Visionary Pam, Algorithms and Housing -- Talking Tech w/ Pam Samuelson & Natasha Duarte
Host Brian Wesolowski sits down with the visionary Pam Samuelson, one of the leaders behind technology law and policy clinics. These clinics are helping answer some of today’s most pressing legal questions around tech and society, and Pam tells us more about how they got started and what they're tackling now.<br><br>Then, Brian brings in CDT's Natasha Duarte to talk about changes the U.S. Department of Housing and Urban Development wants to make to its rules against housing discrimination. HUD wants to make it harder to challenge housing discrimination, especially when it’s caused by algorithms. <br><br>More on CDT's HUD comments: http://bit.ly/CDTHUDalgo<br><br>More on Pam: https://twitter.com/PamelaSamuelson<br>More on Natasha: https://twitter.com/Natasha_Duarte<br>More on our host, Brian: bit.ly/cdtbrian<br><br>Attribution: sounds used from Psykophobia, Taira Komori, BenKoning, Zabuhailo, bloomypetal, guitarguy1985, bmusic92, and offthesky of freesound.org.
Top Keywords
- housing 0.018
- disparate impact 0.014
- disparate 0.012
- clinic 0.011
- anti circumvention 0.010
- circumvention 0.010
- housing discrimination 0.010
- clinics 0.010
- discrimination 0.009
- discriminatory 0.006
- fair housing 0.006
- impact 0.006
Transcript
Welcome to Tech Talk. Bye.
CT. Tea. Welcome to CDT's Tech Talk, where we dish on tech and Internet policy while also explaining what these policies mean to our daily lives. I'm Brian Wasilowski, and it's time to talk tech. We have an amazing episode for you today. First, we chat with the iconic Pam Samuelson, one of the visionaries behind technology law and policy clinics. These clinics are helping answer some of today's most pressing legal questions around tech and society. PEM tells us more about their incredible work. And then after that, we hear from CDT policy counsel Natasha Duarte about changes that the US Department of Housing and Urban Development want to make to its rules against housing discrimination. HUD wants to make it harder to challenge housing discrimination, especially when it's caused by algorithms. Not surprisingly, CDT strongly disagrees. Across the country, there are technology, law, and policy clinics at a number of universities that work with individuals, nonprofits, and consumer groups to address some of the most difficult legal challenges that have emerged because of new technologies. From issues of online speech to privacy to intellectual property, these clinics often provide direct representation on some of the most important civil liberties challenges of today. One of the pioneers of technology law and policy clinics is Pam Samuelson, who also happens to be a CDT fellow. Pam joins us today to talk about the work of these clinics, a number of which bear her name. Welcome, Pam. So glad to be here. So first, thank you for being a CDT fellow all these years. What is it that you love most about CDT?
So I think the wide range of technology, law, and public policy issues that CDT, covers is really excellent. And I think, they articulate, a point of view about these issues in a way that actually is very good for people to be able to follow and also that's very well informed and very accurate and really point in the right direction, for public policy makers. And, they've been so effective, over the years in pressing forward on issues from election security to Internet architecture, net neutrality, free speech, and digital copyright. That's a lot of material to cover, and yet to do excellent in all of it, I think, is really wonderful.
I'm so glad I asked that question, and thank you for indulging me in it. So the first real question for you, those of our listeners who are not really familiar with, what a technology law and policy clinic I mean, you essentially created them. What are they? What do they do?
So, the idea is to treat the law students, who enroll in these clinical programs, which usually have somewhere in the neighborhood of four to six units of credit associated with them. They become like a little mini law firm, and the clinical professor basically supervises them the way that, a senior partner would, basically supervise the work of associates at a law firm. But instead of working for a big company or for some, well financed individuals, we look for, individuals non profit organizations that have some stake in technology, law, and public policy issues, but really don't have the resources to be able to get good representation. So the students learn how to be good lawyers. They learn about cutting edge technology issues, and they learn how to represent the public interest in technology law and to think hard about what that actually means. And they've produced amicus curiae briefs, in cases, testimony for, people who want to appear before state or national legislators, and white papers on issues that probably provide a point of view that the big companies that have some some, stake in the outcome would not actually, be able to, to say in the kind of way that
the the clinic students are able to do. That's awesome. That is awesome. And we've at CEDT been lucky enough to partner with some of the clinics, and so many of of them do such wonderful work. These are now all over the country. What are just some of the places that have, law and policy clinics, tech law and policy clinics?
