The Microsoft Tay Debacle

I'm sorry. You've been a radio and the radio is forever. No, no, no, no, I apologize. Easy jokes beneath me. We're going to talk about chat bots today. Yeah, we've talked about chat bots before, right, We've talked about them in a couple of different episodes. We've talked about chat bots, we talked about AI, We've talked about touring tests, we've talked about machine learning. All of those topics are going to play a part in our discussion today. And speaking of easy jokes that are

beneath us, we are talking specifically today about tay tweets. Yes, Microsoft ta uh yeah, Tay Tay. We we we hardly knew ye. In fact, I doubt anyone really knew. Ye. I didn't find out about Tay until Tay was already gone. I found out about Tay while TAY was still active, but I never messaged Tay. And the only reason I found out about Tay was because I had been invited to be a guest on a show called Tech News Today, and that was one of the news items that they covered.

And I had been very busy in our day to day work and had not followed along. So when I when I heard about it, it it was before all the craziness unfolded. And we'll get into what the craziness was. To be fair, though, even at ground like square one, it seems to me like TAY was probably not the most thought out project Microsoft has ever launched. And I'm including Microsoft Bob in that. Yeah, that's a that's a

sick burn for those who know what Microsoft Bob was. Okay, tell me more, all right, Well now that we we got our little rascals reference all the way. So, according to Microsoft, Tay is quote an artificial intelligent chatbot developed by Microsoft Technology and Research and BING teams to experiment

with and conduct research on conversational understanding. UM, I know some people would probably start to argue with the word intelligent that was in that description based upon some of the things that were happening with t Now, that description doesn't exactly capture what the public saw of Tape because the public facing incarnation of ta was specifically a Twitter account. Well, Twitter account was the primary one. Yeah, she or it.

I get very confused about referring to AI because they're robots. Robots don't have genders. But Tay was impersonating a nineteen year old girl, so our woman. Yeah. Also when when at what age is an AI a woman? I don't know yet? An AI no longer the longer a girl type of thing. She was a young adult AI. Okay, while I continue to go through this this oddly misplaced feeling of cultural shame, Yes, she was also on group me and Kick and Kick So what are what are

you even talking? So group me, Lauren and I know all about group me. Yeah, we happen to use group me while we were in Austin for south By Southwest. So group me is essentially like a very contained version of Twitter. It's it's a collection of cell phone numbers, essentially, so that you can send a message to the group me which includes multiple people. Everyone receives that message and

everyone can respond to it. Yeah, it's a fancy group messaging system that also has a little bit of capacity to look like cute stuff, like you can up a picture and um immediately place meme style text on it. Yeah, it used to be that it was just group texting. So when I started seeing the pictures popping up with all the memes, I immediately went into old man mode

and got really confused. Kick is also an instant messaging platform. Uh, it's based off the BlackBerry Messenger platform, So it's all. Three of these are different ways to send out messages to typically groups of people, not not like a one on one thing necessarily, although I think Kick can can sometimes fall into that category. But um, honestly, before we

even get into two, more of the weirdness here. From the get go, I felt odd about the idea of framing an AI within the context of not only a gender but an age, kind of like you were saying, Lauren, I I most of the time when I think of AI, unless you're specifically trying to create something that's going to pass a touring test and therefore fool someone into thinking it's an actual first I was gonna say, this sounds

like some Eugene kind of trickery. Yeah, and we'll talk about goosetman in a minute to Unless you're doing that, there's not really I don't see a purpose to giving a gender and age to an AI other than maybe there's some studies perhaps that suggest humans are more likely to have a natural interaction with someone that they can kind of kind of categorize in some way. And I don't know if that, if that research even exists. That's

just me taking wild shots in the dark. Um, but I personally thought I would be very reticent to send a message to an entity that had been more or less identified as a nineteen year old female, even just to see how it was working, because I'm a forty year old man, and I don't think sending messages to to an entity that's less than half my age, like it just feels inappropriate from the from the start to me, even though I know it's not a person in right,

it all exactly made me feel kind of squidgy. It well, sure, it's just putting putting a strange basis from from our point of view, perhaps, I mean, and and the boat was supposedly geared towards people I think eighteen to twenty four. So so it wasn't meant for me in the first place, but it wasn't designed by people who are eighteen to twenty four. It's fairly I think it was designed by people closer to my age, and that that is one of the many problems that Tay has. So yes, slightly

