Member-only story
Where Ethics and Caution Collide
Consequentialism and Deontology in AI Ethics
Let’s picture two companies each building a powerful new AI model. Both believe it could meaningfully help people, working towards projects such as accelerating drug discovery, or making bespoke educational tools available to anyone with a phone.
Both also recognise it could be misused. One company runs the numbers: expected benefit to millions of users, weighed against a small, hard-to-eliminate risk of harm to some unknown few, and concludes that developing the model, imperfect as it is, is the more responsible choice than withholding it.
After all, if they don’t develop it and put it out to market, someone with more dishonourable intentions will eventually anyways.
The other company looks at the same model and says: there is a category of harm we will not risk enabling, no matter how favourable the odds look on paper. Until that risk is closed off, the model does not go out the door.
Is either of these companies being reckless? Might one of them be bluffing? You could argue that they are simply reasoning from different ethical starting points, one broadly consequentialist, the other broadly deontological. These two traditions are centuries old, but they have become unusually visible in how AI organisations talk…
