James Brusseau: This is Lane Moore, who runs a stage show called Tinder Live, where the audience watches and hears about her hookup dramas. What does the Tinder app do? It filters your personal information through giant datasets to find your perfect partner. That's the power of AI. Her proposed partners, they all start to look the same. And that's because once the platform identifies your taste, it will keep feeding versions of your model partner with different names. Which is good since that's what you like.
But you're also sacrificing the opportunity to meet anyone really different, which means you are confined in a strange way by the fact that you are getting what you want. And it's your own data creating the prohibition.
Here's another example. AI channeled me this job announcement by aligning my personal information, my education, career history, connections on the site, my comments, all of it to lock me on a career trajectory. But the better the machine works, the lower my exposure to jobs outside the parameters recommended by my own data. So again, I'm losing opportunities to experiment, to find new things to want, and I'm losing them because I'm always already getting what I want.
Let's put this theoretically. Tesla on autopilot was driving down this freeway when a white truck pulled across the road right in front. The AI read the truck's whiteness as a cloud in the blue sky, and the car did not even slow down.
The driver was texting, and in the circle you can see where the trailer sheared off the Tesla top. Beside it, there's the Tesla statement correctly noting that the car is safe statistically. The problem with AI is intellectual schizophrenia. Sometimes superhuman performance but out of the blue there are catastrophic failures at even the most humanly obvious tasks. So faced with this unpredictable unpredictability what counts as safety? Here's another version of the safety dilemma. Would real patients choose the statistically superior AI knives in their own bodies, knowing they could go haywire at any moment like the Tesla crash?
Regardless, this is the point. Novel dilemmas are emerging between humans and AI, and we haven't developed tools to manage them. In fact, we're so far behind that the most penetrating explorations in AI ethics today are not being done in academic libraries but in noisy bars in Brooklyn.
The title of this presentation is AI Human Impact, and the proposal is that one way to develop an ethical response to AI is through financial investors. And we have a successful model which is environmental social and governance investing where companies are rated not only in terms of their financial strength, but in terms of their carbon footprint, their global effect on poverty and hunger, and similar human sustainability metrics.
The problem is that ESG strategies were built for the industrial economy. The digital economy defines the words of human impact differently. The leading challenges are no longer about collectives and their well-being, but individuals wrestling for control of their own personal information and their own lives in the midst of big data machines. So, how would it work?
First, we need metrics customized to measure the human effects of AI. The set employed here is distilled from the European Council and also from academic ethics. Next, how could they be applied to an AI-intensive company?
Say, one blending smart glasses and facial recognition. These glasses let you know about the person who just stepped into your elevator – a celebrity? Pickpocket? Potential client available? With the glasses, you'd know.
But there is a tension. On one side, there's performance – an overlay of knowledge about things and people around you. You see them. On the other side, there's privacy lost. They see you. The result is an AI dilemma about autonomy. Does this AI-intensive product help or harm self-determination? Investors will be able to decide for themselves, they even have a financial interest, if technology serving human autonomy is more likely to succeed in the marketplace.
Here's the company scored ethically on a scale of negative 1-0-1. NM means not material and distinct products and companies will present different salient categories. The larger idea is to delineate the company's ethical profile so it can work alongside traditional financial information.
Finally, and besides helping investors make more informed and better decisions, AI human impact also isolates areas where activist investors may want to focus their attention. There remains a lot of work to do.