So I would say, you know, when my husband and I started thinking about, these, technology clinics, we wanted there to be one in the ninth circuit. And so that's why we started one at Berkeley. Makes sense. Because a lot of because a lot of intellectual property and other technology, policy issues come before the courts, in the ninth circuit. We also wanted there to be one in the District Of Columbia, and American University put together a really great proposal, so that we were able to, fund a clinic, at American University. We also wanted one in the second circuit, and we started one at Fordham University. Fortunately, NYU, has seemed to wanna compete with Berkeley and other schools. So they have one now too. So there's now, two of them in, in the second circuit. And Stanford and USC have joined Berkeley in having a clinic like this at, in the ninth circuit. Harvard has one. Boston University has one. And, of course, Colorado has one, which also is something that we, my cousin and I have been, supportive of. But we're really excited that not only was this a good idea that we thought of, but other people think it's a a good idea. And now, there are opportunities for students to basically
think about the public interest in technology law across the country. That's really incredible. I didn't I didn't I knew there was a bunch of them, but that's even a more expansive network than I fully realized. So bravo on helping inspire, you know, not just just create, but also inspire the creation of other ones. So your clinic at Berkeley, that was the first one. Yes? I believe so. Yes. Uh-huh. And what were what are some of the issues that you, you know, kinda started working on initially, and what are some that you're working on now just to kinda give people a sense of the type of, complex tech issues you're addressing?
So I would say some of the early cases that we were involved in were some of the cases. So the MGM versus Procter case, was one that, the the clinic at Berkeley, did an amicus brief, to the Supreme Court. We were mostly concerned with, supporting the idea, of preserving the safe harbor for technologies that have substantial non infringing uses. MGM was trying to overturn the Sony Betamax ruling from 1984, which created that safe harbor, and we thought that that actually was a good law, and that the Supreme Court should preserve it. So that's one I was really very proud of. Another one where the the tech clinic, did a fantastic job, was in the case, Chamberlain versus Skylink where, Skylink, got sued by Chamberlain for making, a competing garage door opener. And Chamberlain said that violated the anti circumvention laws. And, our clinic basically represented, the Consumers Union and even got, some time to do oral argument Wow. Before the Court of Appeals and, help to persuade the Court of Appeals, that, it wasn't a violation of the anti circumvention laws to basically competing garage door opener. So those are a couple of our kind of older signature, cases, and we're continuing to do, a lot of work on amici, briefs and also, on privacy guidelines. So the city of Oakland is, considering adopting a set of privacy guidelines, for surveillance that, the clinic students actually relatively recently, have, promulgated.
That's amazing. So you actually talked a bit about anti circumvention laws. I mean, part of the reason we're chatting right now is, Deirdre Mulligan, a former CDT board chair and actually one of the founders of this organization, shared with me a little bit of information about a case that, a number of clinics kind of worked on over the years or helped lay the foundation for, and it's specifically related to computer science professor Matt Green and his challenge to section, 12 o one of the Digital Millennium Copyright Act, which I believe has to do a bit with circumvention. And in this case, security researchers, wanting to take a look at some technology. Can you give me a bit of an overview of that case?