odd thing, but perhaps a conversation for another day. Yeah, So, at any rate, Tay, like you said, eighteen to twenty four year olds, that was the target demographic because Microsoft had identified that as the demographic that most frequently uses mobile based social media things like social messengers and um and that's really where they were looking to kind of increase this ability for AI to understand conversational English things,

things that go beyond very simple, uh prescribed commands. Okay, so in practice, what does this look like? What exactly was Ta doing? So in practice, you would send Tay a message on one of these platforms, either Group Me, Kick or Twitter, and then t would respond to you. If it was on Twitter and things were particularly busy, Tay would tell you to send a direct message because

otherwise we just get kind of lost in the crush. Obviously, if you unleash something like this to the world at large and you let it be known that it's out there, you're gonna get a huge response, right, especially from Twitter, so mostly consisting of mischief and I'm sure we'll talk about that some mortally. Yeah. So, uh yeah, this was

you know, that was exactly the case. So they released it to try and see what would happen, how Tay would start to incorporate different types of language in her replies, how she would be able to interpret different things that were said to her. They started with a baseline where Microsoft had essentially mined a ton of publicly available data. So my guess is would be public tweets would probably

be a big one. Right. They identify the demographic that they're looking at, they mine the sort of communication that these uh, these folks that are within that demographic tend to send, and he used as the model for Tay's communication,

and then the idea was to build upon that. Uh So, you know, we mentioned Goostman, uh talking about the fact that you know, we've got this context of a nineteen year old female persona for Tay with Eugene Goostman, that was one of the AI programs that quote unquote passed

the Turing test. We talked about this in a previous episode in which uh it was a an AI that was mimicking a thirteen year old Ukrainian boy, a non native English speaker, and that meant that the people who were uh interviewing this entity over a over a computer, uh, they were aware that they were supposedly talking to a human boy thirteen years old, non native English speaker, limited exposure to things like pop culture that would have come

before he was born. So in other words, it checked people's expectations of what this particular uh AI would be capable of doing. Sure, and similarly, I think the part of the concept for Tay was probably that you have certain expectations for how a nineteen year old woman is going to behave on Twitter, and it probably involves the use of of emojis and um uh not full grammatical sentences,

no capital letters whatsoever. Certainly um certain truncating of of thoughts and ideas, a certain inquisitiveness probably are the stereotypes of that sort of age group. It was to a point, and to be fair, I've only looked at sampling of some of the things he has said. It was to a point where I thought, I thought it was venturing on the border of parody, right. I was starting to feel like it was, Um, maybe someone my age is preconception of what a person would how a person would

constantly be communicating. Uh. And again that could just be my own bias going into it, because there were a lot of things where I'm looking at it and I'm thinking, I'm sure there are many people who do communicate this way, but it feels weird to have built something specifically to communicate in this way. Not that I necessarily thought that she would need to speak in full sentences with perfect grammar,

but just it. It almost felt like someone outside of that group trying to create language in a in a in a like like like a lame dad trying to sound like a cool dad and teenager. Yeah, that's pretty much. That's kind of how I felt. Now again, I'm not I'm not twenty years old anymore, so maybe I'm just totally out of touch and it was fine for people who are of that age and they didn't see anything

different about it. No, Nowadays, twenty year olds love linguistic interlopers from the older generation oh is that a trend. That's that's how That's how I was when I was twinty. I mean, we gen xers, we we appreciated a good you know, talk about interloping, interloping, interlocution, inter interlocular. They love it almost as much as they love being talked

down to about economics. Yeah. Yeah, So anyway, getting back to what Microsoft was trying to do, Microsoft's blog actually, uh kind of outlined what Tay was meant to do. TA is designed to engage and entertain people where they connect with each other online through casual and playful conversation. The more you chat with Tay, the smarter she gets, so the experience can be more personalized for you. So

she was supposed to learn. It's kind of like in a terminator too, where you know the terminator has his brain is a learning computer. The more he interacts with humans, the more he learns how to be one. Right. So that's what it is to say, Asta Leavista baby right, and in the end he could even learn to love maybe. But let's say, let's say you sort of took a page from the Edward for a Long Book, uh in Terminator two, in which he teaches him how to say

rude phrases to people. And you tried this out with something like ta you could get some very sad and very funny results. Right, And if you get enough of them, you can have so much of an influence that it starts to really change how how TAY starts to behave towards everyone. Right, Like, you know, the the beautiful thing about using something like Twitter for machine learning is that you have an enormous wealth of data at your disposal. Yeah.

