Pod Insights Practical AI Analyzing AI's impact on society through art and film About this episode Brett Gaylor joins Chris and Daniel to chat about the recently announced winners of Mozilla’s creative media awards, which focuses on exposing the impact of AI on society. These winners include a film that responds to the audience (via AI recognized emotions) and an interesting chatbot called Wanda.
Sponsors:
DigitalOcean https://do.co/changelog – DigitalOcean is simplicity at scale. Whether your business is running one virtual machine or ten thousand, DigitalOcean gets out of your way so your team can build, deploy, and scale faster and more efficiently. New accounts get $100 in credit to use in your first 60 days. Fastly https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorship – Our bandwidth partner. Fastly powers fast, secure, and scalable digital experiences. Move beyond your content delivery network to their powerful edge cloud platform. Learn more at fastly.com https://www.fastly.com/?utm_source=changelog&utm_medium=podcast&utm_campaign=changelog-sponsorship. Rollbar https://rollbar.com/changelog – We catch our errors before our users do because of Rollbar. Resolve errors in minutes, and deploy your code with confidence. Learn more at rollbar.com/changelog https://rollbar.com/changelog. Linode https://linode.com/changelog – Our cloud server of choice. Deploy a fast, efficient, native SSD cloud server for only $5/month. Get 4 months free using the code changelog2018. Start your server - head to linode.com/changelog https://linode.com/changelogFeaturing:
Brett Gaylor – Website https://www.brettgaylor.com/, X https://x.com/remixmanifestoChris Benson – Website https://chrisbenson.com, GitHub https://github.com/chrisbenson, LinkedIn https://www.linkedin.com/in/chrisbenson, X https://x.com/chrisbensonDaniel Whitenack – Website https://www.datadan.io/, GitHub https://github.com/dwhitena, X https://x.com/dwhitenaShow Notes:
Winners of the Creative Media Awards https://blog.mozilla.org/blog/2018/10/24/keeping-ai-accountable-with-science-fiction-documentaries-and-doodles-plus-225000/Do not track movie https://donottrack-doc.com/The Science behind Cambridge Analytica https://www.gsb.stanford.edu/insights/science-behind-cambridge-analytica-does-psychological-profiling-workMozilla manifesto https://www.mozilla.org/en-US/about/manifesto/Machine bias article from ProPublica https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencingPractical AI episode from Lindsey Zuloaga about bias https://changelog.com/practicalai/17Upcoming Events:
Register for upcoming webinars here https://practicalai.fm/webinars!
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Speaker 00 Baiting for change log is provided by Fastly. Learn more at fastly.com. We move fast and fix things here at change law because of roll bar. Check them out at rollbar.com and we're hosted on Lino servers. Head to linode.com slash change log. This episode is brought to you by Digital Ocean. They now have CPU optimized droplets with dedicated hyper threads from best in-class Intel CPUs for all your machine learning and batch processing needs. You can easily spin up their one-click machine learning and AI application image. This gives you immediate access to Python 3, R, Jupiter Notebook, TensorFlow, scikit, and PyTorch. Use our special link to get $100
0:42 Speaker 00 credit for Digital Ocean and try it today for free. At the do.co slash change log. Once again, do.co slash change log.
1:03 Speaker 00 Welcome to Practical AI, a weekly podcast about making artificial intelligence practical, productive, and accessible to everyone. This is where conversations around AI, machine learning, and data science happen. Join the community and slay with us around various topics of the show at change.com/community. Follow us on Twitter. We're at Practical AI FM. And now into the show.
1:29 Daniel Whitenack? Well, hello, Chris. How's it going today?
1:31 Speaker 00 Doing really well. How are you today, Daniel?
1:33 Daniel Whitenack? I'm doing good. I know in one of our previous news and updates show, we started kind of going down the rabbit hole of AI and art. And I'm really excited today because we have Brett Gaylor here with us from Mozilla. And he's really working at the intersection of AI and media. And so I'm really excited to have you here, Brett. Welcome.
