AI + HUMAN INTELLIGENCE
AI — Changing the relationship between cognitive profile and opportunity
AI can support, augment or substitute cognitive work. Its most important human consequence may be that it changes which cognitive abilities are required to succeed — and therefore changes the relationship between a person’s cognitive profile and the opportunities available to them.
AI changes the conditions of intelligence
Generative AI is not merely another object to compare with human intelligence. It is becoming part of the environment in which human intelligence is used. A person’s performance can now emerge from an interactive system containing human knowledge, model capability, interface design, retrieved information and organisational rules.
This means output alone tells us less about underlying human capability than it once did.
Support, augmentation and substitution
Support reduces incidental cognitive friction while leaving the core operation with the person. Augmentation extends what the person can accomplish while they remain meaningfully engaged in reasoning. Substitution transfers the cognitive operation itself to the model.
The same technology can occupy all three roles. What matters is which operation moved, what remained human, and what the consequences are for performance, learning and responsibility.
Support: reduce friction while the person retains the core cognitive operation
Augmentation: extend what the person can accomplish while they remain meaningfully engaged
Substitution: transfer the cognitive operation itself to the system
AI may change cognitive fit
A person with excellent reasoning but weak organisation may be able to externalise planning. Someone with memory impairment may use an AI system as structured external memory. Someone whose ideas exceed their expressive fluency may use AI to help formulate language. A citizen overwhelmed by bureaucracy may use an agent to organise requirements.
These are not trivial conveniences. They can change whether a cognitive profile is compatible with a task, job or institution. AI may therefore alter access to opportunity without altering the person’s underlying trait ability.
Assisted performance is real performance — but it is not the same construct
If a person successfully completes meaningful work with a tool, the result is real. Humans have always used tools. But for science, education and high-stakes decisions we still need to distinguish what the person can do independently, what the person–AI system can do, and whether the person can supervise the system.
That distinction matters when the tool fails, when circumstances change, when transfer is required, or when the human retains responsibility for the outcome.
Evaluation may become a new bottleneck
AI can generate fluent output cheaply, while verification may remain cognitively expensive. This creates a plausible risk: people can obtain expert-sounding answers without necessarily possessing the domain knowledge needed to evaluate subtle errors. How often evaluation actually becomes the limiting step, for whom, and on which tasks is an empirical question rather than a settled conclusion.
Metacognition, calibration, problem formulation and error detection may therefore become increasingly important. But these too differ across people and can be affected by fatigue, stress and overconfidence.
AI could reduce one inequality and create another
AI may reduce barriers associated with writing, language, organisation or access to specialised knowledge. It may also advantage people who are already better able to formulate problems, supply context, evaluate evidence and detect subtle errors.
The effect on cognitive inequality is therefore not fixed. We need to measure who benefits, on which tasks, under which interfaces and with what longer-term consequences.
Cognitive offloading and development
Offloading is not inherently harmful. Writing, calculators and navigation systems already extend human cognition. The critical question is what happens to the capability that is no longer exercised.
In education, substitution may improve output while reducing practice. At work, automation may remove routine tasks that once built expertise. In daily life, an assistant may increase autonomy while simultaneously creating dependence. These trade-offs need longitudinal evidence rather than assumptions.
AI under stress and overload
One of AI’s most promising roles may be support when human cognitive headroom is temporarily reduced. It could help prioritise, remember, structure and reduce administrative load during illness, caregiving, acute stress or extreme workload.
But reduced headroom can also make a person less able to detect an AI error. The moment assistance is most valuable may therefore also be the moment oversight is weakest. Designing for that tension should be a major research priority.
Human–AI complementarity should be designed, not assumed
The ideal system is not necessarily the most automated one. Some tasks benefit from preserving human deliberation; others are safer when routine operations are delegated. Good design should allocate cognitive work according to comparative strengths, error costs, learning needs and responsibility.
Friction can sometimes be protective. An AI system may need to expose uncertainty, request confirmation, require the user to inspect evidence or deliberately slow a consequential decision.
What HIW should investigate
Which cognitive profiles gain the most from which kinds of AI support? When does AI compensate for a weakness, and when does it conceal a capability that still matters? How do we distinguish independent, supported and substituted performance? How does repeated use change skill, confidence and adaptive functioning?
Can AI preserve agency during illness, disability, ageing, stress and overload? How should assistance adapt to current cognitive state? Which human abilities become more valuable as generation becomes cheap? What should people continue to practise even when machines can perform the task?
SOURCES
References & resources
Selected external sources used across this page. External links open in a new tab.
NEXT
Follow the human question.
AI matters here because it changes tasks, support and the conditions under which people use their intelligence.