Sure. So, Matt Green is, a computer security researcher, and he's developed some computer programs that he uses in his research. And he even though he thinks he has a first amendment right to write the programs that, that he, is concerned about, to do the research and also to, apply, his program to certain technical protection measures, to test their security. He's worried that because section 12 o one, anti circumvention rules are so strict and, the computer security research exception, that's in the statute is so narrow, he is afraid that if he goes ahead and does the research that he has as his agenda and discloses what he learned, either to professional colleagues at a conference, or, publishes, the results of his work in, a computing professional's journal that he might be subject to liability, even criminal liability, for violating section 12 o one. So he brought a challenge, EFS now representing, him, to, 12 o one as, a violation of his first amendment right, to use his own expression to investigate and research, study, analyze, and then disclose what he learned. And, he made a challenge, what's called a facial challenge, basically saying, gee. The the law is over broad as written because it applies in certain ways that, in many ways that actually are, are, threatening to free speech. And I also made an as applied challenge. So an as applied challenge is basically saying the specific thing that I'm doing is first amendment protected, and so the anti circumvention clause ought not to apply to me. And, so the government, basically tried to dismiss the whole case, have it thrown out entirely saying he doesn't have standing to bring a lawsuit like this. And besides which, the the anti circumvention laws are are, are completely constitutional. And, it was a big victory that although, the facial challenge was not, was not upheld, that the district court decided that he did have standing to bring a lawsuit of this sort and his applied challenge, as applied challenge, at least as it was alleged in the, the, in the complaint that has sufficient merit that the case can't be dismissed at this point. So it's not a final ruling, obviously, on the merits, but, it does mean that, that, Matt Green is gonna be able to, proceed with this lawsuit, and to get the court to take seriously that sometimes the anti circumvention laws really do, squelch free speech and freedom of expression. And that's something that the First Amendment says you're not supposed to do.
And I would imagine some a lot of the research that he's doing is also more broadly in the public interest when you think about, you know, his free speech rights, but also security research in general. Most of it is aimed at helping technology be more secure, which would be a net positive for us. Is that an argument that's ever made? And those are just not one that he would make in this case.
Well, I think, that is certainly part of his story, which is that I'm not trying to help somebody engage in any kind of copyright infringement. I'm trying to make sure that systems are secure, and you can't find out that systems are secure or insecure unless you, in fact, are able to test the security of them. And that means that at least as, on its face, the anti circumvention rules seem to make that really a risky thing to do because there's no direct link in the in the statute to, enabling copyright infringement. So, he's certainly trying to say, I'm just trying to make the system better. Now the argument that computer security researchers need to be able to, circumvent, some of these technical protection measures in order to make systems more secure is certainly part of what, the set of submissions that various of the clinics, the Berkeley Clinic, the Colorado Clinic, and the Harvard Clinic, have, joined since, the year 2000. There have been multiple submissions to the copyright office, which holds the kind of what's called a triennial review of the anti circumvention, rules. And this triennial review gives an opportunity to anybody who thinks that the the anti circumvention rules are overbroad as applied to some activity that they think that they should legitimately be able to do. They can basically ask for an exception, and that exception will be valid for a three year period. Okay. So the Berkeley Clinic represented Alex Haldeman and Ed Belton, going back in 2000, 2003, and then 2006. And in 2006, the first of the of the, exceptions for computer security research, was granted after the Sony rootkit scandal, kind of rocked the world, of the computer security research community. Thereafter, the Colorado Clinic, Blake Reed, who was then a clinic student in 2010 and, later, became the head of the clinic, himself.
Oh, very cool. To
represent, the, the Haldeman and other researchers, for expansion of computer security exemptions. So, the clinics have definitely had a major role in basically educating the copyright office about the importance of computer security, research being able to happen so that our systems become more secure over time rather than being vulnerable in ways that people can't find out about because the anti circumvention rules basically killing effect on that reason.
So this is really a story of, you know, not just a a positive, you know, outcome in the case with Matt Green, not final decision, but progress on it. It's really about two decades worth of work by these clinics to kind of lay a foundation and and build to this moment.
Absolutely right.
That's really remarkable. Well, Pam, thank you so much for joining us on Tech Talk. It's a great illustration. I'm glad you shared that story. I didn't even really have to ask the question. You just went into it, which is amazing. Keep up the great work you're doing, and we'll have to have you on again sometime soon.
