Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Sometimes it does, but not automatically. The evidence says AI can raise the value of expertise when a skilled person can frame the problem, supply context, check the output and answer for the result. It does not say every expert gains, and it does not say AI leaves expert work untouched. The 2025 National Academies report on AI and the future of work puts it plainly: AI “is a tool with the potential to enhance human labor and create new forms of valuable work – but this is not an inevitable outcome.”
Why expertise can become more valuable as AI improves
AI makes it cheaper to produce drafts, retrieve information and compare options. When those things get cheap, the scarce part of the job moves. It shifts toward deciding what to ask, noticing what the system doesn’t know, catching errors, choosing between trade-offs, dealing with people and accepting accountability.
That is an interpretation of task-level evidence, not a law. It also helps to separate two things:
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Repair Windows errors before they cause bigger problems3Fix the driver behind crashes, sound loss and screen glitches- Task automation: a subtask gets done faster, but the worker’s judgment doesn’t improve. Output rises while understanding may not.
- Expertise augmentation: the worker uses the system to extend what they can do, and still knows enough to direct it and evaluate what comes back.
The studies below show both outcomes, sometimes within the same body of research.
#1 Best Overall
What the studies actually found
All figures below come from the National Academies’ 2025 consensus report (a roughly 200-page study focused on the United States), which summarizes earlier experiments. Each result belongs to its task and population.
Professional writing: faster, and a narrower skill gap
In Noy and Zhang’s 2023 experiment, college graduates did professional-writing tasks with ChatGPT (version 3.5). The National Academies reports that work time fell by about 40%, average quality rose slightly, and less-skilled writers improved most. The study does not show that people with no relevant background would do well. It measured capable graduates on defined writing tasks.
Customer support: gains concentrated among less-experienced staff
In Brynjolfsson, Li and Raymond’s 2023 study, a tool suggested responses to human support agents. The National Academies reports a 14% average increase in chat resolutions per hour, with less-experienced workers approaching expert productivity. This was a suggestion tool with a human agent in the loop. It doesn’t automatically apply to autonomous customer-facing bots or to other jobs.
Free tools Windows power users keep installed
One-click scans. No signup required.
Radiology: better AI, no better diagnoses
In the radiology experiment the report describes, the AI’s predictions were more accurate than almost two-thirds of participants’ assessments. Yet AI assistance did not improve radiologists’ diagnostic quality on average. Contextual information available to the radiologists did improve quality. The lesson is not that AI is weak. Expertise plus a tool is not automatically better than either alone, and poorly integrated assistance can waste what the tool offers.
Research synthesis: a case study, not a benchmark
A UK Department for Science, Innovation and Technology page, published 23 April 2025, describes a comparative exercise conducted by the Behavioural Insights Team for DSIT and DCMS. AI-assisted rapid evidence review took 23% less time overall and 56% less time for analysis and synthesis. The page itself warns: “This study is effectively a case-study, and therefore the results are not generalisable.” Treat the numbers as an illustration.
Where the pattern breaks down
Notice what the studies imply together. The biggest gains in writing and support went to less-skilled or less-experienced workers, which can narrow the gap between novices and experts on routine tasks. That is a point against a simple “experts win” story. Expertise matters most where tasks are ambiguous, high-stakes or dependent on context, and where someone has to judge whether an output is right.
The National Academies also says rigorous, representative workplace evidence on complementarity versus substitution is limited, and that existing studies neither confirm nor reject its complementarity framework. Outcomes depend on how AI is developed and adopted, plus training and institutional choices.
A quick way to judge your own situation
| Question | Expertise tends to matter more when… | Expertise tends to matter less when… |
|---|---|---|
| Task structure | The task is novel, ambiguous or high-stakes | The task is routine and well defined |
| AI’s role | It suggests, drafts or retrieves for you to review | It acts autonomously with little review |
| Context | Local, historical or interpersonal knowledge decides the answer | Everything needed is in the prompt or documents |
| Accountability | A qualified person can catch errors and owns the outcome | No one is positioned to check the result |
| What is measured | Quality, error rate or customer experience count | Only speed or volume count |
A gain on one measure doesn’t imply gains on others. Faster output can coexist with flat quality, as in the radiology result.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Which skills to build
The OECD’s AI and skills report (2026) says fewer than 1% of workers will need advanced AI-specific skills such as programming or model development. It highlights digital skills, using and interpreting data, managerial skills, problem-solving, creativity and innovation. For most people the practical list is:
- Deep knowledge of your field. It lets you spot a plausible but wrong answer and ask a sharper question.
- Digital and data fluency. Know what a tool can access, how to read its numbers and where it is likely to fail.
- Evaluation practice. Check outputs against sources, test edge cases and compare with your own first-pass judgment.
- Problem framing. Define the goal, constraints and what “good” looks like before you prompt.
- Judgment and ownership. Decide which trade-offs matter and be the person who answers for the outcome.
Don’t assume interpersonal skills are guaranteed to rise in value. The OECD says empathy, communication and teamwork remain essential in many jobs. It also reports early signals that demand for some social skills may fall in parts of Europe under algorithmic management, and cautions it is too early for firm conclusions.
When to trust an AI answer
- Trust more when the task is routine, the answer is easy to verify and a mistake is cheap.
- Trust less when the answer depends on context the system can’t see, or when you lack the knowledge to judge it. Over-reliance is a real failure mode, and the radiology result shows assistance can be integrated badly.
- Don’t use AI to replace understanding you need. If you can’t tell whether an output is right, you are not yet supervising it.
The Bottom Line
Expertise gains value when it is paired with a workflow that lets you direct the AI and check its work. Don’t assume it gains value just because the AI gets better. Build domain depth, data fluency and the habit of checking outputs, and treat any claim that experts will always win with suspicion.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
