In the new episode of the n+1 podcast, David Auerbach discusses his article in Issue 19 about the “Chat Wars” of the late-90s between MSN and AOL. Then Ross Perlin of the Endangered Language Alliance talks about his piece in Issue 19 about endangered languages and the language diversity of New York City. Later on, David Auerbach returns to talk about his article is Issue 13 about the Turing Test and the “Stupidity of Computers.”
Hosted by Elisa Wouk Almino, Moira Donegan, and Eric Wen.
Audio Engineer: Malcolm Donaldson.
Produced by Elisa Wouk Almino, Aaron Braun, Malcolm Donaldson, Moira Donegan, and Eric Wen.
Music from Arthur Verocai, S.E. Rogie, and Cocteau Twins.
Original Music by The Westerlies featuring Andrew Mulherkar, Luke Sellick, and Sammy Miller.
Subscribe to the podcast on iTunes, subscribe via RSS, or download the episode (66.6 MB).
Episode Transcript
Intro
Moira Donegan: Hi, I’m Moira. Welcome to the n+1 podcast. Today I’m going to talk to David Auerbach and Ross Perlin. David wrote a really cool story for n+1’s Issue 19: Real Estate. It’s called “MSN vs. AOL.” David was working at Microsoft in 1999 when Microsoft started developing their online chat technology to compete with AOL’s Instant Messenger and what happened was pretty interesting between the two companies, so here’s David.
Segment 1: Interview with David Auerbach
MD: Hi, I’m Moira Donegan. I’m here with David Auerbach on the n+1 podcast and we’re going to talk to you a little bit about David’s piece in Issue 19: Real Estate, and it’s called “Chat Wars.” So David, tell me about “Chat Wars.”
DA: So I’ve been a software engineer for many years. I worked at Microsoft from about ’97 to 2003 and then Google for about five or six years after that. So back when I was at Microsoft, I was very green. I had actually interned there when I was 20 and then went back after graduation at the age 21. This all happened when I was just a kid, more or less, and it was kind of a crazy time because back then the internet didn’t have anywhere near the sort of penetration and ubiquity that it has today. But AOL was still a titan in those days because it was primary among dial-up services and it still had its sort of gated-community to the internet.
MD: They were sending out all those disks in the mail, you know?
DA: They were, yeah. Millions and millions now all in landfills and one of the things they had was their instant-messaging program, which by far had the biggest number of users at the time because all AOL users were automatically users of their instant-message system and in addition people who weren’t on AOL were downloading their standalone program as well who were on AOL and also who were off AOL. So they had the biggest user base and AOL owned the network. You couldn’t just talk to anyone like you would over email. You had to use their program. So Microsoft was creating its instant-message program, which was called a terrible name—“MSN Messenger Service,” after much cooler and shorter names were considered—and Microsoft wanted to build up its own instant-messenger network. It had purchased Hotmail in the recent past to build up an email network as well. But there was an issue in that because of the network effect in the way you couldn’t start a social network today and automatically be friends with everyone on Facebook on your social network, you couldn’t start up a new instant messaging network and talk to people who were on a different network.
MD: Right, when Microsoft develops their messenger service, are they trying to create their own monopoly or are they trying to hack into what AOL already has?
DA: The goal was to create their own service, to create a competing service, but because of network effects, it wasn’t just a matter of, “Oh, we built a better mouse trap. Just switch over!” because all your friends were already using the old mouse trap and because of the network effect, it’s like saying, “Oh, here’s a social network that’s superior to Facebook. Why don’t you switch over to it?” Well-
MD: It’s not superior if all your friends are on-
DA: Right. Even if it is perfect, even if it is far superior to Facebook, all your friends are on Facebook right now. So unless you move en masse, which doesn’t happen, you’re sort of stuck and companies love to leverage this. This was at least one of the reasons why Google+ got no traction. There were other reasons, but certainly being a late entrant to so-called “sticky social apps” sets you further behind than being a late entrant to not-sticky or not-social apps. So one of the features that it released with was indeed interoperability with AOL, which was to say that the Microsoft MSN Messenger client would talk both to Microsoft-owned services and to AOL’s. Now it was required, of course, that you have one of the Microsoft accounts, but you could also talk to your friends on AOL. So this was sort of a breaching mechanism with the hope that maybe people will start using Microsoft’s service, convert their accounts, and because they can still talk to their friends on AOL, well there’s not a downside to switching clients and over time if people start switching to the client, they’ll all have Microsoft accounts and they won’t need to log on to AOL at all. So because of the protocol the AOL used for the server was private, we, specifically my old mentor and I, looked at what their client was sending to the server, how it was dealing with it, and just sort of figured out what was going on. “Oh, OK, this message-”—so there’s a bunch of just binary bits and we looked at it and said, “Oh, OK, this is what it sends during the log-on process, when you send an instant message to someone, this is how it packages it up and sends it across the wire.” We managed to get it so that the messenger client could basically mimic the AOL clients efficiently, that it could log on and see your buddies with your AOL accounts, so it would effectively talk to two servers: both Microsoft’s and AOL’s. And this was initially just a lark, just sort of, “let’s see how our client handles talking to a different server.” Because while we were testing, sometimes the servers were down, so it was, “Well, let’s try talking to AOL and see what happens.” But somewhere along the line I had the bright idea of just putting it into the finished product. And I assumed Microsoft would say, “Oh yeah, look. Quit the funny business.” But I guess there was enough of a mischievous or profiteering streak they said, “This is pretty cool! Let’s do it!” And so the initial release of the instant-message client actually did talk to both AOL and MSN servers and when you logged on you would see both your buddies from MSN and your buddies from AOL. Great, you no longer needed to use the AOL client. What happened was that AOL did not like this. They saw it as a real threat. And with good reason, but certainly there wasn’t anything illegal about what we were doing, or at least I do not think-
[Laughs]
MD: How certain are you David?
