AI Principles
We build applications that use AI deeply. While we aren’t a frontier lab building models from scratch, we tune and integrate them into the products that people use. These principles guide how we do that in keeping with our vision of human-centered AI.
We seek to enhance human productivity and enjoyment with AI, not replace human workers.
We are aware of environmental concerns and actively seek to minimize our impact and mitigate what remains.
We don't use AI for autonomous decision-making that affects human well-being.
We design AI systems whose behavior reflects the intent behind them, and we test rigorously to confirm it holds in the wild.
We treat data privacy and security as design requirements, not features bolted on after the fact.
We deeply consider the downstream effects of what we build, regardless of who deploys it.
We vet the models and platforms we build on, and avoid those whose risks or provenance we can't verify.
We design for the full range of people who'll actually use what we build, not an imagined average user who resembles us.
We accept that AI is probabilistic, not deterministic, and hold our systems to a standard of being meaningfully better than what they replace—not flawless.