Can you talk about your book?

(upbeat music) - What Hilary Mason, Mike Loukides, and I decided was, well let's try to write this as practitioners and let's keep it super short, simple, and free. And we started with the premise of, well what would we do in our organizations to address these challenges. What have we done, and then let's use this opportunity to reflect back and say, "Is this good enough "or should we do something different?" And so we started with this idea first that we've been talking about for a while, which is oaths. Well, should data scientists take an oath? And we realized, well that could be interesting, but in many other fields where people take oaths, that's not sufficient to actually get things to work. And what we're realizing going through that process is one of the most important things when you're doing data science, especially building a data product where an algorithm is involved, is to actually have a checklist. And the idea of a checklist is very much from the Checklist Manifesto and this idea of, well if you're going into surgery, don't you want this team to actually do a checklist before they operate on you? You want to also empower the nurses or the other parts of the care team to ask simple questions in a very fast, efficient way. That just gives you pause to go, ooh, if there's a problem, we got to course correct. So what's a example of a checklist question? Who's going to own this algorithm afterwards? What happens if the algorithm does something weird? Can we shut it off? Have we actually gotten a team to test this and beat it up and try to break it or see where there gaps are and just find those unknown unknowns. And then we kind of talk about going on to other areas of where are the key tenants that we think through that we often don't talk about around a data product. How do you have what we call the five C's? Things like consent, how do you think about clarity of language, 'cause how often have we gotten this giant terms of service document that comes up and says read, read, read, and you're like "I have no idea what I'm signing up for." How do you have clarity of that language of what that really means and consistency through the product of what the implications are and the data choices and all of these things that kind of come through. And then really starting a dialog through a set of case studies that were developed at the University of Princeton and these case studies come from a backing of real world examples, but they're designed with a technologist and ethicist to really hone it down to the core issues and you can work on these case studies as a team just to see how you think. And we also talk about in there is what would it start to look like to try to drive cultural change and around, as you think about these case studies. And so for example, we talked about always when we interview a candidate, instead of just asking him a cultural question about cultural fit, you ask them an ethics question. Ethics question could be something like, we're designing this algorithm, but we're not supposed to use race. And you find out that you have a proxy for race. What do you do? It's not trying to pin the person down in there, it's just to explore how they think about it. What if everybody who was a data scientist interviewing at a company asked how do you handle ethical issues? How do you do these things? And so it kind of goes through an explore. And what we've tried to do with this book given how much is changing, is to say, one, it's free, it's creative commons license, so if you don't like it you can take our content and then add your own. And we're going to call this the dot one release. And just like open source code, there'll be a dot two and a dot three, and we're encouraging people to contribute their own chapters so that we start to develop a much richer set of content that you could think of as students train or if you're in a company, you can start talking about these things. Putting that collectively together and just finding a way of how do we start fostering the dialog as a community to get ahead of the problems that are coming. (upbeat music)