DA: Well, I think that’s for the courts to decide. But I think AOL certainly wasn’t in the mood to wait for any sort of legal action. They wanted to stop it right now before they started hemorrhaging users. And so one day we came in and the MSN client would connect to the AOL servers and the user would get a message saying, “You are not using an authorized AOL client. Use the superior AOL client here.” And it would send the link. It turned out AOL was looking at the differences between what our client did and what their authorized client did, figuring out when it was our client, and sending this message and disconnecting people.
MD: So you had to disguise your clients?
DA: So what happened was we needed to make our client look as close to the AOL client as possible, because all AOL could do was see whatever was coming across the wire. They couldn’t look at the user’s computer. So there was an iterative process where we would, this was me, I would come in, try to figure out how AOL was distinguishing the client, and try to find any difference whatsoever. Like at one point it turned out the only difference was that AOL downloaded a ton of advertising, which was then showed to the user. So I said, “OK, let’s have our client download the advertising and toss it away, because we don’t need to show it to people, we just need to download it, because that way they would know that it’s us.” So we got to the point where our client was very, very close to AOL’s, to the point that I couldn’t find any remaining differences between how we were acting towards their server and how their own client was acting. And there was a break: suddenly what had happened—you know, usually they broke our client, we would fix it, and then we would release a new version, which would then work. And then there would be a little pause and they’d break it. But this time there was quite a long pause and we were all excited. I was excited saying, “Have we beat them? What’s going on?” There was a mysterious silence. By which point the New York Times had been covering it, CNN had been covering it, talking about how weird this was that AOL claimed that we were doing something unauthorized and yet they actually had to use technical means to prevent us from doing it. They weren’t pursuing legal action, trying to do anything else. It was actually a technical war. And that was awesome from the standpoint of a programmer, from the standpoint of a software engineer. This was tremendously exciting fighting this war.
MD: And they weren’t suing because they weren’t sure it was actionable or they weren’t suing because they thought it would be too slow?
DA: I think both. I think both. They didn’t say anything. They didn’t make legal threats. They said it was bad and that we shouldn’t be doing it and they tried to play the moral high road, which you know you’re losing when you have to play the high-road card. And it was also nice for, you know, Microsoft didn’t have the greatest reputation in the technical community for obvious reasons, but it was nice that people were saying, “Well, you know, Microsoft is closed and evil and whatever, but this time they’re right!”
