Core tenets of trustworthy AI:
- Accuracy – AI systems should have high levels of accuracy appropriate for their intended purpose. There should be ways to validate system performance over time.
- Explainability – It should be possible to understand how AI systems make decisions. There should be transparency around the data and logic used by the system.
- Privacy – The personal data used by AI systems should be kept confidential and secure. Individuals should have appropriate control over how their data is used.
- Accountability – There should be mechanisms to determine who is responsible when an AI system fails or causes harm. Organizations should be accountable for the behavior of the AI systems they deploy.
- Reliability and Robustness – AI systems should consistently perform as intended. There should be fail-safes in place for when something goes wrong.
- Safety and Security – AI systems should be safe, secure and robust. They should not present unreasonable safety risks or be vulnerable to attacks.
- Fairness – AI systems should avoid unfair bias and should not discriminate against certain groups or individuals. The data used to train AI systems should be representative and inclusive.
- Governance – There should be well-defined practices and policies in place for developing, deploying and monitoring AI trustworthy practices to adhere to ethical principles.
AI Trustworthy Impacts
- Societal impact – The broader effects of AI systems should be considered, including impacts to jobs, fairness, democracy, and more. AI should benefit society.
- Human control – Humans should ultimately remain in control of automated systems and autonomous machines. There should always be a way to override an AI system.
- Ethical Values and Alignment – AI systems should be aligned with the moral values of their human users and society in general.
What does Malcolm Gladwell have to do with Trustworthy AI? I’ll start with saying that Malcolm Gladwell is one of my favorite authors of our generation. Malcolm has a special gift of diving deep into the anatomy of human relations. His book How to Talk to Strangers argues that people are hardwired to trust strangers. We as humans default to truthfulness in our interactions and we tend to assume that even strangers are trustworthy unless we are given a strong reason not to. This natural instinct can leads us astray. In many respects, I will argue that this behavior extends to interacting with technology. For example, if you had an urgent task to perform on a computer, it doesn’t matter if you are on your own computer, your friend’s computer or at a public kiosk, your natural instinct is to have trust with computer that is capable for completing your task.