1:58 Speaker 02 Thanks. Nice to be here. Hi, Chris. Hi, Daniel.
2:00 Daniel Whitenack? Hi. Yeah. So Brett, why don't you give us a little bit of a background of how you got involved in AI and media and Mozilla? Just give us a little bit of background about yourself.
2:10 Speaker 02 Sure. I'm a documentary filmmaker is usually how I describe the work that I do. But actually, scratch that. I would call myself a documentarian. I'm kind of, you know, platform agnostic to be to be nerdy. So I make documentary.
2:26 Speaker 02 Yeah, I make creative nonfiction work on the internet about the internet, using the internet. So kind of a three-part trifecta there. So I've made some feature documentaries. One of them was about remix culture. It was called RIP, A Remix Manifesto. And that was, you know, kind of a, in the early aughts, you know, around 06, 07, 08, we were sort of seeing the internet become a place where kind of an emerging folk culture of people taking media and remixing it, downloading it to their computer. And that whole process was really creating a lot of anxiety in traditional
3:09 Speaker 02 legacy media, but also pointing kind of a way towards a more participatory culture of the internet. So I made a documentary about that sort of period of time. And, you know, it was a really hopeful time for those of us who, you know, were involved in the internet's early days. It kind of felt like this really democratic moment where anybody who wanted to participate in the culture around them was suddenly able to do that much more easily and to be able to find a good, kind of a global audience all around the world.
3:42 Daniel Whitenack? The internet was going to democratize everything and solve all our problems.
3:47 Speaker 02 It wasn't it? It was like, it was just...
3:49 Daniel Whitenack? Did it do that?
3:50 Speaker 02 Well, and Nazis, you know, that's... So yes, the internet did do this, but then we sort of saw a few cracks in the utopian kind of landscape. And around 2013, I started thinking about, you know, the ways in which the business model of the internet was leading to an increasing amount of a collection of personal data about people that use the internet and sort of this creeping surveillance capitalism sort of model that was beginning to emerge as the sort of de facto, especially since in that earlier moment that I
4:35 Speaker 02 described in the internet's evolution, it was like, oh, wow. And here's another great platform. Facebook, it's going to encourage us to share everything and
4:44 Daniel Whitenack? Connect all the people.
4:46 Speaker 02 It really did. They were really, really good at encouraging us to share things. And so at that time, I had started to work at Mozilla and had created with a lot of amazing community members, a system called Popcorn JS, which was essentially a way to synchronize a piece of media, like a video or an audio file with web events. So you could say, you know, at this time, make the webpage do this, or pull in this piece of data, or fire this JavaScript command. And I thought, what would it be like to make a documentary about online privacy that would play like a movie, but it included your data? And so that was the, that was the inspiration behind Do Not Track, which turned into quite
5:31 Speaker 02 a large documentary series that was produced with the National Film Board of Canada, several partners in France. And it did really well. We, that kind of played around the world. It was, received a Peabody Award, which is sort of a recognition of media.
5:46 Daniel Whitenack? Congrats.
5:47 Speaker 02 Yeah, thank you. It was fun. That year was David Letterman and the folks behind Mr. Robot and, you know, just some amazing media. And it was kind of cool to see documentary work about the internet sort of in the mix of important social issues. And so
6:05 Daniel Whitenack? It probably also illustrates people's genuine, like, interest and concern over the topic as well, I would imagine.
6:13 Chris Benson? And I also just wanted to note that I find it ironic that as we record this today, it happens to be Election Day in America.
6:19 Speaker 02 Yeah.
6:20 Chris Benson? And a lot of these issues are very top of mind as we're going through that.