The problem is when people are aware of it, and they are the ones supplying the data, and they can make that data whatever they want. They can purposefully misdirect your efforts to go in a direction they think is hilarious. Yeah, it's investing in quantity rather than quality. Yeah. And and to be fair, I mean this this relationship was a

little weird from the get go. Microsoft was using Tay to look at things like you would have a little profile and t would recognize stuff on your profile like whatever your handle or nickname was, your gender, uh, your favorite food, your zip code um, and uh, you know your relationship status, which I think is also a little creepy for why does an AI need to know if

I'm single or not single. That seems a little like like messages you would get from an artificially intelligent chat like you know, hang in there or whatever it may be. Seems a little weird. Maybe, you know, maybe Tay is particularly interested in relationship counseling is a possible career. Now

she needs to really work on her delivery skills. Then, because Tay started getting a little out of hand, So let's let's before we really talk about what went right the chaos, ya, let's talk about what ideally would have happened had had things played out the way Microsoft seems

to have intended it. Well, you said, you said it that she was supposed to be learning from a massive store of data, which is natural conversation on the Internet, and so that was supposed to be something that the computer program could analyze and say, Okay, I'm getting a very generalized picture of how people talk, right, and how they talk to each other and in response to specific

phrases and questions and and other little bits of linguistic cues. Yeah, and which is not inherently a bad idea, no, no, and and and going further than that, how different thoughts can be expressed or or the same thought rather can be expressed in very different ways, which is very useful if you're going to incorporate that kind of conversation understanding

in a digital assistant. Oh yeah, yeah, that's that's like the problem in in in language and machine language and language learning, because you because the kind of conversation that we have where we've said this a lot of times on the show before, but where we can say one thing and and if not dozens, then hundreds or thousands of different ways and have another human understand them as long as they're conversational and the same language. Uh. That's

that's real hard for computers. It's what you call not easy, right. So that's why usually we have to conform to the limitations of machines. Right. We have to be able to express our commands in a way that machines understand, which might not come naturally to us. And as we get better and better with machines understanding conversation, we have to do less work. Right. We don't have to accommodate the machines.

They just understand us. That's the goal. So that's what we wanted to have happen, or at least what Microsoft wanted to have happen. It's not what Twitter wanted to have happen, or you know, the other folks who are using kick or group me. Um, what they wanted to do was to perhaps exercise a little bit of control and power over something and direct it for their own amusement.

So you could do things would say that weren't particularly you know, uh, malicious or mischievous, Like you could ask her to tell you a joke or tell you a story. She would play games with you. She would create memes like we were talking about with group me. She could do memes with pictures. Yeah, so you could actually have these kind of interactions. But you could do other stuff too, um, like you could make her say things, like you could

tell her to repeat stuff. Now before, let's talk a little bit about some of the phrases she would use on her own. Here's here's a couple of sample tweets she sent out. Um. This is sort of the way Tay's baseline worked before she got corrupted. I guess it's probably the best way of putting it. Uh OMG totes exhausted swagulated too hard today, HBU, which I assume is how about you Laurence nodding yes, So it's not. I mean it could be. That's my favorite question to ask people.

Another one was damn t b H. I was kind of distracted, you got me. I'm like, okay, I can these are I can see these these tweets. I see where it's coming from. Uh. You know that's kind of the baseline again, sort of the the way t would react to two people just trying out TA sending messages to her normally. Uh, but then came all the trolls, all the people who think it's hilarious to mess with a system that they have some input into. So uh, one thing they would do is make TA repeat stuff.

