Ethical and Societal Challenges in Intelligence Enhancement
Linas JuozenasShare
Equity · Emerging technology · Responsible innovation
Who Gets to Benefit? Building Innovation That Is Fair, Safe and Shared
A breakthrough is not fully successful merely because it works in a laboratory or reaches the market. Its real value also depends on who can use it, whose needs shaped it, who carries its risks and whether people can question, refuse or seek remedy when it causes harm.
- Meaningful access
- Human rights
- Proportionate safeguards
- Public participation
The real question is not only “Can we build it?”
Emerging technologies can widen human possibility. They can also reproduce old exclusions at greater speed and scale. Responsible innovation asks what kind of progress is being created, for whom and under whose control.
Artificial intelligence can translate lessons, assist diagnosis and make complex tools easier to use. Gene and cell therapies can address biological causes of illnesses that were once treated only through continuing symptom management. Neurotechnology may restore communication or movement for some people with serious impairments. Each development can be genuinely valuable without being universally available, equally effective or ethically complete.
The gap between invention and public benefit is where many ethical questions live. A service may technically exist but remain unusable because it is unaffordable, inaccessible, unavailable in a person’s language or dependent on infrastructure their community does not have. A system may work well on average while repeatedly failing a smaller group. A treatment may be clinically impressive yet require specialist centres, prolonged care, travel and follow-up that make access far more complicated than its price alone suggests.
It means removing unjust barriers, examining how benefits and risks are distributed, accommodating different needs and protecting each person’s ability to understand, choose, refuse and seek remedy.
In 2025, the International Telecommunication Union estimated that 2.2 billion people remained offline, mostly in low- and middle-income countries.1 Connectivity has expanded enormously, but a headline count of internet users does not reveal connection quality, device access, affordability, digital skills, accessibility or whether useful services are available locally. Similar layers appear in healthcare, education and public services.
A breakthrough works
Researchers establish that a method can produce a desired effect under defined conditions.
A system reaches people
Infrastructure, institutions, trained workers and sustainable financing make use possible beyond a trial.
People can benefit fairly
Access, outcomes, choice and accountability are examined across populations—not assumed from averages.
Access is a chain with at least six links
Calling a product “available” can hide the practical distance between its existence and a person’s ability to benefit from it safely.
If any essential link fails, nominal access may produce little benefit. A free learning platform is not meaningfully accessible to a household sharing one unreliable device. A diagnostic model is not equitable if it performs poorly for a subgroup and clinicians are not warned. A highly advanced treatment does not reach a patient merely because an insurer covers the medicine if travel, fertility preservation, time away from work or long-term care remain impossible.
Equality and equity are related but not identical
Equality often means providing the same resource or rule. Equity examines whether different starting conditions and needs prevent that formally equal offer from producing a fair opportunity. Sometimes the fairest design is universal: captions, clear language and secure defaults help many people. Sometimes additional support is necessary: an adapted interface, travel assistance, a community health worker or an offline route.
A system can improve the average outcome while worsening results for a smaller population. Evaluation should report who was included, how performance differs across relevant groups and whether anyone experiences a new or disproportionate burden.
Inequity can enter at every stage of innovation
Bias is not only a defect in an algorithm. It can begin with which problems receive funding, continue through research and design, and appear again when a tool meets an unequal world.
-
Problem selection
Investment follows incentives, political priorities and purchasing power. Needs affecting people with less market influence may remain understudied even when their social cost is high.
-
Evidence and data
Who participates affects what researchers can learn. Missing groups, poor measurement, historical discrimination and labels based on flawed past decisions can shape results before a model is trained or a product is tested.
-
Design choices
Language, defaults, device requirements, sensory assumptions and definitions of “normal” determine who can use a system comfortably and whose behaviour is treated as an exception.
-
Deployment conditions
A tool validated in one institution may behave differently where staffing, infrastructure, culture, disease prevalence or incentives change. Context is part of performance.
-
Outcomes and burdens
Measure not only accuracy or adoption, but waiting time, false decisions, privacy loss, unpaid labour, environmental cost and the consequences of being unable to opt out.
