Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), founder of The SuperSkills Intelligence Company. The research is published openly with every source graded. Found an error? Tell me and it is corrected on the page.
Home / Keynotes / AI and human judgement
The signature topicThe question underneath every AI decision a leadership team makes. As the machine takes over the tasks, where does human judgement belong, and how do you stop it leaking away one reasonable decision at a time?
Watch the showreel · 2 minutes
A feel for the room before you put me in front of yours.
A preregistered meta-analysis of 106 experimental studies and 370 effect sizes, published in Nature Human Behaviour, found that human and AI combinations performed significantly worse on average than the better of the human alone or the AI alone, at a Hedges' g of minus 0.23, with the losses concentrated in decision-making rather than in content creation.
Two limits travel with it, and they are stated on stage rather than left out. The comparison is against an oracle who always picks the better performer, which nobody can do in advance. And the studies were published between 2020 and 2023, so the analysis predates the current generation of models. It is an argument against assuming the pairing is free, rather than proof it cannot be made to pay.
Endoscopists averaging twenty-eight years of experience saw their unassisted detection rate fall from 28.4 to 22.4 per cent after routine exposure to an AI tool, measured on procedures performed without it. Students with unrestricted GPT-4 scored 48 per cent higher while they had it and 17 per cent below a control group once it was removed.
The room usually goes quiet at the second one, because a guardrailed tutor in the same experiment largely removed the harm. Same model, different interface, opposite outcome. That is the whole argument: the design of the relationship decides whether the tool raises capability or quietly spends it.
A way to see which mode the organisation is running. Most teams do not decide how much judgement to hand over; it moves on its own, one reasonable step at a time, until nobody is sure which decisions are still theirs. The signature keynote, Drift versus Design, sets the choice out and hands the room the controls.
Then the working artefact: which decisions stay human, which the machine may take, and who remains accountable, written down rather than settled by whoever is busiest that afternoon. That is at the delegation boundary map, and the board version at what a board should ask about AI.
The distinction is deliberate. This is about adoption and capability rather than compliance: where judgement should stay in the workflow and how to keep it strong, not model safety, regulation or what the law should say. The operational side of oversight has its own page at human oversight and accountability.
The risks named along the way have their own research: synthetic seniority, the missing rungs, and the accumulation of unexercised judgement examined at capability debt.
Everything above is set out at length, with every source graded and its limits stated, at does AI weaken human judgement: roughly 7,000 words, 34 graded sources, including the evidence that runs against the argument and where it remains uncertain. A booker who wants to know what will be said before the room fills can read all of it.
Rahim Hirji specialises in where human judgement belongs as AI takes over the tasks. He is the author of SuperSkills (Kogan Page, 2026) and publishes the underlying research openly, including a roughly 7,000-word review of the evidence with 34 graded sources. His signature keynote, Drift versus Design, is built around judgement allocation: deciding in advance which decisions stay human, which the machine can take, and who is accountable.
A preregistered meta-analysis of 106 experimental studies found human and AI combinations performed significantly worse on average than the better of human alone or AI alone, with losses concentrated in decision-making. Separately, endoscopists' unassisted detection fell from 28.4 to 22.4 per cent after AI exposure, and students scored 17 per cent below a control group once an unrestricted tool was removed. None of that argues against using AI. It argues that the design of the relationship decides the outcome.
Deciding deliberately which decisions are made by people, which by AI, and how accountability is assigned, rather than letting that boundary move on its own as tools spread. An organisation that has never written the list has already answered by default.
It concerns adoption and capability rather than compliance: where human judgement should stay in the workflow and how to keep it strong, rather than model safety, regulation or governance. Speakers who have written AI policy are better placed on the latter, and the guide to choosing an AI keynote speaker says so.
Written by Rahim Hirji, author of SuperSkills (Kogan Page, 2026), founder of The SuperSkills Intelligence Company. The research is published openly with every source graded. Found an error? Tell me and it is corrected on the page.
Tell me the room, the date and the decision you are trying to make. A reply within 24 hours, and a straight answer on fit even when the answer is somebody else.
Also for boards and leadership offsites and on human oversight and accountability. Browse every topic, audience and region.
If this was useful and you want a small amount of it each week, that is what the letter is for. Weekly essays on AI, capability and the future of work. Read by 25,000 people, every week since 2017. Free, and one click to stop.
Opens Substack to confirm. No pitch in it, unsubscribe in one click, and nobody follows up because you read something.
Running an event, or responsible for how AI arrives in your organisation? Keynotes · Advisory and coaching · Enquire