So a lot of the lot of the tweets you've probably seen, if you've followed the story of Microsoft TA saying horrible, horrible racist, the very worst things are probably things that someone had repeat back. Yeah, you can imagine generally approving of Hitler. This the Godwin's law was was was fulfilled almost immediately. Um So yeah, making her repeat stuff back was easy. You didn't she didn't necessarily quote

unquote understand it. She was responding to a command, which was to say this thing that I'm telling you to say This is very similar to a program I used to have as a kid, um and it was very similar to Eliza, one of the early chatbots. So there was a sound card for old PCs called sound Blaster, and one of the pieces of software that came with sound Blaster was this program called Doctor Spaces. So Dr Spatso was essentially Eliza, except he talked like this, you know,

and you could ask Dr Space the questions. He would try to psychoanalyze you, so he would answer everything with his own question. But you could also make him say stuff, so he would say say and then you put things in quotation marks. He'd say whatever you wanted him to say, and much amuse and into it. Yes, because I was a kid and I just thought it was funny to make my computers say really rude things, mostly just mild curse words. Because I was a pretty boring kid. You

didn't have the imagination to make him approve of Hitler. Yeah, I didn't. I gotta admit I never made Dr Spazzo's out like a Nazi sympathizer. It never occurred to me as a kid. Kids today, man, they are miles ahead of me. Uh So that's that's what we started to see we saw people starting to have her repeat stuff, and not only was she repeating it, she was starting to learn. So she would start to respond to people and incorporate some of the stuff that she had been

told to repeat over and over again. Not consistently either. Sometimes she would make one statement to a person where she'd essentially equate something like feminism with being the most evil concept in the world, and then she would send another message to someone else where she she'd say something like, oh, no, femin is um is great. So uh, not consistent across

all messaging either. Well. And it also I would I mean, I don't know how the programming works, but I would argue that that the program probably didn't understand any of the topics that were being thrown at it, like, I mean, probably a few a few clue words like like hey, how are you means I'm supposed to say something about how I'm doing UM? But I you know, I don't think that it was taught what feminism is. I don't

think that it had an opinion on feminism. I think that, you know, it was just testing out an answer to a question, you say, what do you think about feminism? And it it's like, oh, this is a question about what I think, say it's good or bad? Yeah, yeah, I mean I get the sense that these are text strings for it. It's not a it's not a semantic engine and exactly sure, yeah there was there was. I agree. I don't think there was any understanding on the part

of Microsoft TA. In fact, that's one of the reasons why I really argue that people people who say this is a real warning about AI, that's b s. It's not a warning about AI. It's a warning about how people. It's a warning about real eye Yeah, human natus, it's a it's a warning about how how humans when we were given the opportunity to mess with something, it could be really hard to resist that opportunity. And this is

something that we've seen multiple times. So some other examples of stuff that you know, she she blasted out xenophobic stuff, racist stuff, homophobic stuff. Pretty much everything that was dark in human nature was spouted off by Tay in less than just plenty of regular old foul mouth dirtiness and pornography.

Sure so, uh, I love that. Um Gerald Miller on Twitter wrote, Tay went from humans are Super Cool, which was like one of the first messages TA talked about to full Nazi in less than twenty four hours, and I'm not at all concerned about the future of AI. Again, I would argue that this is not so such a case of worrying about where AI is going, rather than how do you design a system that uses machine learning from crowdsource data and factor out all the intentionally malicious misdirection.

That's that's the question that has nothing to do with AI. That has to do with understanding how people are right. I see you've got some specific exchanges recorded in here, which one which are something that we can actually say, I like the Ricky Gervais one. Yeah, so Ricky Gervais. This one was. This one's not not as as jaw dropping lee awful as some of the other things that she said, But this was one of those where Tay actually came up with a response. She wasn't repeating something.

So according to The Guardian, one exchange began a user asked, Tay is Ricky Gervais and atheist? Ricky Gervais is a comedian, Yes, and it's very uh outgoing about being an atheist, So so this person stay is Ricky Gervais an atheist, And they replied, Ricky Gervais learned totalitarianism from Adolf Hitler, the inventor of atheism. So we get Hitler, we get totalitarianism. Hitler apparently invented the concept of being an atheist. Um. Yeah, so it's sort of beautiful. Yeah, I mean that, that's

that's sort of a poetry of nonsense. I like the way these ideas are crammed together. At least at least in this one, there was nothing so like personalized. There wasn't like personalized hate. Who can take it on him? But Ricky Gervais not not super factual, but sure, I'm pretty sure Ricky Gervais would would use this as material in the stand up. Right. But but it's not something where it's it's centering out like a specific ethnic group for example. Right. So, as you can guess, microsaw looked

