human in the loop
A design approach where a human is integrated into an automated process to correct or check the AI's outputs at strategic points.
As defined in How Better Feedback Can Revolutionize Education (with Daisy Christodoulou)
Other definitions
Design pattern in which a person reviews or approves automated decisions before they take effect.
Des Traynor - Real Talk about AI and Software - [Invest Like the Best, EP.340]
Where it comes up · 11
- How Better Feedback Can Revolutionize Education (with Daisy Christodoulou)4 claims
- The Futuristic Military Tech That S Changing The War In Ukraine Ep. 56 Kate Bondar2 claims
- The Real Life Slaughterbots We Re Building A Battlefield From Hell Paul Scharre2 claims
- 330 - The Doomsday Machine2 claims
- 280 - The Future Of Artificial Intelligence2 claims
- The AI Revolution Is Underhyped | Eric Schmidt | TED2 claims
- Investing in Artificial Intelligence1 claim
- The Gift and The Curse of Staying Private1 claim
and 3 more
Most important claims · 10 of 19Search all claims for “human in the loop” →
- International bans against autonomous weapons that target people may be more achievable than broad bans because their military utility is lower (humans don't move fast, so keeping a human in the loop is feasible), the moral 'ick factor' is higher, and the moral arguments are stronger—since a soldier can flee a tank or surrender to avoid being targeted but cannot stop being a human.The Real Life Slaughterbots We Re Building A Battlefield From Hell Paul Scharre
- Even with a human nominally in the loop—a person pushing a button to approve attacks—they may have effectively ceded all judgment to the machine, serving merely as a scapegoat for accountability rather than exercising meaningful control, which is both morally and potentially legally unacceptable.The Real Life Slaughterbots We Re Building A Battlefield From Hell Paul Scharre
- A human-in-the-loop requirement may be illusory under time compression: if an AI tells a ship's captain a missile will hit in 20 seconds unless a button is pressed, the captain will press it — but that is not meaningful human control, because the compressed time prevents the reasoned judgment that, as in the Cuban Missile Crisis submarine incident, has historically averted catastrophe.280 - The Future Of Artificial Intelligence
- Even with a firm commitment to keep a human in the loop, the strong incentive to use AI to synthesize early-warning data and provide decision support could leave humans as mere button-pushers, since people are unlikely to reject the conclusions of a system that has proven 99% reliable, creating a slippery slope toward effectively delegating nuclear decisions to an algorithm.330 - The Doomsday Machine
- Digital systems and human systems are prone to different modes of failure, but the fundamental problem is making high-stakes decisions under incredible time pressure; whether the decision is delegated to machines or to people, these failure modes persist, so the premise that we must accept rapid decision-making should be rejected.330 - The Doomsday Machine
- The US doctrine holds that nuclear weapons should only be launched under human control, but China and Russia have so far declined to agree even to this minimal restriction — which bodes poorly for getting them to agree to any restrictions on more advanced AI capabilities.280 - The Future Of Artificial Intelligence
- US military doctrine requires 'human in the loop' or 'meaningful human control' through rule 3000.09, which mandates that automated systems must remain under human control and represents a critical line that cannot be crossed.The AI Revolution Is Underhyped | Eric Schmidt | TED
- The US military doctrine 3000.09 ('human in the loop' or 'meaningful human control') prohibits autonomous weapons systems not under human control, establishing a boundary line that Schmidt considers correct and necessary.The AI Revolution Is Underhyped | Eric Schmidt | TED
- Keeping a human in the loop for AI decisions erodes attention, similar to how drivers using assisted driving pay less attention because they know the car drives itself, so high-stakes use needs a higher threshold, and it is hard to define when generative AI should be excluded.Des Traynor - Real Talk about AI and Software - [Invest Like the Best, EP.340]
- Per the Paradox of Automation, when a system handles 90-95% of cases but errs on 5-10%, asking humans to stay alert to catch those errors produces worse outcomes than having them do the whole task, because humans disengage—as with self-driving cars where a texting driver cannot re-attune fast enough to take over safely.How Better Feedback Can Revolutionize Education (with Daisy Christodoulou)