As I wrote about already, last Friday I attended a one day workshop in Montreal called FATML: Fairness, Accountability, and Transparency in Machine Learning. It was part of the NIPS conference for computer science, and there were tons of nerds there, and I mean tons. I wanted to give a report on the day, as well as some observations.
First of all, I am super excited that this workshop happened at all. When I left my job at Intent Media in 2011 with the intention of studying these questions and eventually writing a book about them, they were, as far as I know, on nobody’s else’s radar. Now, thanks to the organizers Solon and Moritz, there are communities of people, coming from law, computer science, and policy circles, coming together to exchange ideas and strategies to tackle the problems. This is what progress feels like!
OK, so on to what the day contained and my copious comments.
Sadly, I missed the first two talks, and an introduction to the day, because of two airplane cancellations (boo American Airlines!). I arrived in the middle of Hannah Wallach’s talk, the abstract of which is located here. Her talk was interesting, and I liked her idea of having social scientists partnered with data scientists and machine learning specialists, but I do want to mention that, although there’s a remarkable history of social scientists working within tech companies – say at Bell Labs and Microsoft and such – we don’t see that in finance at all, nor does it seem poised to happen. So in other words, we certainly can’t count on social scientists to be on hand when important mathematical models are getting ready for production.
Also, I liked Hannah’s three categories of models: predictive, explanatory, and exploratory. Even though I don’t necessarily think that a given model will fall neatly into one category or the other, they still give you a way to think about what we do when we make models. As an example, we think of recommendation models as ultimately predictive, but they are (often) predicated on the ability to understand people’s desires as made up of distinct and consistent dimensions of personality (like when we use PCA or something equivalent). In this sense we are also exploring how to model human desire and consistency. For that matter I guess you could say any model is at its heart an exploration into whether the underlying toy model makes any sense, but that question is dramatically less interesting when you’re using linear regression.
Anupam Datta and Michael Tschantz
An issue I enjoyed talking about was brought up in this talk, namely the question of whether such a finding is entirely evanescent or whether we can call it “real.” Since google constantly updates its algorithm, and since ad budgets are coming and going, even the same experiment performed an hour later might have different results. In what sense can we then call any such experiment statistically significant or even persuasive? Also, IRL we don’t have clean browsers, so what happens when we have dirty browsers and we’re logged into gmail and Facebook? By then there are so many variables it’s hard to say what leads to what, but should that make us stop trying?
From my perspective, I’d like to see more research into questions like, of the top 100 advertisers on Google, who saw the majority of the ads? What was the economic, racial, and educational makeup of those users? A similar but different (because of the auction) question would be to reverse-engineer the advertisers’ Google ad targeting methodologies.
Finally, the speakers mentioned a failure on Google’s part of transparency. In your advertising profile, for example, you cannot see (and therefore cannot change) your marriage status, but advertisers can target you based on that variable.
Sorelle Friedler, Carlos Scheidegger, and Suresh Venkatasubramanian
Next up we had Sorelle talk to us about her work with two guys with enormous names. They think about how to make stuff fair, the heart of the question of this workshop.
First, if we included race in, a resume sorting model, we’d probably see negative impact because of historical racism. Even if we removed race but included other attributes correlated with race (say zip code) this effect would remain. And it’s hard to know exactly when we’ve removed the relevant attributes, but one thing these guys did was define that precisely.
Second, say now you have some idea of the categories that are given unfair treatment, what can you do? One thing suggested by Sorelle et al is to first rank people in each category – to assign each person a percentile in their given category – and then to use the “forgetful function” and only consider that percentile. So, if we decided at a math department that we want 40% women graduate students, to achieve this goal with this method we’d independently rank the men and women, and we’d offer enough spots to top women to get our quota and separately we’d offer enough spots to top men to get our quota. Note that, although it comes from a pretty fancy setting, this is essentially affirmative action. That’s not, in my opinion, an argument against it. It’s in fact yet another argument for it: if we know women are systemically undervalued, we have to fight against it somehow, and this seems like the best and simplest approach.
Ed Felten and Josh Kroll
After lunch Ed Felton and Josh Kroll jointly described their work on making algorithms accountable. Basically they suggested a trustworthy and encrypted system of paper trails that would support a given algorithm (doesn’t really matter which) and create verifiable proofs that the algorithm was used faithfully and fairly in a given situation. Of course, we’d really only consider an algorithm to be used “fairly” if the algorithm itself is fair, but putting that aside, this addressed the question of whether the same algorithm was used for everyone, and things like that. In lawyer speak, this is called “procedural fairness.”
So for example, if we thought we could, we might want to turn the algorithm for punishment for drug use through this system, and we might find that the rules are applied differently to different people. This algorithm would catch that kind of problem, at least ideally.
