10 Principles of Good AI
An ode to legendary designer Dieter Rams
50 years ago, legendary designer Dieter Rams laid out his 10 Principles of Good Design. Today, I propose a variant: The 10 Principles of Good AI.
1. Good AI is innovative
AI should bring new capabilities to the world, pushing the boundaries of science, technology, and knowledge. AI should not be built for hype.
2. Good AI makes a product useful
AI should only be added to a product if it enhances its experience.
3. Good AI is accessible
AI should be accessible to all — across devices, languages, health conditions, cultures, beliefs, and socio-economic backgrounds.
4. Good AI makes a product understandable
AI should make the product more intuitive. It should be self-explanatory and easy to pick up by any user.
5. Good AI is unobtrusive
AI should enhance the user experience, without any distraction or overwhelm. It should integrate seamlessly into the products we work and live with — always respecting privacy.
6. Good AI is honest
AI systems should be transparent about their training, capabilities, and limitations. They should be as open and explainable as possible.
7. Good AI is long-lasting
AI systems should be modular, designed for sustainability, and adaptable to future needs and technologies.
8. Good AI is thorough down to the last detail
Nothing in a model should be arbitrary — care and accuracy in the thought process show respect towards the user. Particular attention should be paid towards bias and representation.
9. Good AI is environmentally friendly
The development and deployment of AI should not come at the expense of our planet. Good systems should strive to minimise their carbon footprint, and use smaller models wherever possible.
10. Good AI is as little AI as possible
Models should be streamlined and efficient, avoiding unnecessary complexities, and elevating the essential functions of the product they serve. As simple as possible, but not simpler.
Which ones would you add?