-
Feedback, repair and exit
People closest to a failure often notice it first. They need a visible reporting route, a timely human response, correction of records and a safe alternative while the problem is investigated.
A diverse launch team or one consultation cannot guarantee equitable outcomes. Participation must influence decisions, and monitoring must continue after deployment as users, conditions and the technology change.
The same ethical principles look different across technologies
AI, biotechnology and neurotechnology should not be governed as one undifferentiated category. The depth of intervention, reversibility, evidence, affected rights and consequences of error all matter.
Access is more than giving every learner an app
AI can offer translation, examples, feedback and additional practice. Yet learning depends on curriculum quality, teacher support, reliable devices, language coverage, accessibility and whether the student is still doing the thinking. Systems used for admissions, grading, discipline or learner profiling require much stronger scrutiny than an optional brainstorming tool.
- Provide a non-AI route for essential services.
- Keep educators responsible for consequential judgments.
- Test for accessibility, uneven error and false mastery.
- Minimise collection of children’s and learners’ data.
Efficiency cannot erase due process
Automation may help process cases or identify patterns, but a person can be seriously harmed by an incorrect decision about healthcare, employment, credit, benefits or public safety. The right question is not whether a model is impressive; it is whether its role, evidence and safeguards are appropriate to the decision.
- Notify people when automated systems materially shape decisions.
- Provide understandable reasons and meaningful human review.
- Prevent the reviewer from merely approving the machine’s result.
- Record and investigate patterns of error and appeal.
The medicine is only one part of access
The FDA approved the first CRISPR/Cas9-based treatment for sickle cell disease in December 2023, not 2025.5 Its arrival was an important scientific milestone. It also illustrates why access cannot be reduced to a list price: treatment can involve specialised centres, stem-cell collection, intensive conditioning, extended care, fertility considerations and long-term follow-up.
- Compare benefits and burdens with existing care.
- Plan capacity where disease burden is greatest.
- Include travel, time, support and follow-up in financing.
- Maintain registries and transparent long-term safety oversight.
Intimate data requires durable responsibility
Neurotechnology ranges from non-invasive research and consumer devices to implanted medical systems. These uses do not carry identical risks. More intrusive or consequential systems raise questions about mental privacy, changing consent, cybersecurity, repair, software support, ownership of derived data and responsibility if a company withdraws a product.
- Separate therapeutic evidence from enhancement marketing.
- Treat consent as ongoing, not a single signature.
- Plan maintenance, replacement and safe discontinuation.
- Limit secondary use of neural and behavioural data.
An ethical compass for emerging technology
Ethical principles do not generate one automatic answer. They make important values visible, reveal conflicts and help institutions explain why a decision is justified.
Treat people as persons
Do not reduce someone to a prediction, score, genome, diagnosis or source of data. Technology should operate within human rights, not redefine them for convenience.
Make choice meaningful
Consent requires understandable information, freedom from improper pressure and a realistic alternative. An unreadable notice is not genuine control.
Test the complete effect
Compare promised benefits with physical, psychological, social, economic and environmental risks—including harms created by errors or misuse.
Examine distribution
Ask whether people with similar needs are treated consistently and whether existing disadvantage creates unequal access, exposure or outcomes.
Collect less and protect it well
Data that is useful can also be sensitive. Limit collection, access, retention and reuse; protect systems throughout their operating life.
Keep responsibility identifiable
Someone must have authority and duty to investigate, correct, compensate, suspend or withdraw a system. “The algorithm decided” is not accountability.
From principles to governance
UNESCO’s Recommendation on the Ethics of Artificial Intelligence places human rights and dignity at its centre and calls for fairness, inclusion and human oversight.2 WHO’s work on AI for health and human genome editing similarly connects scientific opportunity with safety, effectiveness, ethics and accountable institutions.4 These frameworks are valuable because they cover a technology’s life cycle rather than treating ethics as a final approval box.
A practice may be technically lawful yet still be exploitative, inaccessible, misleading or unsupported by sufficient evidence. Conversely, different jurisdictions may regulate the same risk differently. Organisations remain responsible for reasoned judgment.