David Robinson and Harlan Yu
Next up we talked to David Robinson and Harlan Yu about their work in Washington D.C. with policy makers and civil rights groups around machine learning and fairness. These two have been active with civil rights group and were an important part of both the Podesta Report, which I blogged about here, and also in drafting the Civil Rights Principles of Big Data.
The question of what policy makers understand and how to communicate with them came up several times in this discussion. We decided that, to combat cherry-picked examples we see in Congressional Subcommittee meetings, we need to have cherry-picked examples of our own to illustrate what can go wrong. That sounds bad, but put it another way: people respond to stories, especially to stories with innocent victims that have been wronged. So we are on the look-out.
Closing panel with Rayid Ghani and Foster Provost
I was on the closing panel with Rayid Ghani and Foster Provost, and we each had a few minutes to speak and then there were lots of questions and fun arguments. To be honest, since I was so in the moment during this panel, and also because I was jonesing for a beer, I can’t remember everything that happened.
As I remember, Foster talked about an algorithm he had created that does its best to “explain” the decisions of a complicated black box algorithm. So in real life our algorithms are really huge and messy and uninterpretable, but this algorithm does its part to add interpretability to the outcomes of that huge black box. The example he gave was to understand why a given person’s Facebook “likes” made a black box algorithm predict they were gay: by displaying, in order of importance, which likes added the most predictive power to the algorithm.
[Aside, can anyone explain to me what happens when such an algorithm comes across a person with very few likes? I’ve never understood this very well. I don’t know about you, but I have never “liked” anything on Facebook except my friends’ posts.]
Rayid talked about his work trying to develop a system for teachers to understand which students were at risk of dropping out, and for that system to be fair, and he discussed the extent to which that system could or should be transparent.
Oh yeah, and that reminds me that, after describing my book, we had a pretty great argument about whether credit scoring models should be open source, and what that would mean, and what feedback loops that would engender, and who would benefit.
Altogether a great day, and a fantastic discussion. Thanks again to Solon and Moritz for their work in organizing it.
I’m back from Haiti! It was amazing and awesome, and please stand by for more about that, with cultural observations and possibly a slide show if you’re all well behaved.
Today, thanks to my math camp buddy Lenore Cowen, I am going to share with you an amazing blog post by Pamela Ribon. Her post is called Barbie Fucks It Up Again and it describes a Barbie book entitled Barbie: I Can Be a Computer Engineer
Just to give you an idea of the plot, Barbie’s sister finds Barbie engaged on a project on her computer, and after asking her about it, Barbie responds:
“I’m only creating the design ideas,” Barbie says, laughing. “I’ll need Steven and Brian’s help to turn it into a real game!”
What the fucking shit, Barbie?
Today I want to tell you guys about core-econ.org, a free (although you do have to register) textbook my buddy Suresh Naidu is using this semester to teach out of and is also contributing to, along with a bunch of other economists.
It’s super cool, and I wish a class like that had been available when I was an undergrad. In fact I took an economics course at UC Berkeley and it was a bad experience – I couldn’t figure out why anyone would think that people behaved according to arbitrary mathematical rules. There was no discussion of whether the assumptions were valid, no data to back it up. I decided that anybody who kept going had to be either religious or willing to say anything for money.
Not much has changed, and that means that Econ 101 is a terrible gateway for the subject, letting in people who are mostly kind of weird. This is a shame because, later on in graduate level economics, there really is no reason to use toy models of society without argument and without data; the sky’s the limit when you get through the bullshit at the beginning. The goal of the Core Econ project is to give students a taste for the good stuff early; the subtitle on the webpage is teaching economics as if the last three decades happened.
What does that mean? Let’s take a look at the first few chapters of the curriculum (the full list is here):
- The capitalist revolution
- Innovation and the transition from stagnation to rapid growth
- Scarcity, work and progress
- Strategy, altruism and cooperation
- Property, contract and power
- The firm and its employees
- The firm and its customers
Once you register, you can download a given chapter in pdf form. So I did that for Chapter 6, The firm and its employees, and here’s a screenshot of the first page:
The chapter immediately dives into a discussion of Apple and Foxconn. Interesting! Topical! Like, it might actually help you understand the newspaper!! Can you imagine that?
The project is still in beta version, so give it some time to smooth out the rough edges, but I’m pretty excited about it already. It has super high production values and will squarely compete with the standard textbooks and curriculums, which is a good thing, both because it’s good stuff and because it’s free.
Specifically, he looked at the 2013 cab rides in New York City, which was provided under a FOIL request, and he stalked celebrities Bradley Cooper and Jessica Alba (and discovered that neither of them tipped the cabby). He also stalked a man who went to a slew of NYC titty bars: found out where the guy lived and even got a picture of him.
Previously, some other civic hackers had identified the cabbies themselves, because the original dataset had scrambled the medallions, but not very well.
