomous agency.

Bleeding, infection and inflammation are not the only risks associated with implanted microelectrode arrays. They may not always be biocompatible with surrounding neural tissue, which is critical for signal recording and transmission. Activation of electrodes may reorganize and induce changes in this tissue and neural circuits. These changes may promote neuroplasticity and endogenous repair and growth mechanisms that would result in increased axonal connectivity. But they could also be adverse and cause further neurodegeneration. A stable and effective array that would function for a long period without replacement would be one that integrated into the surrounding neuropil (Kennedy et al 2011). This would ensure accurate signal recording and transmission for desired BCI output.

One of the first successful applications of BCI technology was the BrainGate 2 neural interface. This system consisted of a 96-microeclectrode array implanted in the motor cortex of a volunteer with tetraplegia from a spinal cord injury. It enabled him to move a robotic arm and open and close a prosthetic hand with his thoughts mediated by the computer algorithm (Hochberg et al 2006). In a later study involving a volunteer with tetraplegia from a brainstem stroke, the system enabled him to use a robotic arm to grab a foam ball (Hochberg et al 2012). Experiments with similar systems have shown that the interface can allow activation of electromyographic signals by subjects to move prosthetic arms and legs (Collinger et al 2012, Gilja et al 2015). More advanced forms of this technology will enable people to have greater control over these prosthetics (Vilela and Hochberg 2020). But it remains an experimental technology, and patients using interfaces are in important respects research subjects.

In 2016, a 24 year-old quadriplegic used a BCI to bypass a spinal cord injury and move his right hand and fingers (Bouton et al 2016). In this first human experiment of limb reanimation, an array consisting of 96 microelectrodes was implanted in his motor cortex. By intending and trying to move his hand and fingers, the patient activated the electrodes and caused the transmission of motor cortical signals to produce the movements. Although he regained some degree of reach and grasp, he did not regain his sense of touch. But a more recent BCI experiment partly restored his touch by increasing bi-directional signalling between his central and peripheral nervous systems (Ganzer et al 2020).

BCIs connecting neural ensemble activity to a exoskeleton are a form of whole-body neural prosthetic designed to allow a greater degree of movement in tetraplegia (Lebedev and Nicolelis 2011). A recent proof-of-concept study of this technology is a major advance in translational BCI research from primates to humans, In this study, activation of a four-limb neuroprosthetic exoskeleton through a wireless ECoG BCI enabled an individual with tetraplegia from a cervical spine injury to make upper-limb movements and perform reach-and-touch tasks over a two-year period. Exoskeletal prosthetics have therapeutic potential for the 20% of cervical spinal injuries resulting in extensive paralysis (Benabid et al 2019).

Another significant development in BCI technology in restoring motor function has been the DEKA LUKE arm (George et al 2019). This is a bi-directional neuro-myoelectric prosthetic with biomimetic sensory feedback from the arm to the CNS. It has allowed some patients with amputated hands and arms to regain some degree of movement, grasp and touch. This and the other BCI applications I have mentioned can restore varying degrees of movement and associated sensory functions, increase the scope of agency and improve quality of life in recipients.

The neurological and cognitive ability of individuals to learn to use the interface and its effects on proprioceptive, somatosensory, interoceptive and exteroceptive processing and feedback can influence whether or to what extent they can benefit from or be harmed by BCI-mediated action. When a person with paralysis or limb loss uses a BCI to move a prosthetic limb or external object, the interface liberates the mind from a disabled body (Nicolelis 2011, Steinert et al 2019). Through the interface, the BCI user can perform intentional actions when she cannot move her body on her own. These are not disembodied actions but extended bodily actions mediated by the BCI. They are not alien to the subject but events she can identify with as she integrates the BCI into her body schema (Lebedev 2014, Gilbert et al 2017). Although the hardware, software and electrical mechanisms in implantable and implant-free BCIs are different in certain respects from those of DBS, both enable shared control between the user and the system. Both can restore some degree of agency limited by cognitive, affective, volitional and motor impairment (Drew 2019).

The ability to perform BCI-mediated actions requires a high level of operant conditioning, patience and sustained attention. Not all patients who might want to use a BCI to restore movement are candidates for it. Some with TBI or neurodegenerative disease may be too cognitively impaired to learn to use the interface to produce motor output through it. They may not even have the cognitive capacity to give informed consent to be trained to use it. A more important ethical issue is that they may be vulnerable to harm from unreasonable or unrealizable expectations of restoring motor control. The belief that using a BCI might restore some of this control may alleviate anxiety and depression experienced by some patients with paralysis or limb loss. Yet this belief and the desire to move a computer cursor, for example, may result in harm if the inability to use the BCI prevents the realization of that desire. This could exacerbate the feeling of loss of agency from one's condition.