[Laughs]
DA: Which was sort of perversely gratifying in a way. But anyway, so we got to the point where like a week elapsed without us breaking it—without AOL breaking our client—and we were all very excited saying, “Have we won? What happens now?” And then abruptly it broke again. And I came in one morning, just people saying, “It’s broken again. Take a look at it.” And I saw that our client was getting disconnected. I couldn’t see exactly why, but I looked at what their client did and there was some serious gobbledygook going on. This was something I’d never seen before. It was completely unlike any sort of protocol that had been exchanged between the AOL client and the server before. And I was baffled and I spent quite a while staring at it and I tried just mimicking this with our client and in spite of it, they changed it a little and it broke again. But I couldn’t understand what was going on. And finally a really sharp engineer on the team took a look at it and said, “Oh, they’re hacking their own client.” So what was happening was that they were taking advantage of a bug of a security hole in their client that allowed them to execute arbitrary code on the client and say, “Here are some instructions to execute on the computer.” Not within the program, but on the whole computer, do this and send back the response. So they were basically hacking the computer. It was akin to installing malware or spyware. Now AOL wasn’t going to do anything too malevolent, but they were doing something that was at least a little unorthodox, sketchy, and certainly brilliant from our perspective. This was certainly a brilliant tactic to use because our client didn’t have a security leak. This was a security hole that was specific to their client, specific to their program. We couldn’t mimic that, at least, not just with our program. And we were very frustrated. And AOL was doing other things to drive us crazy as well, like they did it so that any MSN client that connected to their servers from inside of Microsoft’s corporate headquarters would not be cutoff. You actually had to access it from outside of Microsoft. This was very puzzling because one day we came in and we said, “Everything looks fine.” And they were saying, “No, no, I’m getting disconnected.” And it took us an hour or two to realize, “Oh, they’re filtering by IP so that they’re making it look like everything’s fine and dandy if you’re at Microsoft! Sneaky, sneaky bastards!” We tried another thing, which was the security hole they were exploiting was only in the PC client and we said, “OK, well, how about the Macintosh client? Let’s just pretend to be the Macintosh client!” So we did that and they immediately actually deprecated all their Macintosh clients forcing all their Mac users to upgrade, which we didn’t think they would do, because that’s a big deal back in 1999. It’s not big deal to download an upgrade today, but back then, you’ve got dial up; it takes a while to upgrade. So they were hell-bent on stopping us. And at that point we looked at the security leak and we said, “How can we get around this?” And we came up with some ideas and they were very intricate and I get into them a bit in the article and on the Reddit thread that popped up around this article saying, “OK, well, we considered this.” But the decision was made higher up to say, “No, this is too much work.” I’d pretty much been the only engineer working on this, so from Microsoft’s standpoint this wasn’t costing a lot of resources, but at this point it was like, “OK, we might need more people to do this.” And it sounded like people just weren’t keen on dumping more resources into it. So the decision was made, “OK, we aren’t going to try to get around this technically.” Many years later I met someone from AOL who had moved on just like me and he had been there at the same time on the opposite side and I said, “Well, good game. Congratulations.” [Laughs] And he said, “Yeah, it was crazy. It was a crazy time.” And I said, “It really was.” It was just beautiful nostalgia and we sort of both reflected on the fact that no one outside the technical community really had any idea what was actually going on. [Laughs] And that it was so strange to see these two levels going on—the technical story, which nobody knew about, and how it was reflected in the national press.
Segment 2: Interview with Ross Perlin
MD: That was David Auerbach, author of “MSN vs. AOL: Chat Wars” in n+1’s Issue 19. Next up I’m going to talk to Ross Perlin. Ross is a writer. His book Intern Nation did really well. He is also a linguist and he works with a really cool organization in New York called the Endangered Language Alliance (ELA) and Ross wrote an interesting piece in n+1’s Issue 19 about his work with the ELA and he sat down to talk to me a little bit about that.
MD: So I’m here with Ross Perlin who wrote an amazing article in n+1’s Issue 19 called “Endangered Speakers” about his work with the Endangered Languages Alliance in New York City. Ross, thank you so much for being here.
RP: Thanks for having me, Moira. I am a writer and a linguist and I’ve been working as the assistant director of the Endangered Languages Alliance here in New York for almost the past two years. It was founded just a few years before that by two linguists and a poet and basically came out of the recognition that New York City right now is probably the single-most linguistically culturally diverse place on the planet. Perhaps in the history of the planet. We estimate that there may be as many as eight hundred languages spoken in the New York area. Numbers are always provisional and difficult, but that’s just kind of an indication of the deep, deep diversity that’s here. That really goes beyond the kind of—of course people recognize that there are several million Spanish speakers, a huge Chinese-speaking population, of course people know those large groups, but I think there’s relatively little recognition of these much smaller languages that are represented here in New York that are indeed endangered both here in New York certainly and also in their homelands where they’re originally spoken. And here in New York they are represented by immigrants, refugees, all kinds of people who’ve come to the city for whatever reasons and it’s our attempt both to recognize something particular about New York and about these hyper-diverse megalopolises of today, but also a kind of response to this larger crisis of language endangerment that’s unfolding across the world.
MD: So the Endangered Language Alliance was nice enough to let us use some recordings of rare and endangered languages that they’ve captured here in New York and we’re going to let you listen to a couple of those, starting with a story told in Garifuna, an indigenous language from what’s now Honduras.
[Garifuna spoken.]
MD: This is Acehnese, an Indonesian language.
[Acehnese spoken.]
MD: This is Breton. This is a French language.
[Breton spoken.]
MD: And this is Tontemboen, an Indonesian language.
[Tontemboen spoken.]
MD: How can people donate to the ELA if they want to?
