Creating AI Solutions That Users Trust
Designing trust into AI systems
A black-box that can be trusted
The Oracle of Silicon Valley and the so-called web guru and trend spotter, Tim O’Reilly says:
“The great question of the 21st century is going to be 'Whose black box do you trust?”
AI is increasingly influencing our social, cultural, economic, and political interactions. More and more of our lives are in the hands of “black boxes,” i.e., algorithms making decisions and shaping choices but whose inner workings are often a mystery even to their creators (deep learning models are an example of such black boxes). These black boxes are not only affecting us at the individual level, but algorithms are now shaping organizations and societies as a whole (hint: U.S. elections and Trump). In such a world comes the key question of trust.
Sure, Alexa doesn’t make life-changing decisions for us. Or at least not yet! But if we consider autonomous cars, for example, it becomes crucial that we trust that system.
This means that understanding how to evaluate algorithms without knowing the exact rules they follow (missing transparency) is a key discipline in today’s world. O’Reilly proposes four tests to determine whether a black box algorithm can be trusted:
Clarity of intended outcome: The creators should make clear what outcome they are seeking, and it should be possible for external observers to verify that outcome. Take the example of an airplane autopilot. A clear outcome could be to respond correctly to wind and weather, in accordance with known principles of aeronautics.
Measurability: Success should be measurable. Alignment between the actual behavior of the plane and the laws of physics and aeronautics.
Goal alignment: The goals of the algorithm’s creators should be aligned with the goals of the algorithm’s consumers. For instance, a smooth ride without crashing but not added fuel consumption that increases global warming.
Long term decisions: The algorithm should lead its creators and its users to make better longer-term decisions. For example, does it mean that all the pilots lose their jobs?