For Part 1 of this series, click here.
I recently joined a small group of E&C professionals who decided to complete the free online course on AI Ethics for Business, offered by Seattle University.
We agreed to complete a module or two every week and to share our insights. Here are some of my insights from Module 2:
- In these early days of AI, it might be best to create technology-specific regulations rather than impose a general regulatory framework. Let’s think specifically about the regulations required for self-driving cars, face recognition and nurse robots, and then extract general concepts.
- Most data-collection efforts start with good intentions. A magazine needs your home address for delivery, and later they ask for your age and your income to attract the right advertisers to place adds in the magazine. But then, the advertisers offer to buy that information from the magazine to send direct-mail marketing to subscribers, without their consent. What’s a magazine to do? How transparent must they be with their subscribers? How much agency should the subscriber have? Similar scenarios (and questions) are now being played with data collected by our phones about where we’ve been, by our watches about our resting heart-rate, by our cars about how fast we are driving, and this data is being being fed to AI engines.
- Machines learn to make decisions based on datasets that humans provide. These datasets almost always contain biases. Let’s say I want a machine to learn how to identify a good poker player. I will feed this machine with all the data that we have about the players who reached the final table of the Main Event at the World Series of Poker since its founding in 1970. The machine will see that only one woman ever made it, back in 1995. What will the machine learn from this?
- The concerns that humans have about AI and machine learning revolve around agency: we want to have control over the types of decisions machines make; we want to understand how those decisions are made; and we want to be able to override those decisions.
A side-note: I find the end-of module quizzes very poor. If all you remembered of these modules was the information included in the quizzes, you would have a dismal understanding of AI ethics.