What Does It Mean to Remain Human?

In 1953, a surgeon removed both medial temporal lobes from a young man with severe epilepsy. The seizures came under better control. Something else did not. For the rest of his life, the patient known in the literature as H.M. could not reliably build new memories of events. He could hold a thought while he attended to it and lose the thought the moment his attention moved. He also remained, to everyone around him, distinctly himself: the same temperament, the same dry humor, the same old stories, the same person other people could recognize across a room.

That case sits underneath the question this article takes up. If a person’s continuity depends on memory, and memory turns out to be a biological process that can break, then a technology that partly restores it is doing something far more intimate than replacing a hip. Yet H.M. is also the standing refutation of the simplest version of that worry. A man who could not lay down new memories had not stopped being a person. Something else was carrying the continuity: habits, character, relationships, a body that kept appearing in familiar places, a name that other people kept using.

So the question of what defines a human person when memory, perception, emotion, and reasoning are increasingly technologically supported resists a single-faculty answer. It has to be answered by asking which parts of a person are being supported, what kind of support the technology actually provides, and who holds authority over the result.

An old question with new hardware

Philosophers have been arguing about persons since before there was any hardware to argue about, and the argument has a useful shape. The Stanford Encyclopedia of Philosophy separates at least five questions that the phrase “personal identity” hides: what it is to be a person at all; what it takes for one person to persist through time; which properties make someone “the person they are” in the sense of an identity crisis; how we could ever find out who is who; and how many persons are present at a given moment (“Personal Identity,” Stanford Encyclopedia of Philosophy).

The third of those — what the encyclopedia calls the characterization question, following Marya Schechtman — is the one most people mean when they say “I would not be myself anymore.” It concerns the properties a person is attached to and takes to define her. The second question, persistence, is the one metaphysics cares about: it asks what makes an earlier and a later being numerically the same individual, not merely similar.

Technologies that reach into memory, mood, and attention press on both. They also expose a gap that the philosophical discussion has always contained: the difference between a person and the mechanisms a person runs on. Support a mechanism and you may change the person’s range of action without changing who she is. Change the content the mechanism carries and you may alter something more like authorship. The two cases look similar in a product demo and are different in almost every other respect.

Three answers, three things each one protects

There is no consensus view of personal identity, and pretending otherwise would make the governance questions that follow look easier than they are. Three families of answers do most of the work, and each one picks out a different thing that neurotechnology could threaten.

Psychological continuity

The psychological view descends from Locke, who described a person as “a thinking intelligent being, that has reason and reflection, and can consider itself as itself, the same thinking thing, in different times and places,” and who located identity in the continuity of consciousness, typically glossed as memory and connected mental life. In its contemporary form, associated with Sydney Shoemaker and Derek Parfit, a person at one time is the same as a person at another if the later one is uniquely psychologically continuous with the earlier one — a chain of overlapping connections among memories, intentions, beliefs, and character.

Parfit pushed this line to a conclusion many find uncomfortable. In Reasons and Persons (1984), he argued that what matters in survival is psychological continuity and connectedness — what he called Relation R — and that strict numerical identity is a further, and less important, fact. His thought experiments, including branching cases in which a person’s psychology continues along two paths at once, are designed to show that continuity can come apart from identity. Two successors can each be fully continuous with you while neither is you.

Notice what the psychological view protects: memory, deliberation, and the coherence of a life story. It also makes memory the most philosophically loaded faculty a device could touch. If memory carries identity, then a memory technology is never merely a memory technology.

Biological continuity

The main rival is the biological or animalist view, defended by Eric Olson, which holds that we are essentially human organisms and persist as long as the organism persists. Something can keep living through deep psychological disruption because biological continuity, not psychological content, does the work. Animalism plays a smaller role in neuroethics than it deserves, partly because it is unglamorous. It is also the view most at peace with the H.M. case: the man kept being the man, whatever his amnesia did to his mental connections.

This view is not immune to prosthetics. A hippocampal prosthesis that stimulates surviving tissue does not interrupt biological continuity; a hypothetical replacement of a brain region with manufactured components would test animalism much harder, because the question becomes whether the organism is still doing what organisms do. The distinction between augmenting an existing biological process and substituting for it is doing real work in that argument, and it is exactly the distinction that current clinical results do not yet cross.