Okay. Thanks very much, Brian.
The US Department of Housing and Urban Development is proposing a major rule change that would weaken housing discrimination protections and puts blind faith in algorithms to be fair and equitable. Seriously. CDC filed comments with HUD opposing these proposed rule changes. Policy counsel, Natasha Duarte, joins us today to talk about what we said and why. Welcome, Natasha.
Thank you.
So top level, what are the rules that HUD is proposing changes to, and why are these proposed changes so concerning?
So HUD is trying to change the rules for challenging housing discrimination under what's called the disparate impact standard. Under the Fair Housing Act, it's illegal to deny housing to someone based on their race, national origin, religion, sex, familial status, or disability. But it doesn't just protect against intentional discrimination. It also protects against policies and practices that have the appearance of being neutral but result in housing discrimination or segregated housing. So what HUD is proposing is to make it more difficult to challenge housing discrimination based on that standard, what's known as the disparate impact standard. Okay. So for people who experience housing discrimination and want to sue under the disparate impact theory, the new rule would create extremely high hurdles for them to jump through, to be able to have their cases heard in court. And in fact, housing attorneys we've talked to, who work on these cases for a living had set have said these new hurdles would have made it impossible to bring the very landmark cases that have shaped what it means to discriminate in housing. And in particular,
the proposal would prevent people from challenging housing discrimination when it's caused by using an algorithm. Okay. So tell me a little bit more about disparate impact just because that's a little bit difficult of a concept to wrap your head around. What does that look like in the real world? Yeah.
So the types of things that disparate impact protects against are, for example, if a city passed an ordinance that limited every home in a particular neighborhood to three occupants or fewer, which would have a disproportionate which would disproportionately prevent if a building whose occupants were already mostly white maintained that racial imbalance by requiring any new tenants to have references from current residents. Or a zoning ordinance that restricted the building of new affordable housing developments to neighborhoods that were mostly occupied by people of color, thereby perpetuating or deepening racial segregation among neighborhoods in that city. So things that are not where you can't point to a specific, intent to deny someone housing based on their race or sex or national origin, etcetera, but where the effect of the policy
is to do that. Okay. So a little bit more complicated to prove, but still, you know, huge impacts then. Yeah. I mean, this has been an extremely
important tool in preventing and challenging housing discrimination for, many decades, because, you know, where people live, is not an accident. We still have segregation in our cities and neighborhoods that's a result of, a legacy of, racism and segregationist policies in The US. And so, a lot of these practices can, perpetuate and exacerbate those patterns. And the Department of Housing and Urban Development, which is the agency that enforces against housing discrimination, is not just there to stop people from explicitly intentionally discriminating. It's also there to, they're tasked with ending, housing segregation.
Alright. So how would these new rules exactly then make it more difficult, for disparate impact claims to move forward?
Yeah. So under the current disparate impact standard, which HUD and the federal courts have recognized that disparate impacts are harms that the Fair Housing Act, can remedy, for decades, and HUD specifically codified the standard in 2013, in a rule, by the standard by which, disparate impact cases are litigated. So, there are basically, three parts to a disparate impact case. First, when someone believes that a practice is causing a disparate impact, they can file a complaint that alleges what the policy or practice is and that shows that a disparate impact exists, or is predictably likely to result from that practice. So often this is shown as a statistical disparity, but it doesn't have to be. And when a plaintiff shows that a disparate impact exists, then the burden shifts to the defendant, the person who's being accused of discriminating, to show that the practice they're engaging in or the policy that's being challenged is necessary to accomplish a legitimate business purpose that is nondiscriminatory. And if they do that, then the plaintiff has a chance to show that actually there are other less discriminatory means of accomplishing that business purpose. Okay. So while all this is happening, the parties are able to, engage in discovery. So, the plaintiffs can request documents to learn more about the defendant's policy or practice to be able to build their case that it is, in fact, causing this disparate impact, and they can, you know, the parties and the courts can look into whether this really is being done for a nondiscriminatory purpose such as making sure that neighborhoods are clean, for example, you know, other types of city business, and that it's not in fact just, a way to discriminate. What hud is proposing is to put incredibly high burdens on the plaintiff at step one when they are first trying to make out their case. So for example, under the proposal, a plaintiff would have to show that, there is a robust causal link between the practice and the discriminatory impact that they are alleging. These kinds of things are would be very difficult, if not impossible, for someone to show without being able to engage in the discovery I was talking about to do fact finding and to actually, you know, figure out what's going on. And so the plaintiff would have to show all of these things to be able to even have their case heard. And so it would cut off a lot of, of alleged discrimination before, courts or HUD ever gets to look into it. And so as I mentioned, you know, this would prevent a lot of cases from going forward that have been sort of bedrock cases in establishing, what housing discrimination means. Some of the things I was talking about before, like zoning ordinances that keep families out of neighborhoods.