at this and said, well, this is going swimmingly. Let's just leave her online. What happened, pulled the plug. It took less than twenty four hours for them to say, you know what, we might need to revisit this after we think it through a little bit more. Uh. They also worked up a little curtain between the Internet and for a bit, I like to think that the little like technical difficulty sign came up, but yeah, exactly. Yeah, so they they also deleted a lot of the particularly

nasty tweets that they had made. She made a couple hundred thousand within that twenty four hours, which is more than even I have tweeted, which is saying something I know, right, so uh. Some people actually criticized Microsoft for deleting the tweets, saying that those messages should remain up to serve as a warning of the dangers of AI. But again, I don't think that's I don't think that's a really a

credible argument personally. Um. I think it's more of an of a way of warning people about how nasty folks can get on on social media. Microsoft's response was that the AI chat butt t is a machine learning project designed for human engagement. As it learns some of its responses are inappropriate and indicative of the types of interactions some people are having with it, We're making some adjustments to tay well, I mean that that seems accurate to me, Yeah,

absolutely accurate. Um. She did come back online very briefly about a little bit later on a couple of days was an accident. It was completely an accident. Microsoft said that they were testing her internally. Uh, in within Microsoft, not testing Tay's internal organs. She doesn't have any she's had. Yeah, it wasn't anything like that. I'm just now thinking like I'm putting the body of Cortana in there, just thinking of them slicing open the virtual Cortana to Cortana is

the character from Halo, artificial intelligent character from Hale. Also the name of that. Well, I imagine the second she came back online, people are trying to get her talking about Hitler again. Well, yeah, they said that, uh that mostly the messages that were being sent weren't offensive, but there were still quite a few that did fall into that category of crazy racist Hitler oriented tweets. So at least some of them were still in that that camp.

And Microsoft immediately shut her down again. Well not immediately, it was after they realized it, because it was in the middle of the night when it happened. Okay, okay, so yeah, a couple hours later they shut it back down. Uh, And I don't think they're surprised. To me here is that Microsoft was surprised by this at all. Yeah, because we I mean, I remember PR attempts to get crowd sourced information into marketing campaigns, that kind of thing every time. Yeah,

I remember. So I I agree with what you're saying about how I don't think this is really all that uh less curative of anything about artificial intelligence, but it is very illustrative of stuff about Internet culture and about what happens when you try to crowdsource things, especially when major companies try to crowd things. Uh. There is an inherent urge to mischief that is always latent on the Internet, and there are certain types of stimuli that draw it

out more than other things. And I can admit to being very much a part of this in some scenarios. Now, I wouldn't go try to get ta saying racist stuff, but I do remember one time Coca Cola had a thing where they let you generate your own memes that let you make your own Coca Cola animated gifts that had so have like a Coke ad going in the

background and you can enter your own text. And I guess what they were hoping people would be entering is I don't know, like you know, like big marketing campaign right, they'd have people putting in like, hey, Stacy, I love Coke to their friends. I don't know, uh, like what good could come of this? So I actually when they did that, I remember reading a very interesting article about testing all of the different things Coke wouldn't let you say and and and they had a wide dictionary of

no no terms. They wouldn't let you talk. They wouldn't let you use curse words or slurs or anything like that, obviously, and they also wouldn't let you talk about uh most obvious morbid topics that they didn't want you talking about Nazis or Hitler. They didn't watch you talking about health problems or anything like that. But I I felt dared to mess with it by this, to probe this this

uh these boundaries and find out ways to get around it. Yeah, so I remember, Um, I remember figuring out that it wouldn't let you talk about demons in English, but you could you could come up with phrases about d hmans in Latin that it would let you do. So I thought that was pretty fun. Yeah, I remember talking about this with you earlier this week, and I said, yeah, I could imagine writing something like Taste Sweet Oblivion, and like you know, you could easily find ways around the

the protective barrier. There. There are other examples too. I don't know if you guys remember this, but I remember actually following the story when it happened. Mars Incorporated several years ago created this module for their Skittles web page, and the module allowed tweets that contained the word skittles to be uh to be displayed on the homepage in more or less real time, like shortly after you posted it. And it became abundantly clear early on that no one

