written by Henry Lieberman from MIT
“So, what can we do to win the trust of our customers?”
That wasn’t the question you were expecting — not from the CEO, on this rare occasion when you get to meet them. Here’s someone who probably makes a couple of dozen, or hundreds of times as much as you — and they’re asking for your advice. This could be your big chance.
You’re a respected scientist or engineer in your organization. But you usually stay out of management, and stick to technical questions. You want to be proud of your work, and see your company do the best it can for its customers and its employees. But you know that the company doesn’t always live up to the ethical standards it could, and you’ve got a bunch of ideas for how it could do better. The CEO might or might not take your advice, but if they do, you’ve got a rare opportunity to do good for the world.
You tense up, and reflect for a moment before you start to answer. You know the CEO is always under intense pressure to increase profits, and that motive is responsible for a lot of the shortcomings you perceive. You can’t just dump all your best ideas, because you expect that a lot of them would get shot down. But that doesn’t mean you can’t have a positive impact. You just have to pick your battles. You have to figure out what would be acceptable and feasible, given the position that the executive is in. So what do you do?
If the products in question have something to do with artificial intelligence (and these days, it’s getting more and more probable that they will), here’s what you do: You give him or her Ben Shneiderman’s article, “Bridging the Gap Between Ethics and Practice” in ACM Transactions on Intelligent Interactive Systems.
And “gap” is a good way of framing the problem. Nobody’s fooled by vague and lofty ethical “mission statements”. What your CEO needs is practicalities: Best practices that are already used (if perhaps not widely) in large organizations. Details about how to get them into practice. Shneiderman delivers.
He talks about AI safety by making analogies to safety in industries like aviation or auto safety. Backed up with references from the experience of the industry. Sure, the airlines and car companies were pressured into it by the government, but that doesn’t mean that other industries can’t learn. How to modify traditional software engineering workflows to the very different workflow of machine learning development. How to go from traditional quality control frameworks to the best practices for machine learning evaluation, especially evaluation for ensuring fairness. He presents analogies to feedback mechanisms like military After Action reviews and pharmaceutical Adverse Event reporting. He gives even the most territorial of CEOs a framework for dealing with external review and government regulation without obstructionism. A few things stretch credulity. He touts insurance companies as bulwarks against risk; I think the experience that most people have with insurance companies won’t inspire confidence. Mostly, though, this is sensible, practical advice, if a little on the conservative side.
If, miracle of miracles, the CEO actually takes all of your (or Shneiderman’s) advice, would that result in AI that can be reliable and trusted by customers? Alas, it probably won’t get us all the way to that. The problem is, there will still remain the fundamental conflict between the narrow pursuit of profit and the ethical principles of service to stakeholders. Not even the CEO can change that. In the short term. But in the long term, hopefully, we’ll evolve towards a wiser and more just society. Hopefully, again,
AI and other technologies might help us do that.
And in the meantime? Articles like Shneiderman’s will help, and that’s enough reason to give it to your CEO. And I wish you good luck with that CEO meeting.