So algorithms play a role in this too. How did those jump into this equation or this proposed new equation?
Right. So there is a part of this rule that focuses on, when the alleged discrimination is being caused by an algorithm. So if a plaintiff is saying that there's a disparate impact and that case relies on an algorithm. So, for example, someone is using an algorithm to screen their tenants to decide who, is eligible to live in a particular building. There are three defenses that HUD has created, and these defenses can be used to have the case dismissed again before it ever, gets heard in in court. And those three defenses are, one, the defendant can show that none of the inputs in their algorithm are close proxies for protected classes. So close proxies, for things like race, sex, familial status, disability. Two, they can show that, the algorithm was not created by them, that it was created by a third party, or a third party. Enough? Yeah. Someone else created it? Right. So, basically, any algorithm that most people use was created by a third party.
Yeah. I've never used one I've created but Right. Okay.
And the third defense is that a third party, analyzed the algorithm and found that it is basically valid, that it is, a valid predictor of something that is, you know, related to a business purpose. Mhmm. So, that it is predictive of credit risk, for example.
So it sounds like, you know, just to me, this is editorializing a bit, that these are pretty garbage proposed rules. What what exactly would be the reason for this? What is, HUD trying to accomplish or what do they say their goal is with these changes? Yeah. It's not immediately clear in HUD's proposed rule.
These defenses appear to be coming kind of out of nowhere. HUD does not really attempt to explain them in the rule or just they are, you know, the language used in the defenses is not something that is established in law, really. And so this is important because, you know, algorithms are increasingly being able being used to make decisions that determine whether housing is available to people or at what cost it's available. So these are areas, like I mentioned, tenant screening. So, there are lots of services out there that any landlord can sign up for and use to do, background checks and other sort of screening on tenants. There are a few that even, look into a tenant's, social media and do a background check based on that and, you know, may score a tenant or give a recommendation to the landlord about whether they should accept that tenant. There are algorithms that are used in, lending for mortgages, mortgage market where, it is determining whether someone can access housing that's covered under the Fair Housing Act and of course in advertising housing opportunities the Fair Housing Act also covers the advertising of housing so if housing is advertised in a way that works to deny the opportunity of housing to people based on a protected class, that's a a fair housing violation. And so, of course, in online advertising, that's all powered by algorithms and algorithmic predictions about who should see an ad and and who it's most relevant to. So these are all areas that, are raised the exact type of risks of disparate impact Mhmm. That HUD is supposed to be protecting against because, you know, as we talk about all the time at CDT, you can have an algorithm that doesn't is not designed explicitly to make distinctions based on race or gender, or income or what have you, but, because it's learning from patterns and data and those patterns are shaped by all kinds of societal forces, including discriminatory ones, those algorithms can unintentionally, exacerbate those discriminatory patterns. And so, you know, it's possible that HUD is trying to protect, certain industries. There are multiple references in the rule to, to lenders and, you know, the insurance industry. It could be that they're trying to provide a safe harbor just across the board, for, you know, the the third party defense in particular points toward, you know, trying to create just generally this safe harbor where people can, use an algorithm that someone else created and not have to, worry about how it was built and whether it's discriminatory, not be, having to ask that third party how their algorithm was designed and whether it was tested for disparate impacts. And those questions are exactly the things that people should be asking. And so not only is this gonna cut off a lot of fair housing litigation, but it's also going to, disincentivize a lot of the,
responsible corporate practices that we are always asking for. Yeah. So I stand by my initial. These are garbage proposed changes. Yeah. So this is why, of course, you filed comments. We weren't the only ones on them. You did a great job of, getting people to sign on and join in our comments. Who are some of the folks that signed on and what perspectives did they bring to this work?