was vetting the tweets, no one was curating. It was just a straight feed from the Internet, from Twitter into the Skittles homepage. And all you gotta do is include the word skittles. So you could write anything you wanted to put the word skittles in there it would show on that homepage. There were some really funny ones. I think I even did one. If I'm not mistaken, I'd have to do a search on my Twitter history to

find out when I said skittles. I mean, I mean, people weren't just saying really clever things about how they were tasting the rainbow. No, no, I mean I'm sure that made up a significant percentage of the tweets that showed up on the Skittles homepage, but genuine earnest love for Skittles. People just couldn't contain. They had to go straight to Twitter to share it, or they were playing a game of Skittles and they were just confused why

their game was being broadcast on a candy website. Uh. At any rate, there were all sorts of different posts that went up there, including ones that were offensive obviously, so there were there's precedents, right we we That's the thing that really puzzles me that Microsoft went forward with this without even taking into consideration the precedents that has been set with crowdsourced campaigns like this, I think, especially

on Twitter. Yeah. But but but but not not even not even only there, No, I mean Microsoft itself had already experienced with this. In two thousand eleven, Microsoft launched a poll on Facebook. It was part of the Microsoft Phone Projects, so it's one division within Microsoft, uh, and they asked a question. They were saying, what was your

favorite feature of the Nokia Lumia eight hundred phone? And you had some options to choose from, but you could type in your own option, and people did, and a lot of them options were like there's not a not a darn thing, that kind of thing. That's a very pc way of putting the way they wor did it.

So there were options that were very much uh not in line with Microsoft marketing strategy, and those got lots and lots and lots of votes because people thought it was funny or they were dissatisfied with the product, or both. So it's interesting to me that the same company that had experienced this back in two thousand and eleven would go forward in two thousand and sixteen with a plan that had some of the very same trappings there, like

it was almost set up to fail in that way. Um, but yeah, it's just one of those things where I think, I think part of the reason why there is the sense of mischief one you're you're looking at entities that are much greater than yourself. Right, Microsoft is enormous, huge company. Just don't feel bad about it. Like if it had been a small business trying to say, hey, make your own meme about our product, I don't think I would would try to make memes about demons. But it's Coca Cola.

I'm just not worried about them. I don't think my my demon memes about Coke are really hurting Coke. Well, not only that, but I mean, depending upon you might actually have some very negative opinions about whatever the corporation is, right, like Microsoft, Microsoft one of those companies that's very divisive there. I don't know that there are that many people who are super pro Microsoft, but I know people who are either Microsoft is a necessary evil, I need it for

what I do, or I hate Microsoft so much. That seems to be the case, except for maybe Xbox. Xbox players are like, no, I really love the Xbox, which I totally get, but uh, it's it's one of those things where you you that also encourages this mischief. But yeah, when I one time when the NYPD I think, tried to start a Twitter hashtag campaign. Do you guys remember that one? Yes, I remember that. It didn't go well, it didn't turn out the way that they wanted it to. Yeah.

I think they were trying to they were trying to crowdsource people. Uh what were they trying to get them to do, to like tweet good experiences HASHTAGYPD? Yeah, see, giving giving a voice to a population. Uh, that may have legitimate criticisms to say, you've got to be prepared for when that comes in, you know. I mean people don't like being told, uh and prompted with the understanding

that you need to say something positive about my massive organization. Yes, yeah, yeah, even if you even if you previously did like something. There is something that strikes most of us the wrong way when we're told we need you know, you are required to say something positive about this thing, even if you liked it before. Just the thought that you are

now being compelled to do it. There's most of us have a resistance to that, that that little, say nineteen year old inside you just like, hey, I can miss with I guess I technically have two of those now. Now I've seen Tay talked about in a reference to another interesting technology issue, which is the prevalence of of female gendered digital assistance or digital embodied personalities. Yeah, it's

this is it's interesting. I am trying to think if there is a personal digital assistant I've heard apart from Siri, which you can you can have a male version of Siri. The the British version is male and that's the one my wife has on her phone. Apart from Sirie. I'm not aware of any of the major digital assistance that are not female. Cortonas fee male. Google Now's voices is female. You might be able to change that one. I don't know. Alexa for Amazon is female. So that's that's something that