Responsible progress uses safeguards proportionate to risk
The useful choice is rarely “innovation or ethics.” The real task is to match evidence, oversight and reversibility to the seriousness of the possible consequences.
A spelling assistant, a classroom grading system, an AI diagnostic device and an implanted brain interface should not pass through identical gates. Governance becomes more demanding when a system is difficult to reverse, affects fundamental rights, reaches many people, acts without meaningful choice, uses highly sensitive data or could cause severe harm.
| Context | Illustrative example | Minimum questions | Safeguard intensity |
|---|---|---|---|
| Low consequence and reversible | An optional tool that reorganises personal notes. | Is data protected? Can the output be checked? Can the user leave easily? | Clear information, secure defaults and simple correction. |
| Moderate consequence | A learning system recommending exercises or content. | Does it work for different learners? Does it narrow opportunity or create dependency? | Educator oversight, accessibility testing and continuing outcome review. |
| High consequence | A system influencing employment, education access, credit, benefits or clinical care. | Is the use lawful and necessary? What evidence supports it? Can a person appeal effectively? | Independent evaluation, documentation, trained human authority, audit and enforceable remedy. |
| Physical or deeply personal intervention | Gene therapy or implanted neurotechnology. | Are safety and benefit sufficiently established? Is consent durable? Who provides lifelong support? | Clinical and ethics review, specialist regulation, long-term monitoring and continuity planning. |
Stage-gated innovation
-
Define the intended benefit and unacceptable harm
Name the population, setting, problem and boundaries before building. Record who helped define them.
-
Test assumptions before scale
Use representative evidence, realistic conditions and comparison with existing alternatives—not only a favourable demonstration.
-
Deploy gradually where uncertainty remains
Limit scope, monitor closely and protect participants. A regulatory sandbox should be a controlled learning environment, not immunity from rights or safety duties.
-
Keep rollback possible
Define pause thresholds, manual alternatives, data correction, incident response and responsibility for people affected during suspension.
-
Re-authorise with evidence
Review performance, distribution of outcomes and new risks. Permission to continue should not be automatic merely because a system has become familiar.
The European Union’s AI Act illustrates a risk-based approach: it entered into force in 2024 and applies in stages, with different duties for prohibited practices, general-purpose models and certain high-risk systems.3 The exact legal duties depend on the system, role, use and current implementation timeline, so organisations should consult the official text and qualified legal guidance rather than rely on simplified labels.
Design inclusion into infrastructure, research and delivery
Equitable access cannot be repaired solely with a discount at the end. It requires coordinated choices from the first research question through financing, deployment and long-term support.
Build for meaningful connectivity
Coverage is only the beginning. Reliable electricity, affordable service, suitable devices, maintenance, secure identity systems, digital skills and offline or low-bandwidth options determine whether connection becomes capability.
Treat accessible design as core quality
Use recognised accessibility standards, involve disabled people in testing and support multiple ways to perceive, understand and operate a service. Retrofitting after exclusion is usually harder and less complete.
Localise more than words
Translation does not automatically create cultural or practical relevance. Examples, assumptions, risk communication, payment routes and support must fit the setting in which people will act.
Share decision-making power
Consultation matters when it can alter requirements, budgets, timelines or whether a project proceeds. Compensate community expertise and report how input changed the design.
Improve representation responsibly
Recruitment should address justified evidence gaps without treating communities as data sources. Consent, benefit sharing, governance and return of useful findings matter alongside sample diversity.
Fund the complete pathway
Consider price, delivery capacity, workforce, travel, supporting care, monitoring and opportunity costs together. A financed product without a functioning pathway remains inaccessible.
Use purchasing power to reward responsible design
Public agencies, hospitals, schools and large employers can make accessibility, data protection, interoperability, subgroup performance, incident reporting and exit support part of procurement. Contracts can require vendors to disclose material changes, assist data migration and continue critical support for an agreed period.
For high-cost therapies, financing experiments may connect payment to outcomes or coordinate negotiation across payers. The United States’ voluntary CMS Cell and Gene Therapy Access Model, initially focused on sickle cell disease, is one example: participating states and manufacturers use CMS-negotiated outcomes-based agreements, while the model also recognises practical supports associated with treatment.6 It should be evaluated as an ongoing model, not treated as proof that every access problem has been solved.