Yeah. So importantly, we, collaborated in writing and filing these comments.
Better word than sign on. Yes. Thank you for correcting that.
With, some, folks who are experts in, machine learning, in fairness in machine learning, in how, algorithms and statistical systems can discriminate. So, you know, we have statisticians. We have computer scientists. And, of course, we have other organizations, civil rights organizations, other tech policy organizations. You know, we've got some great representation from, networks like our our wonderful CDT academic fellows on there. So we're very grateful to have had, all of that collaboration and to be able to bring to HUD, such a robust explanation, and record of why this rule doesn't make sense, from a legal perspective, but also from a technical perspective completely misunderstands
how algorithms work and how they can discriminate. And there was a wave of separate comments too that people filed in addition to the kind of joint ones that we did. What were some of the other organizations that were reaching out or filing comments? Yeah. So there's a really broad range of organizations and individuals
filing.
There Against, of course.
Yeah. I haven't looked at the number in a while, but there are, a lot of comments, filed by individuals. There are, of course, fair housing groups and housing experts, filing comments, people who have who litigate these housing cases, who, again, have argued that, these are going to prevent them from bringing the types of cases that are exactly what the Supreme Court says are the core of disparate impact liability. And, industry actors also, have filed comments. The, BSA, the software alliance, is a group that represents a a large number of software companies, and they filed comments, talking about how these rules aren't actually good for business either. They create more confusion than the the existing disparate impact standard. Under the existing standard, the rule is that you can't have a discriminatory effect. So, businesses can test their systems and, figure out if they are discriminatory or if they're likely to discriminate, and have, you know, some idea of whether they are complying Yeah. With the standard, with civil rights law. With this new standard of looking at the individual inputs to the algorithm and trying to figure out if they're close proxies, that's harder to comply with, and it also, can hinder efforts to actually do affirmative equal opportunity, housing initiatives. So, pretty much everyone who works in, algorithms and machine learning has, has accepted the fact that if you want to actually ensure that your model is not discriminatory, that you can't just ignore, protected or sensitive characteristics like race and gender. You have to actually use them and try to figure out if you are having a discriminatory effect based on those characteristics. And so robust efforts to, make algorithms less discriminatory actually rely can rely on those characteristics. And so if, if under the new standard, how the courts are gonna be scrutinizing not the effects of a system, but what inputs went into it, companies are more worried that their actual efforts to, improve, anti discrimination are are, going to be potentially in jeopardy.
Interesting. So positive takeaway, I guess, a lot of groups and individuals are rallying against these of sending comments. Do you have any hope that HUD will listen?
So, HUD is required to consider the comments that are filed. I think in this proposal, it's particularly clear that many of the things we raised were not considered at all. They there were no real justifications in the rule for, these algorithmic defenses. So HUD has to go back and consider them. So, that that gives me hope, I suppose. I think we have that we have that going for us. And, again, the fact that the opposition to this spans across advocates, academics, and industry, hopefully, that will send a message that really needs to go back to the drawing board. Well, fingers crossed on that and,
brilliant work on those comments. I got to read them. They're very insightful insightful, so folks should take a look at them on our website. Keep up the great work, and thanks for joining us, Natasha. Thank you. That's it for this episode of Tech Talk. If you're interested in technology, law and policy, which of course you are because you're listening, give Pam Samuelson a follow on Twitter. She has an amazing account. And of course, for more from CDT, follow us on Twitter, like us on Facebook,