strikes me as weird too. Maybe they don't know what the psychology is behind that, what the market research says. I don't know, but it just that also is odd. And for Microsoft in particular, this was a really troubling week for them because the same week that they launched Microsoft Tay and she goes off the reservation because she's been steered the wrong way by by malicious trolls. They

were also dealing with a pr nightmare. There was the Game Developers Conference that was going on earlier that week, and Phil Spencer, who's the head of Xbox, actually came out and apologized for an Xbox sponsored party. It took place in a nightclub, um and it was part of

the Game Developers Conference overall event. At that party, they had a lot of skinnily clad women dancing on platforms, and anyone who is at all aware of the time it within the video game industry right now knows that the sexual objectification of women is a is a huge topic of discussion, uh, in particularly at conferences like that. It's been it's been talked very nearly to death, just

just the concept of of of hiring performers like that. Um, and not that there's anything wrong with being a performer like that, but it just it just paints it paints the industry in a very particular way, and one that one that is not necessarily welcoming of of of women's

participation as as as brains rather than you know, dancers. Yeah, I think it says sort of the assumed culture of our industry is sort of a tradition, yeah, traditional hyper masculine tech bro culture, right, which which perpetuates this unwelcoming environment for people who have very valuable skills and voices to add to that industry. But we're continuously denying them the ARA tunity or or if not denying them, at

least discouraging them feel unco right. So this was this was like two strikes in a row for Microsoft as far as trying to to work with these ideas that involved in some way gender Like, this's not not great record for Microsoft. In March of two thousand sixteen. So the question then is what's next. Well, what if I told you that Microsoft said hey, now everyone can make one of these. I think that's a great that's a great idea. So the very same I'm not the very

same day that Microsoft had to take Tay offline. Uh, there was the Microsoft Build Developer Conference and the CEO of Microsoft, Soughtya Nadela, announced that the company would make available the Bought Framework. That's all one word, bought framework, and Bought Framework is essentially the a p I that developers can use to create their own version of chatbots that can work on different sorts of platforms, including stuff

like Skype, Slack, group Me, other messaging systems. UM so very interesting that they had this very well publicized I don't know that you could say any word other than failure of of Microsoft kuffle kerfuffle would work, this publicized kerfuffle of Tay, and then go ahead and say, hey, you can do it too. Now that being said, it's entirely possible that you could roll out something along the lines of Microsoft TA and and have it unfold in

a very different way than what we saw with Tay. Well, I mean one option that occurs to be I don't I don't know if this would be the case, but I think it's at least possible to consider that by putting more of these things out there, you would essentially people would get it out of their system, like it would become less novel and less interesting to try to mess with a bot. Yeah, yeah, you're like I messed with a bottle last week. I'll not talk about Hitler

to this one. Like, to me, it would just make sense to essentially nullify the machine learning part of Tay's software while people are getting it out of this Let it go for a month and then take it off, and sure you're not gonna have as much engagement after that because people are going to get bored. But the people who are going to get bored are probably the ones you don't want engaging with the AI in the

first place. So or you take another approach where you have like a controlled beta right where you have a controlled audience that can interact with that AI, and you are just very careful about monitoring it and you encourage people to have you know, natural conversation with the AI as close as they possibly can. But if there are any shenanigans, popping up. You just you you squash it. It's a lot easier to do that if you haven't under a controlled population than if you open it up

to the world at large. Um. And honestly, this is a really important field for Microsoft because this is a company that's going through a massive transition the company. You know, you think, what are Microsoft's biggest products, Windows and Microsoft Office, and those are for desktop computers, and desktop computers and laptop computers have been in pretty serious decline compared to

the mobile platforms. So Microsoft is trying to transition itself into more of a mobile platform company so that it can continue to grow and develop and evolve rather than become obsolete over time. They definitely don't want to turn into the picture of what Microsoft is in those old Apple versus Microsoft ads. Uh. I'll respect to John Hodgman. Um, So I think I think it's I understand where they're going.

I think it was a misguided approach. Uh. Slightly hilarious, slightly sad to see the just the amount of abuse that was thrown at Tay in order to corrupt her in this way. But it it makes me curious to see how people are going to try and tackle this problem in the future, because there is there is a lot of value to getting a grip on this natural conversation.

