Collective Intelligence

Human factors engineering improves people’s lives by making technology work well for them. Most broadly, human factors engineering aims to improve human interaction with systems by enhancing:
• Safety: Reducing the risk of injury and death
• Performance: Increasing productivity, quality, and efficiency
• Satisfaction: Increasing acceptance, comfort, and well-being

In considering these goals, safety is always a critical concern, but the relative emphasis of each goal depends on the particular area of application: high-risk systems, such as a B787 cockpit; workplace design, such as a manufacturing assembly line; consumer products, such as an iPhone. Design of high-risk systems must focus on safety. In contrast, design of workplaces focuses more on performance, and design of consumer products focuses more on satisfaction.

A systematic review of human-AI interaction in autonomous ship systems

Область исследований, которая изучаю отказоустойчивость и безопасность сложных социо-технических систем + пара интересных ссылок

https://github.com/lorin/resilience-engineering

You can think of robustness as being able to deal well with known unknowns, and resilience as being able to deal well with unknown unknowns.

https://en.wikipedia.org/wiki/Resilience_engineering

https://en.wikipedia.org/wiki/Safety_engineering

https://en.wikipedia.org/wiki/MTTFd

https://www.resilience-engineering-association.org/

Подход для сравнения методов оценки АБ

Интересный подход: логировать полезность фичей в моделях в общем хранилище, чтобы потом ускорять подбор для новых моделей

Training Data Attribution (TDA) - метод, который помогает понять какие примеры из обучающий выборки сильнее влияют на знание модели

https://medium.com/people-ai-research/scaling-training-data-attribution-f7d1eddd85da

Еще один материал про купед. На этот раз от мордокниги и с примером кода


https://medium.com/@AnalyticsAtMeta/how-meta-scaled-regression-adjustment-to-improve-power-across-hundreds-of-thousands-of-experiments-624e08aaf560

Что такое "хорошо используемый ИИ"? Зависит от задачи.

Концепт от одного информационного архитектора https://jarango.com/2024/10/01/exploring-the-ai-solution-space/

TL;DR – Spotify повышает CTR по рекомендованным блокам, когда объясняет при помощи LLM почему он показывается

https://research.atspotify.com/2024/12/contextualized-recommendations-through-personalized-narratives-using-llms/