How Safe AI is trusted and responsible: 5 Quantexa responsible AI principles
Sign in to watch the recording
The write-up below stays open to everyone. The video itself needs a free account.
About
How Safe AI is trusted and responsible: 5 Quantexa responsible AI principles
Recap
Felix Hoddinott, co founder and Chief Analytics Officer of Quantexa, and Siobhan, Quantexa's AI hub lead, walked through the five principles Quantexa uses to decide whether an AI system is safe and responsible. The principles grew out of work with Quantexa's clients and partners, and this year a fifth was added at the front of the list.
The first is being risk based. Look at both the technology (what it is, who owns it, where your data goes) and the use case (what happens if the answer is wrong). An AI scheduling mistake is a nuisance. An AI recommendation in a building survey could cause financial or physical harm. The second is accountability: the AI is never accountable for its own output, so you should be able to name the person responsible for each AI tool you use, and say whether a person checks its work.
Third, fair and justifiable. Using data about people is often unavoidable, but it has to be justified. A simple test: look closely at the cases where the AI got it wrong and check they don't fall unfairly on one group. Fourth, transparent and explainable: tell people when AI is being used, and be able to explain how an AI driven decision was reached. Fifth, robust and secure: the tool should give consistent answers if you ask the same thing several times, and your data should be protected, with multi factor authentication where it's available. If data is misused, the responsibility sits with your business, not the AI provider.
Each principle was linked back to the AI Security Health Check, including a demo of someone running it, getting an action plan and ticking off improvements. The presenters also practised what they preached: the session materials were drafted with AI help and carefully checked by a person before use.
Key takeaways
- Judge risk by both the tool and the use. The same AI can be fine for one job and dangerous for another.
- Make sure a named person is accountable for every AI tool you use.
- Check where your AI gets things wrong, not just where it gets them right.
- Tell customers and staff when AI is involved, and be able to explain its decisions.
- Test for consistency: ask the same question several times and compare the answers.
- Using AI to help spot something you missed is responsible. Letting it decide who to hire is not.