6:24 Speaker 02 Yeah, that was definitely not the case when we started to make Do Not Track, but over the course of making the series, you know, it became much more relevant. Like, for instance, we had one of the episodes, the third episode was about, it was sort of like, if you look at it now, it's kind of naive. It's like, what could you know about somebody from their Facebook profile? And we're like, we found this, you know, sort of niche study by researchers at Cambridge University who feel that they're able to predict your personality based on your likes on Facebook. And that was the AI researcher Mikhail Kosinski, who did, who basically created this model where you could correlate
7:04 Speaker 02 persons likes on Facebook to a really high degree with their psychological profile. So it was a vetted and peer reviewed study where I think what they did was, you know, took interviews with people's family members and their friends and, and psychologists and sort of plotted you on what's called the ocean model, which stands for openness, conscientiousness, extraversion, I want to say agreeableness and neuroticism. So that's your, that's your ocean model. So you, you exist somewhere in an axis of all of those points. And so what Kosinski was able to do was correlate, like say, the, the one that we always use was, was apparently like if you like
7:45 Speaker 02 The Dark Knight, that correlated really highly with a low score in extraversion, for example. And so basically they took a large group of likes on Facebook and then did these interviews with folks and sort of said, like, here it is. I could, I could log in with my Facebook ID and it would look at all my likes and then say like, here it is. You're, you're this open or you're this neurotic. The nuts thing is like, it is pretty accurate. And so anyway, we use that, that, that became an API called Apply Magic Sauce. And we use that in the Do Not Track documentary to sort of, you know, use people's personal data to sort of show how these emerging,
8:29 Speaker 02 you know, this, this is AI really was starting to, to work. Now, then what happened after the film came out, or probably while we were making it, was this startup called Cambridge Analytica requested the licensing of that API. They were denied by the University of Cambridge. And then they were like, well, we're just gonna basically copy their approach. And out of spite, we'll call our company Cambridge Analytica. And the rest is history.
8:58 Daniel Whitenack? Seems like some of us are familiar with that, that the rest of that story.
9:02 Speaker 02 Yeah, the rest of the story is, it's interesting. He actually has another very controversial paper. I don't know if you've seen it, but attempts to guess a sexual orientation based on a Facebook profile photo.
9:17 Daniel Whitenack? So I've seen this.
9:18 Speaker 02 Yeah, he tends to create these like highly controversial sort of mind bomb sort of studies that kind of illustrates some of the bleeding edge dragons of these emergent like machine learning or AI systems.
9:35 Daniel Whitenack? So is that kind of where, I mean, I know that kind of up, you know, recently in your profile, you're kind of heading up these creative media awards for Mozilla was kind of all of that that you've mentioned that was kind of in the mix, leading up to Cambridge Analytica and all of those things. Is that part of the driver for that work and how you got involved in that?
9:55 Chris Benson? And also, could you take a second and just kind of describe Mozilla for anyone in our audience who isn't familiar with it? They may know Firefox, but just kind of give a quick intro as you answer.
10:05 Speaker 02 Yeah, you guys got to keep me on track too, because I can.
10:07 Chris Benson? All good stuff. This is good, man.
10:11 Speaker 02 This could be a long answer, but I'll keep it short. Yeah, so most people are familiar with Firefox, but Firefox is one aspect of the mission of the Mozilla Foundation, which is a project basically to keep the internet open and accessible public resource for all of humanity. So we're guided by a manifesto, which I would encourage anybody to check out online. Yeah, and it really comes out of the recognition that the Firefox project and the open source code is a useful instrument in the market to make sure that there is some web browser and sort of user agent that exists in the world that is independent and is not one
10:54 Chris Benson? Corporate driven.
10:55 Speaker 02 Yeah, exactly. And however, we don't disparage the commercial interest on the Internet, and we value independence at the same time. So Firefox is one thing that Mozilla does, but the Mozilla Foundation that I work with also does various interventions to ensure what we call Internet health. So we see like the Internet as an ecosystem, and for that ecosystem to be healthy, we need to kind of tend it. You know, it's like, imagine that the Internet was like the ocean. Well, if you want the ocean to continue to be healthy, you got to make sure people aren't polluting it, or somebody hasn't overfished it, or, you know, and so sometimes that requires work. And so for us, that work is giving
11:39 Speaker 02 out grants and awards to sort of promising approaches to Internet health. We also have a really robust fellows program. So we support leaders in this effort to keep the Internet healthy. So that could be like a policy person, like a lawyer, or it could be a technologist, or it could be an activist, or in my case, it's media makers who are trying to explain this stuff to the public.