In some cases, a user's failure to sustain attention to the task of moving a robotic arm might have a different outcome from failing to move it. Allowing one's mind to wander in the process of transmitting signals from the brain to the computer may result in an unintended action. This might include striking a trainer with the arm and causing serious bodily harm. This could be an example of negligent BCI use with legal implications (Bublitz et al 2019). A subject could also use to BCI to intentionally harm the trainer or another person, which would be an aggravating condition regarding criminal responsibility. Ethical and legal assessments of BCI use would depend on the degree of control the subject had in using the system.

All patients with severely reduced agency from paralysis or limb loss have the same need for motor restoration. But some may not have the cognitive, emotional and volitional capacities necessary to use a BCI. Training them to use the system may not be in their best interests if it failed to achieve their goals It would be fair for investigators to exclude them from training on these grounds, This would be consistent with investigators' obligation of nonmaleficence to prevent psychological harm to subjects from failing to successfully use them (Beauchamp and Childress 2019, chapter 5). Including only patients with the requisite capacities for operating a BCI and excluding those who lack them be a fair form of discrimination. Among patients deemed cognitively and emotionally capable of learning to operate the interface, some still may fail to generate intended motor output. As part of their ethical obligation to protect research subjects from harm, investigators informing subjects about using a BCI would be required to point out variability of outcomes with BCIs and ensure that they have reasonable expectations about their goals in using them (Glannon 2014, 2016).

BCIs can bypass damaged brain regions and activate preserved regions mediating language processing to enable behaviorally nonresponsive patients to communicate. Impairment or loss of this capacity from CNS injury or disease can entail significant functional limitations BCIs may overcome some of these limitations for patients with locked-in syndrome (LIS) or advanced amyotrophic lateral sclerosis (ALS). They may enable them to express attitudes about their condition or wishes about how they should be treated. Specifically, the interface may enable them to communicate momentous decisions about whether to continue or discontinue life-sustaining medical care. These systems enable communication by recording and decoding neural signals and converting them into letters and words as synthesized speech. Electrodes placed on the scalp, epidurally or subdurally, or intracortical microelectrode arrays record electrical signals in motor and language-processing areas (Saur et al 2008). They are then transmitted to a receiver and computer for language output.

BCIs utilizing electrodes on the scalp to record neural signals through EEG have enabled some locked-in patients to activate slow cortical potentials, sensorimotor rhythm and the P300 event-related potential and respond affirmatively or negatively to questions about their quality of life (Birbaumer et al 2008, 2014). These responses have had approximately 70% accuracy in assessing patients' attitudes about living with their condition. But BCIs are limited in allowing subjects to translate an intention to speak into a communicative act. This is mainly because the cranium can deflect the transmission of neural signals from sensorimotor areas to the computer, Epidural and subdural ECoG can avoid this problem. As noted, ECoG has higher spatial and temporal resolution than EEG . One group of researchers used an ECoG BCI in a dual-stage decoding process to reconstruct speech and produce spoken sentences (Anumanchipalli et al 2019). The system they used could allow fluid natural speech, which corresponds to 150 words per minute. Significantly, the subjects in this study were all capable of speaking. The critical test is whether this BCI could generate speech from verbally nonresponsive or compromised patients. Further studies will be necessary to demonstrate that it could have this effect.

In 2014, neurologists Leigh Hochberg and Merit Cudkowicz stated that, 'though both scalp-based EEG and corticography signals have been recorded in people with ALS and total LIS, we are aware of no reports of restoring communication using a neural signal-based BCI in this most severely affected population' (2014, p 1852). The authors appear to be referring to communication involving more than binary 'yes' or 'no' responses to questions. More robust language production would be necessary for these patients to express decisions about continuing or discontinuing life-sustaining treatment. Neural interfaces consisting of implantable microelectrode arrays may enable some patients to do this. If the implant were in cortical regions associated with language processing, then it could more directly record and transmit signals from these regions to the software for letter and word production. Like other brain implants, though, there would be a risk of failed biocompatibility between this implant and surrounding neural tissue. Even if it facilitated efficient recording and transmission of neural signals to the computer, it is unclear how long the implant would function and whether it would have to be replaced.