RP: We have a big “Donate” button on our website, elalliance.org. It’s a shoestring operation and we’re always happy for donations, tips, and volunteer effort.
MD: Ross, thanks so much for being here.
RP: Thank you very much, Moira.
Segment 3: Interview with David Auerbach Part II
MD: So I talked to David Auerbach in our first segment, and David also wrote a piece in n+1’s Issue 13 called “The Stupidity of Computers” where he talked about some of the things that computers do well and some of the things that they don’t do very well and he gave me one example of a chat program that attempted to convince users that it was a human and here is our conversation about that.
MD: To wrap it up, do you want to tell us a little about “The Stupidity of Computers”? Your piece from Issue 13.
DA: “The Stupidity of Computers?” Oh God. Yeah, so the stupidity of computers I think is hard to grasp if you haven’t been on the inside. Just sort of what it is that computers do well and what they don’t do well. It’s clear that computers can analyze, process, and calculate at a speed unfathomably beyond what humans can do, and yet there are some problems that you might think would be easy that they haven’t done. The most famous of these being the Turing Test, which was Alan Turing’s challenge in 1950 to have a computer that could talk to someone over instant-message style and convincingly pretend to be a human, such that a person wouldn’t be able to reliably tell whether they were talking to a human or a program over an instant-message window. And this has not been substantively passed in any meaningful way. Turing predicted that it would be achieved by 2000. He wasn’t even close. And…I don’t know, well, maybe it’s possible that everyone says, “Oh, well of course I know why that is.” But at least from my perspective, I think it’s an interesting question. I think it’s something that a lot of people don’t sort of intuitively grasp and it has to do with the sort of data that computers process and one of the things that I get into a lot is computers are great with explicit taxonomized concretely represented data and they are terrible with nuance, ambiguity, incompleteness, and all those things. And those are the sorts of things you find in human language and human interaction. This immense degree of imperfection and heuristic thinking and the lack of ability to specify things. And if you compare human language to computer languages, they barely deserve to be called the same thing because computer languages are so simple and elegant and human languages are so messy, exception-ridden, and overloaded. You know, if you take a word such as “get” or “as” or “like,” try defining that in a complete way and try specifying it in a complete enough way for a computer. It cannot easily be done; possibly it can’t be done in any sort of explicit way. And these are the sorts of issues I talk about. And yet the most successful programs at passing the Turing Test—the ones that have come the closest, and they haven’t come close—but the most successful ones have not in fact been ones that were most algorithmically sophisticated. There was one program called MGonz that managed to convince a user on a university internet that it was a human at least for a while. And a programmer at the university had set it up as a chat bot and left it running on his account for anyone to talk to. And it wasn’t—it was actually an incredibly primitive program. So you may ask, “How on Earth was this program able to do better than the most sophisticated natural language generators?” Well, this program was incredibly abusive, and-
MD: It was written to speak like a jerk, right?
DA: Yeah. So, actually, would you like to-
MD: Would you like me to play a role?
DA: Yeah, you can be the program called MGonz and I’ll be the guy who thought he was talking to a real person.
MD: OK, so MGonz: “ When was the last time you had sex?”
DA: “Yesterday!”
MD: “OK, honestly, when was the last time you got laid?”
DA: “I told you! It was yesterday!”
MD: “You are obviously an asshole. OK, honestly, when was the last time you got laid?”
DA: “OK, OK, it was over twenty-four hours ago. For you it must’ve been twenty years!”
[Laughs]
DA: And really all the thing ever did was keep spitting back questions, assaults, insults. Like, you could hear even in that small excerpt it was repeating itself.
MD: It was like two or three things.
DA: Yeah. And the author said that it succeeded in convincing this person for an hour and he never did figure it out. The techniques were quote: “profanity, aggression, fervid queries about the user, and implying he was a liar when he made responses to these.”
[Laughs]
MD: Which is the way that real humans behave.
[Laughs]
DA: Well, and I think that tells you something about exactly what it is that computers don’t quite get and perhaps indeed what computer scientists often don’t get.
[Laughs]
DA: Nonetheless, to be serious about it I think the thing is that there is so much in nuance and tone in human language that you aren’t just building something that understands human language. You have to build something that has something like what we would call personality or human thinking, the conception of thinking. And that’s very difficult. So that is the way in which I think computers are stupid and the other thing I get into in the piece is how because computers have become so ubiquitous, the computers are now sort of shunting us to act more like them and their stupidity is sort of becoming our stupidity.
Outro
MD: Thanks for listening! The n+1 podcast is produced by Malcolm Donaldson, Moira Donegan, Eric Wen, and Elisa Wouk Almino. Special thanks to Carla Blumenkranz, Dayna Tortorici, David Auerbach, and Ross Perlin. Thanks!