Partnerships are stronger when they develop local research leadership, technical skills, maintenance, governance and decision-making authority. Donating a tool without the ability to adapt, repair or question it can create a new form of fragility.
A practical roadmap for the people shaping technology
Responsibility is shared, but it must not become so diffuse that no one is answerable. Each institution has distinct levers and duties.
| Stakeholder | Before development or purchase | During deployment | After launch |
|---|---|---|---|
| Governments and regulators | Set rights-based rules, fund neglected needs, assess infrastructure and require evidence proportionate to risk. | Inspect, enforce, publish guidance and protect independent research, complaints and whistleblowing. | Track outcomes, close loopholes, coordinate across borders and provide effective remedy. |
| Companies and investors | Define intended users and excluded uses; budget for accessibility, security, support and responsible exit. | Document limitations, monitor distribution of errors and give product teams authority to pause unsafe work. | Report serious incidents, correct harms, maintain critical products and avoid abandoning dependent users. |
| Researchers and universities | Choose socially relevant questions, improve representation and establish meaningful consent and governance. | Preserve research integrity, register appropriate studies and report limitations and adverse findings. | Share useful knowledge responsibly, support replication and examine real-world effects beyond publication. |
| Health and education institutions | Compare the technology with existing practice and include professionals, patients, learners and caregivers. | Train users, maintain human responsibility and provide accessible alternatives and appeals. | Measure learning, health and equity outcomes—not merely usage—and stop systems that do not justify their burden. |
| Communities and civil society | Articulate needs, non-negotiable rights and locally meaningful definitions of benefit. | Observe implementation, support public understanding and identify hidden costs or excluded groups. | Use complaints, public-interest research, advocacy and participatory review to demand correction. |
Publishing a long technical document is not enough if affected people cannot understand it or do anything with the information. Useful transparency connects a clear explanation with choice, oversight, correction and consequence.
A compact equity impact review
Use these questions at the proposal stage, before launch and at scheduled review points. The answers should be supported by evidence and assigned to named owners.
Twelve questions before scaling
If the evidence is missing, record the uncertainty and decide how it will be resolved. Silence is not evidence of low risk.
- What exact problem is being solved, and who defined it?
- Who is expected to benefit, and what would meaningful benefit look like?
- Who may be excluded by cost, language, disability, location, device or institutional rules?
- What groups are absent or too small in the evidence?
- What data is collected, inferred, retained, shared or reused?
- Can people participate without surrendering unnecessary privacy?
- Is consent genuinely voluntary, understandable and reversible?
- How do errors and outcomes differ across relevant populations?
- What happens to a person when the system is wrong?
- Can a qualified human review and change a consequential decision?
- Who can pause, repair or withdraw the system, and at what threshold?
- Who pays for support, correction, transition and long-term obligations?
Evidence worth keeping
- Purpose record: intended use, prohibited use, target population, alternatives and success criteria.
- Participation record: who was involved, how they were selected, how they were compensated and what changed.
- Evidence record: datasets, study populations, testing conditions, uncertainty and known limitations.
- Impact record: benefits, errors, complaints, appeals, access barriers and outcomes across relevant groups.
- Decision record: named owners, risk acceptance, corrective actions, review dates and reasons for continuing or stopping.
Documentation is not valuable merely because it exists. It should help a responsible person understand the system, trace a decision, identify an emerging pattern and act before preventable harm spreads.
Common myths and practical questions
Simple slogans can make ethical disagreements appear easier than they are. These distinctions create room for better decisions.
Does equity slow innovation?
Thoughtful safeguards can require time and resources, especially for high-consequence systems. That does not make them opposed to innovation. Discovering inaccessible design, unsafe performance or public rejection after large-scale deployment can be far more costly. The right pace depends on the seriousness and reversibility of the risk.
Will breakthrough technologies naturally become affordable?