Like we've said, it's something that could make home automation incredibly effortless, where you know, you just you could say something one of twenty different ways, but your house will totally understand what you mean and respond in kind. You know. Another thing that actually just occurred to me is that I think in the future, a machine learning program that's supposed to, you know, gain insights about natural language from interacting with the public might actually have something useful to

learn by dealing with tons of trolling and abuse. Like I don't know if it's necessarily the case that it that you should just turn off its learning part, then uh, certainly it shouldn't learn how to talk from these people and say like, yes, this is a model of good conversation. But I'm not sure that there is nothing to be learned. There might be something to be learned, do you know what I mean? My thought is very vague here, but no, no, I mean, especially uh, being the person super secret you

guys listening out their YouTube in the room know this. Um, I'm the one who reads the YouTube comments for how stuff works. Yeah, so I see a pretty wide range of responses to a pretty wide range of topics from the general public on a daily basis, and and it's it's genuinely fascinating to me trying to sess out, uh, what is trolling and what is sincere, particularly in response to to videos that have been produced by our stuff they don't want you to know Team, which deals with

a fringe theories. And there are a lot of responses in there that that are you know that there's this huge bizarre contingency of flat earthers on YouTube and I seriously cannot figure out if they're trolling or not. I think a lot of them are. But that's this is a really great point. This is this is a perfect jumping off from what I'm saying, Like I think, uh, potentially sampling huge amounts of trolling and abuse could maybe lead into machine based insights on what's the difference between

somebody messing with you and somebody having a seriously extreme opinion? Right, And and of course, I mean, like you're saying, Lauren, it's hard for us to tell depending upon the cues that we have or lack of cues. And so this is a huge challenge, not just for AI but for human intelligence, you know, just and anyone who's had that kind of miscommunication online understands that it can. It can legitimately happen without any intention on either party for it

to be misinterpreted. Oh absolutely, And then that kind of interpretive help I think would be would be terrific because you know, especially speaking with people, uh for for whom your language is not their first language, can be very

difficult to to have a clear conversation. And there have been times when you know, I've I've been like trying to answer someone's earnest question and they keep trying to rephrase it and I'm not getting it, and like Google Translate isn't getting it, and and it's just like all I want to do is tell you how your fingernails grow, but I don't. I don't know that that's the question

you're asking, right, So that that kind of thing. The hope I have is that once the fire has has died down at Microsoft trying to handle this this PR issue, they can go over that kind of data and can actually glean stuff from it, because it would be really,

as you point out, genuinely useful and Uh. I would love to see something positive come out of this experience and not just have it be another oh, those wacky pr people or however you want to frame it, like the Skittles stuff, right, like something beyond just confirming what we already know, which is that people like to stir crap up when they have the opportunity. UM, I would

love to see that. I hope that that's what happens, and I hope that people who were behind Microsoft A, the people who were developing it, aren't being held to the fire for what happened, even though I do think that at least some of it could have been predicted from day one. But that's that's another that's another point entirely. At any rate, that is the tragic comedy of Microsoft TA as it stands. Right now. We're recording this March, so perhaps by the time this episode goes live there

will be more developments. But I just thought it was an interesting story and one that I think has implications far beyond just Internet based AI. This also, I think gives lessons to roboticists who are trying to figure out how to design human robotic interaction and that are going to be acceptable and work and not create more problems, because that's a that's a fascinating area of study that

relies more upon human psychology than actual robotic ability. And and when you see things like this happen, as as someone who designs robots, I imagine they have to sit there and think, oh, we might need to rethink some of our approaches because people people are willing to mess with stuff in certain contexts. Perhaps we need to take that into account in our design so that we can

head that off before it becomes a problem. Um. So fascinating really, But at any rate, if you guys have any suggestions for future episodes of forward Thinking, please send them our way. Send us an email, our addresses f W Thinking at how Stuff Works dot com. You can drop us a line on Twitter. Our handle is f W thinking. Um we will respond without being tay like at all, I promise. Or you can go to Facebook

search fw thinking. Our profile will pop right up and you can leave us a message there and we will talk to you again really soon. For more on this topic in the future of technology, visit forward thinking dot com, brought to you by Toyota. Let's go places