Narrative and relational identity

A third family treats identity as something built in a story and sustained in relationships rather than located in a single faculty. On this account, personality change matters less because some inner essence altered and more because the change disrupts a narrative and the social recognition that goes with it.

The philosopher Françoise Baylis made this argument directly about deep brain stimulation (DBS) in Neuroethics, concluding that describing DBS as a threat to personal identity is “false, misdirected or trivially true” (Baylis, “‘I Am Who I Am’: On the Perceived Threats to Personal Identity from Deep Brain Stimulation,” Neuroethics, 2013). Her reasons are worth stating carefully, because they cut against the most common framing of the ethical problem. The claim is false, she argues, when it misunderstands how dynamic identity formation is; misdirected when aimed at a device rather than at the discriminatory attitudes that people with disabilities meet; and trivial when it treats any major life event as an identity threat. On a relational reading, the thing to protect is agency and the ability to keep narrating one’s own life, not an untouched interior.

What the clinical literature actually shows

Before generalizing from DBS to consumer neurotechnology, it helps to look at what the studies measured.

A 2024 cohort study in JAMA Network Open enrolled 52 pairs of patients with Parkinson’s disease and their care partners, tracking them before DBS and again 6 and 12 months afterward; two patients and one care partner were later lost to follow-up, leaving 49 complete patient records (Merner et al., JAMA Network Open, 2024). The design is the interesting part. Rather than administering a generic personality inventory, the researchers first asked each patient to name the three personal characteristics they most feared losing, then tracked those specific traits on visual analog scales. Ratings rose over time for both patients and care partners, reflecting what the authors describe as more positive manifestation of valued characteristics, and most individual scores stayed the same or increased. The patients evidently did not experience the treatment as being replaced by a stranger.

An earlier interview study with a different population complicates the picture in a more interesting direction. In PLOS ONE, Sanneke de Haan and colleagues analyzed interviews with 18 patients receiving DBS for obsessive-compulsive disorder and reported that many described themselves as becoming more themselves, not less (de Haan et al., PLOS ONE, 2017). The authors use this to argue that the philosophical vocabulary of “personality change” is too coarse to describe what patients actually go through, since a person freed from disabling compulsions has not lost an attribute so much as recovered access to one.

Both results sit inside a literature that is thinner than public discussion assumes. A systematic review by Frederic Gilbert, J. N. M. Viaña, and C. Ineichen examined more than 1,500 articles and found only 64 that reported first-hand empirical data on changes to personality, identity, agency, authenticity, autonomy, or self — while the theoretical commentary citing those changes was abundant (Gilbert, Viaña & Ineichen, “Deflating the ‘DBS causes personality changes’ bubble,” Neuroethics). Their conclusion is not that nothing happens. It is that a confident narrative about devices eroding selves has been running well ahead of the measured evidence, and that better instruments for detecting such changes are needed before anyone can say how often they occur.

That is the honest state of play: a small number of careful studies of how patients and partners describe themselves, a substantial theoretical literature built on a thin empirical base, and no validated way of measuring “the self” that everyone accepts.

Memory as the sharpest test case

Memory is where the philosophical stake is highest, so it is worth asking what the most ambitious memory work actually accomplished.

In 2018, a Wake Forest and University of Southern California team reported results from DARPA’s Restoring Active Memory program. Volunteers were epilepsy patients who already had electrodes implanted for seizure localization. During a memory task, the researchers recorded firing in hippocampal subregions, built an individualized nonlinear model of how input patterns in one region transform into output patterns in another, and then, on randomly chosen trials, delivered stimulation derived from that model (Hampson et al., “Developing a hippocampal neural prosthetic to facilitate human memory encoding and recall,” Journal of Neural Engineering, 2018).

Stimulated trials showed a 37 percent improvement in short-term memory performance and a 35 percent improvement on a later recognition test. Twenty-two subjects enrolled; eight received stimulation. DARPA’s own account of the work notes that the codes “did not directly reference specific memories” but instead reproduced the firing patterns associated with correct encoding, and that the stimulation supplemented natural activity rather than replacing it (DARPA, “Progress in Quest to Develop a Human Memory Prosthesis,” 2018).