Some technologies become cheaper through scale, competition and learning. Others depend on scarce materials, specialist labour, complex manufacturing, continuing support or strong market exclusivity. Affordability is an outcome of technology, institutions, financing and policy—not an automatic law.
Can algorithmic bias be solved by adding more data?
More representative and higher-quality data may help, but volume alone cannot repair a poorly defined goal, unjust historical labels, missing context, unsuitable deployment or a harmful use. The full decision system—including institutions and incentives—must be examined.
Is open source automatically equitable?
Open code or knowledge can lower some barriers and enable inspection or local adaptation. People may still lack computing resources, specialist skills, suitable data, support or legal freedom to use it. Openness can also make unsafe capabilities easier to distribute. Its effects depend on what is opened, for whom and with which safeguards.
Is transparency enough to make a system ethical?
No. A system can be clearly documented and still be unsafe, coercive or unfair. Transparency matters when it supports informed choice, independent evaluation, accountability and remedy. It complements rather than replaces substantive limits.
Should everyone receive equal access to every enhancement?
Not every proposed enhancement is safe, beneficial or socially desirable, and healthcare priorities must consider need, evidence and opportunity cost. Equity does not require indiscriminate distribution. It requires fair reasoning, protection from discrimination, meaningful participation and special care when access could create durable forms of power or exclusion.
Can one global ethical framework fit every culture?
Implementation should respect language, context and legitimate cultural difference, while fundamental rights protect people from having coercion or discrimination excused as custom. Global principles and local participation should inform one another rather than forcing a choice between them.
Who is responsible when many organisations built the system?
Responsibility can be distributed, but it should still be explicit. Developers, deployers, professional users, purchasers, executives and regulators may hold different duties. Contracts and governance should identify who monitors, responds, corrects, compensates and decides whether operation can continue.
Progress becomes greater when more people can shape it
A technology’s technical capability is only part of its achievement. The rest lies in the institutions and relationships around it: whether research reflects real needs, whether evidence includes the people who will rely on it, whether access is practical, whether choice is respected and whether errors can be challenged and repaired.
Ethics should neither decorate innovation nor freeze it. It should help society steer—making risk visible, setting boundaries, protecting those with less power and preserving space to learn. Inclusion improves this process because communities contribute knowledge that distant designers, investors and regulators may not possess.
The aim is not a world in which everyone uses the same technology. It is a world in which valuable advances do not become privileges by default, vulnerability is not treated as permission, and people remain participants in decisions that alter their education, health, work, bodies and future.
It works. It solves a worthwhile problem. People can reach and use it. Its burdens are justified and fairly distributed. Those affected retain dignity, choice and a meaningful route to correction.
This article is for general educational purposes and does not constitute legal, medical, regulatory, investment or policy advice. Requirements and evidence change; consult current official sources and qualified professionals for consequential decisions.
Evidence lens and further reading
The rewrite replaces the original article’s unsupported statistics and nonexistent or misdescribed programmes with primary institutional sources.
-
International Telecommunication Union — Facts and Figures 2025
Global estimates on internet use and continuing gaps in connectivity, affordability and quality. -
UNESCO — Recommendation on the Ethics of Artificial Intelligence
A global framework centred on human rights, dignity, fairness, inclusion, transparency and human oversight. -
European Commission — Regulatory framework for AI
The official overview and current staged implementation timeline for the European Union’s AI Act. -
World Health Organization — Human Genome Editing: A Framework for Governance
Values, principles, institutions and governance tools for somatic, germline and heritable human genome editing. -
U.S. Food and Drug Administration — First gene therapies approved for sickle cell disease
The FDA’s December 2023 announcement, including the first FDA-approved treatment using CRISPR/Cas9. -
Centers for Medicare & Medicaid Services — Cell and Gene Therapy Access Model
Current model details on outcomes-based agreements and supporting access to sickle cell gene therapy. -
World Health Organization — Ethics and Governance of Artificial Intelligence for Health
Guidance connecting AI’s health potential with ethics, human rights and public accountability. -
U.S. National Institute of Standards and Technology — AI Risk Management Framework
A voluntary framework organised around governing, mapping, measuring and managing AI risks.