The point he was trying to make was that we should not assume that “anonymized” datasets actually protect privacy. Instead we should learn how to use more thoughtful approaches to anonymizing stuff, and he proposes a method called “differential privacy,” which he explains here. It involves adding noise to the data, in a certain way, so that at the end any given person doesn’t risk too much of their own privacy by being included in the dataset versus being not included in the dataset.
Bottomline, it’s actually pretty involved mathematically, and although I’m a nerd and it doesn’t intimidate me, it does give me pause. Here are a few concerns:
- It means that most people, for example the person in charge of fulfilling FOIL requests, will not actually understand the algorithm.
- That means that, if there’s a requirement that such a procedure is used, that person will have to use and trust a third party to implement it. This leads to all sorts of problems in itself.
- Just to name one, depending on what kind of data it is, you have to implement differential privacy differently. There’s no doubt that a complicated mapping of datatype to methodology will be screwed up when the person doing it doesn’t understand the nuances.
- Here’s another: the third party may not be trustworthy and may have created a backdoor.
- Or they just might get it wrong, or do something lazy that doesn’t actually work, and they can get away with it because, again, the user is not an expert and cannot accurately evaluate their work.
Altogether I’m imagining that this is at best an expensive solution for very important datasets, and won’t be used for your everyday FOIL requests like taxicab rides unless the culture around privacy changes dramatically.
I’ve been reading Head First Java this past week and I’m super impressed and want to tell you guys about it if you don’t already know.
I wanted to learn what the big fuss was about object-oriented programming, plus it seems like all the classes my Lede students are planning to take either require python or java, so this seemed like a nice bridge.
But the book is outstanding, with quirky cartoons and a super fun attitude, and I’m on page 213 after less than a week, and yes that’s out of more than 600 pages but what I’m saying is that it’s a thrilling read.
My one complaint is how often the book talks about motivating programmers with women in tight sweaters. And no, I don’t think they were assuming the programmers were lesbians, but I could be wrong and I hope I am. At the beginning they made the point that people remember stuff better when there is emotional attachment to things, so I’m guessing they’re getting me annoyed to help me remember details on reference types.
Here’s another Head First book which my nerd mom recommended to me some time ago, and I bought but haven’t read yet, but now I really plan to: Head First Design Patterns. Because ultimately, programming is just a tool set and you need to learn how to think about constructing stuff with those tools. Exciting!
And by the way, there is a long list of Head First books, and I head good things about the whole series. Honestly I will never write a technical book in the old-fashioned dry way again.
Yesterday was a day filled with secrets and codes. In the morning, at The Platform, we had guest speaker Columbia history professor Matthew Connelly, who came and talked to us about his work with declassified documents. Two big and slightly depressing take-aways for me were the following:
- As records have become digitized, it has gotten easy for people to get rid of archival records in large quantities. Just press delete.
- As records have become digitized, it has become easy to trace the access of records, and in particular the leaks. Connelly explained that, to some extent, Obama’s harsh approach to leakers and whistleblowers might be explained as simply “letting the system work.” Yet another way that technology informs the way we approach human interactions.
After class we had section, in which we discussed the Computer Science classes some of the students are taking next semester (there’s a list here) and then I talked to them about prime numbers and the RSA crypto system.
I got really into it and wrote up an iPython Notebook which could be better but is pretty good, I think, and works out one example completely, encoding and decoding the message “hello”.
Yesterday was the end of the first half of the Lede Program, and the students presented their projects, which were really impressive. I am hoping some of them will be willing to put them up on a WordPress site or something like that in order to showcase them and so I can brag about them more explicitly. Since I didn’t get anyone’s permission yet, let me just say: wow.
During the second half of the program the students will do another project (or continue their first) as homework for my class. We’re going to start planning for that on the first day, so the fact that they’ve all dipped their toes into data projects is great. For example, during presentations yesterday I heard the following a number of times: “I spent most of my time cleaning my data” or “next time I will spend more time thinking about how to drill down in my data to find an interesting story”. These are key phrases for people learning lessons with data.
Since they are journalists (I’ve learned a thing or two about journalists and their mindset in the past few months) they love projects because they love deadlines and they want something they can add to their portfolio. Recently they’ve been learning lots of geocoding stuff, and coming up they’ll be learning lots of algorithms as well. So they’ll be well equipped to do some seriously cool shit for their final project. Yeah!
In addition to the guest lectures I’m having in The Platform, I’ll also be reviewing prerequisites for the classes many of them will be taking in the Computer Science department in the fall, so for example linear algebra, calculus, and basic statistics. I just bought them all a copy of How to Lie with Statistics as well as The Cartoon Guide to Statistics, both of which I adore. I’m also making them aware of Statistics Done Wrong, which is online. I am also considering The Cartoon Guide to Calculus, which I have but I haven’t read yet.
Keep an eye out for some of their amazing projects! I’ll definitely blog about them once they’re up.