Read precisely, this is a result about the capacity to encode, not about the content of what is encoded, and it was demonstrated in a small group under clinical conditions with electrodes already in place. The philosophical worry about memory prostheses usually concerns a different scenario: implanting or rewriting memories that were never experienced. Nothing here does that. The engineering distance between improving how efficiently a person’s own hippocampus writes a record and writing a record on the person’s behalf is the distance between a hearing aid and a voice in your head telling you what you said last week.

Even the capacity-side result has identity implications, and they are mostly permissive. If a device helps someone retain what she would otherwise lose, it is extending the reach of the psychological continuity the Lockeans care about, in the same spirit as eyeglasses for perception. The harder questions arrive when the record itself can be authored or edited, or when access to it depends on a service that can suspend it.

Where a mind ends and its toolkit begins

Long before implants, philosophers noticed that thinking already spills out of the skull. Andy Clark and David Chalmers argued in 1998 that when a person reliably couples with an external resource — their central example is Otto, who has Alzheimer’s and carries a notebook that functions as his memory — the resource can count as part of the cognitive system rather than as a mere tool (Clark & Chalmers, “The Extended Mind,” Analysis, 1998). Their parity principle asks whether a process would be counted as cognitive if it happened inside the head; if the answer is yes, its location outside the head should not by itself disqualify it.

That argument is doing more work in contemporary life than in the 1998 paper. Search engines, notes applications, and language models now hold material a person would once have memorized, and the coupling is often reliable enough to be trusted without inspection. If the extended-mind thesis is right even roughly, then a substantial part of what any of us calls “my thinking” has been running on rented infrastructure for years.

The thesis also opens the strongest objection to itself. Coupling that is reliable when a company keeps the server running is reliable in a different sense than coupling that travels with the body. An external component can be revised without notice, monetized, subpoenaed, or discontinued, and a person whose cognitive system includes it inherits those dependencies. This is a practical concern about ownership and continuity rather than a metaphysical objection, and it points somewhere specific: the properties that matter for extended cognition are the same properties that matter for infrastructure — portability, provenance, and the ability to leave.

Practical answers follow from that. Records of one’s own thinking should be exportable in ordinary formats. A system that supplies memory or perception should disclose what it added, when, and on whose instruction. When a tool is doing part of the work, the person using it should still be able to do the work slowly and badly without it, which keeps the underlying skill from quietly disappearing.

Dignity that does not scale with capability

There is a temptation, once augmentation becomes plausible, to treat human worth as something like an engineering specification that devices could raise or lower. The philosophical and religious traditions mostly reject that move, for reasons that hold up under scrutiny.

In the Christian tradition, human dignity rests on the claim in Genesis that human beings are made in the image of God, a status conferred rather than achieved, and therefore unaffected by changes in capacity. On a Buddhist analysis, the difficulty runs the other way: the self that people believe they are protecting is itself a construction, a bundle of processes without a permanent owner, so the anxiety about preserving a fixed essence rests on a mistaken premise. Neither claim is a scientific hypothesis, and neither can be confirmed or refuted by a laboratory result. They are frameworks for deciding what matters, and both happen to produce the same practical conclusion: dignity does not rise and fall with memory performance, mood stability, or processing speed.

Secular accounts can reach a similar place by other roads. Kantian respect attaches to rational agency rather than to the amount of it, and disability scholarship has long argued that the value of a life is not a function of the abilities it displays. Baylis’s point about discrimination lands here as well: when a person’s identity is described as threatened by a device that restores mobility, part of what is doing the describing is a discomfort with disabled bodies.

This matters for neurotechnology because the alternative — a sliding scale of humanity indexed to capability — would make every assistive device a partial demotion and every enhancement a claim to superiority. It would also make the ethics of the field impossible to state, since there would be no fixed subject whose interests were being weighed.

None of this makes the technologies benign. It relocates the concern. The problem with a mood-altering system is not that it makes someone less human; it is that it may make someone less the author of her own life while insisting that she consented.