A study published in 2016 shows how implanted BCI system may allow verbally unresponsive patients to communicate to some degree. Investigators describe a 58 year-old ventilator-dependent patient locked-in from ALS who lost her ability to speak but had preserved cognitive functions (Vansteensel et al 2016, Vansteensel and Jarosiewicz 2020). She used a fully implanted BCI system consisting of a computer typing program to communicate by moving a cursor to select two letters per minute. Subdural electrodes were implanted over her sensorimotor and dorsolateral cortex. Electrical cortical signals passed through a transmitter implanted under the collarbone. An antenna sent the signals to a receiver and then to wheelchair-mounted computer tablet from which she selected the letters. 'The detection of signals from the cortex requires computational processing to separate ('decode') them from background noise. Decoding of sufficient quality provides the input for a computer system that directs typing software, thereby enabling communication' (Vansteensel et al 2016, p 2060).

The two letters per minute that this patient was able to produce is far below the level of 150 words per minute necessary for fluid communication. Limited word production could generate ambiguity about the patient's intentions and allow attending physicians and families to misinterpret them. In the case of a patient with advanced ALS, this could involve questions about initiating or terminating mechanical ventilation. It could result in continuing life-sustaining treatment that would conflict with a patient's wish to discontinue it. Or it could result in discontinuing treatment when the patient wanted to remain alive. The patient could be harmed in either of these respects. The ethical stakes in reliable BCI-mediated communication would be high. BCIs may advance to the point where patients with the necessary neurological, cognitive and emotional capacity could use them to coherently and reliably express attitudes and decisions about medical care to physicians and family members. These would be autonomous decisions made by patients rather than for them by others. Because they would be autonomous, they would allow them to have some control over their lives.

Not all ALS or LIS patients with preserved cognitive functions could learn to use a BCI for this purpose, A BCI with a wheelchair-mounted computer tablet might overcome the dysphagia, anarthria and aphonia from pseudobulbar palsy if the patient has intact consciousness and cognition (Bernat 2020, p 232). Yet advanced ALS may involve not only degeneration of motor neurons but also cognitive impairment and loss of rational and decisional capacity (Crockford et al 2018). This could preclude them from using a BCI. Still, some patients retain this capacity and could use a BCI to express wishes and decisions about actions by others that could benefit or harm them. A fully implanted BCI with direct recording of electrical signals in language areas could overcome the limitations noted by Hochberg and Cudkowicz and allow these neurologically compromised patients to make these decisions on their own without having to rely on substitute decision-making that may not be in their best interests.

It is also possible that some patients at the higher end of the MCS spectrum may retain enough cognitive capacity for BCI-mediated communication. Some commentators note that loss of the ability to communicate with family is the main source of suffering among minimally conscious patients (Fins 2015). Many neurologists and neuropsychologists have maintained that binary 'yes' or 'no' responses to questions confirmed by imaging and neurophysiological recording would not be sufficient to establish that a minimally conscious patient could make and express informed and deliberated decisions, especially regarding life-sustaining artificial nutrition and hydration (ANH). They have expressed doubts that such a patient would have the cognitive and emotional capacity to make such complex decisions (Fernandez-Espejo and Owen 2013, p 808). But BCIs may enable more robust communication that would include these decisions. Two recent studies suggest that this may occur sooner rather than later. One study used an intracortical BCI to record brainwaves associated with the thought of writing and translate them into computer-generated letters (Willett et al 2021). A second study used a subdural multielectrode array implanted over the sensorimotor cortex of a person with anarthria from a brainstem stroke to decode sentences from cortical activity in real time at a rate of 15.2 words per minute. The median word error rate was 25.6% (Moses et al 2021).

In 2019, Adrian Owen claimed that 'communication with entirely physically nonresponsive patients is still very much in its infancy. Yet, 20 years from now, so-called BCIs, will likely be as commonplace as smart phones, flatscreen TVs and mobile touchscreen devices' (Owen 2019, p 527). Some may be skeptical of this claim. But current and future BCIs with advanced recording and decoding of neural signals in language processing areas of the brain may enable verbally nonresponsive patients to reliably communicate decisions about actions that affect them and their lives. The interface would allow decisions about continuing or discontinuing life-sustaining care to be made 'by proclamation rather than by proxy' (Hochberg and Cudkowicz 2014, p 1853).