One of the most important insights that I’ve gotten in working with biologists and ecologists is that today it’s actually not really known on a scientific basis how well different conservation interventions will work. And it’s because we just don’t have a lot of data.
Archive
The question is what are we doing in the industry, or what is the machine learning research community doing, to combat instances of algorithmic bias? So I think there is a certain amount of good news, and it’s the good news that I wanted to focus on in my talk today.
We have increasingly smart, surveillant persuasion architectures. Architectures aimed at persuading us to do something. At the moment it’s clicking on an ad. And that seems like a waste. We’re just clicking on an ad. You know. It’s kind of a waste of our energy. But increasingly it is going to be persuading us to support something, to think of something, to imagine something.
Machine learning systems that we have today have become so powerful and are being introduced into everything from self-driving cars, to predictive policing, to assisting judges, to producing your news feed on Facebook on what you ought to see. And they have a lot of societal impacts. But they’re very difficult to audit.
Quite often when we’re asking these difficult questions we’re asking about questions where we might not even know how to ask where the line is. But in other cases, when researchers work to advance public knowledge, even on uncontroversial topics, we can still find ourselves forbidden from doing the research or disseminating the research.
The smartphone is the ultimate example of a universal computer. Apps transform the phone into different devices. Unfortunately, the computational revolution has done little for the sustainability of our Earth. Yet, sustainability problems are unique in scale and complexity, often involving significant computational challenges.
I have a few examples of things that are starting to happen. And you can see that AI normally stands for artificial intelligence, but I’ve often concluded it stands for “artificial idiot.” And the reason is very simple. It turns out that these systems are good, but they’re good in kind of narrow ways.
Social referencing is so great robots should do it, too. But first there are some technical challenges that we need to solve. For instance, low energy consumption. Throughout these technical challenges, what these robots really need to do is understand the social environment that they are in.
Most companies […] don’t deliberately want to be malicious in the technology that they’re making. However, if you don’t think about the fact that your solution is not the only one, and is going to enter into a whole host of different things, then you are going to end up causing problems and it might as well be malice.