Governance for a question that does not settle

If philosophers cannot agree on what a person is, a policy document cannot be asked to settle it either. What policy can do is protect the conditions under which any of the competing answers remain live.

The most concrete international instrument so far is the OECD’s 2019 Recommendation on Responsible Innovation in Neurotechnology, adopted on 11 December 2019 and described by the organization as the first international standard in this domain (OECD, Recommendation of the Council on Responsible Innovation in Neurotechnology, OECD/LEGAL/0457). It sets out nine principles covering responsible innovation, safety assessment, inclusivity, scientific collaboration, societal deliberation, oversight capacity, safeguarding of brain data, stewardship and trust, and anticipation of misuse. Two of its definitions are worth quoting because they encode a position on the question this article is about. It defines autonomy as the freedom to make one’s own choices, and cognitive liberty as the right to mental self-determination. It also defines personal brain data as data related to the structure or function of an identified or identifiable person’s brain, including information about physiology, health, or mental state.

That is a thinner commitment than it sounds. A right to mental self-determination does not specify what the self is; it specifies that whoever the person turns out to be, decisions about her mental life belong to her. The same logic runs through the UNESCO Recommendation on the Ethics of Neurotechnology adopted in November 2025 and UN Human Rights Council Resolution 58/6 earlier that year, which frame neural data and mental states as protected interests without adjudicating the metaphysics.

For engineers and product teams, the unsettled question translates into a list of testable properties. A system that touches memory, perception, mood, or reasoning should be able to answer, in plain language:

What did it change, and how do I see the change in my own record?
Can I reverse it, and what happens to me if the provider disappears?
Who else can read the data it produces, and under what legal process?
If it altered my preferences, was that alteration disclosed before I formed the preference?
Can I still do this unaided, at reduced quality, without penalty?

These are not metaphysical tests. They are the operational conditions under which the various theories of personhood can continue to have something to describe.

What would count as evidence

Because the field is prone to confident extrapolation, it is worth naming the boundary between what has been demonstrated and what has been imagined.

Demonstrated, in controlled clinical settings with implanted hardware and consenting patients: closed-loop stimulation can detect a pathological brain state and intervene selectively; mood and recovery state can be decoded from intracranial recordings in individual patients with custom models; patterned stimulation derived from a person’s own hippocampal activity can improve encoding performance by roughly a third in a small group; and patients with Parkinson’s disease and OCD largely describe themselves as intact or improved after stimulation.

Plausible engineering, on current trajectories but unbuilt: non-invasive deep modulation precise enough to be useful outside the clinic; portable, wearable memory aids reliable enough to count as part of a cognitive system; validated instruments that can detect self-related changes reliably enough to support consent conversations.

Speculation: memory editing that installs experiences never had; consumer devices that read complex emotional states from the outside; the transfer of a person’s psychology to another substrate. These may eventually be engineering problems. At present they are thought experiments, and the useful thing about thought experiments is that they clarify concepts, not that they predict products.

The single most informative study that could be run is also the least glamorous: a long-term, adequately powered, standardized comparison of self-related outcomes in patients before and after stimulation done at multiple centers, with instruments agreed in advance. Until something like that exists, the honest position on device-induced identity change is that it is possible, sometimes reported, weakly measured, and probably less common than the ethical literature implies.

Where this leaves the question

Remaining human turns out not to be a property that technology erodes at a fixed rate, and it is not located in any one faculty that a device could substitute. The traditions disagree about what carries a person through time — psychological connection, biological life, a narrative sustained with other people — and they disagree in ways that no implant will resolve. What they converge on is narrower and more useful: a person is the kind of thing that can hold a life together, be held accountable for it, and have it recognized by others.

That convergence suggests where to look when a new capability arrives. Assistance preserves or widens the range of a person’s own action, and the person remains the author of the result. Substitution narrows that range while making the result feel effortless, and authorship quietly moves to whoever set the objective. A memory aid that helps someone keep her own life is assistance. A system that decides which of her experiences to keep is substitution, whether or not she notices the difference.

For most of the coming decade, the interesting questions will not be about whether a machine can think. They will be about who holds the keys to the processes that make a particular human life continuous, what that person can still do when the keys are unavailable, and whether the answer to “who is this?” still begins with her own account of it.