12:04 Chris Benson? So that sounds incredibly dynamic as a mission for Mozilla, because I mean, as this is evolving so quickly, then I guess it has to really keep track of new developments that are coming out. Obviously, AI, as we're talking about that, how is Mozilla involved in AI? How does it use it internally? And where is it taking that? How is it choosing to participate?
12:24 Speaker 02 At Mozilla, we kind of recognize that these emergent AI systems are just becoming part of our computing environment. Like, what's the quote? It's like, when you're fundraising, it's AI. When you're hiring, it's machine learning. And when you're implementing, it's linear regression. I think this is what they say. So we have some of the, you know, best minds in engineering that think about this or technical infrastructure of the Internet, and just a lot of the elements of this computing environment now feature some of these computing principles, you know? So trying to look at large patterns, trying to build systems that evolve over time, that's just kind of Part of the way that you make software these days.
13:09 Speaker 02 And so we want those systems to have the same kind of values that we expect and push for in other aspects of the internet. We want it to be transparent. So you understand how it works. We don't want more data collected about you than is necessary. We want other engineers to be able to see the code so that they can confirm that there isn't bias in those systems. We want the internet to be built by as wide a cross section of society as possible. So by that, I mean, it isn't just engineers in Silicon Valley who are creating these systems.
13:50 Speaker 02 They are consulting with civil society. They are consulting with potentially the groups that are evaluated by these by these systems. So it's a really, that's a complicated answer to kind of a big question is like, what does Mozilla think about AI? It's like, we think that it needs to serve humanity and we think that it needs to be open and free and healthy.
14:14 Daniel Whitenack? Okay, sign me up. I'm ready. Yeah, that's a good answer. I'm wondering if like, and maybe this is a more personal question as well. But you know, in terms of like the current trajectory of AI, you know, as a community, as you know, practitioners, as researchers, do you see the community, you know, embracing those sorts of values? Or do you see it, you know, kind of steering steering in another direction that's maybe concerning to you?
14:41 Speaker 02 I think there's some really positive signs that, you know, that let's call it the computer science community is realizing that, you know, you need to think about the social implications of what you build. And that's why Mozilla is making efforts to support that. So we just launched a program to support promising approaches to ethics in computer science education. For instance, you know, if you look at what's come out of Google a lot lately, you know, there's some promising signs from both like employees and
15:15 Speaker 02 Management that they recognize that there needs to be some really like bright lines that separate where artificial intelligence technology should not be monetized or you know, like the examples of, you know, not wanting to have their technology used in military contracts, for example, or, you know, you're seeing a lot of employees of Amazon, for instance, not want those facial recognition technologies used in immigration or other really sensitive areas where you need a lot of public oversight and transparency and how those systems are built. I think that's happening. What maybe concerns me is the sort of speed
15:54 Speaker 02 at which all this is is changing and kind of a feeling of that there's sort of a manifest destiny in the way that these technologies are built. So it's like, oh, yeah, any, any place where we could collect data, let's just collect it. And then we'll, we'll assume that there will be a use for that, that machine learning algorithm yet to be invented will, will solve. I think you sort of see that a little bit in the maybe what's promised to governments or cities about how, you know, they can save money or make difficult decisions at scale using machine learning or artificial intelligence. I don't know if you guys are
16:36 Speaker 02 familiar with this report that ProPublica did on the US justice system where some states were using machine learning systems to predict people's probability of reoffense.
16:49 Chris Benson? I, I am familiar with it. I've read up on it.
16:51 Speaker 02 Yeah. And so that's like an area where it's like, you know, maybe we should just let the humans continue to make these decisions because it's very difficult to sort of see where exactly bias can occur because these algorithms are so complex and it's so difficult to give them data that isn't collected in a manner that doesn't reinforce an existing bias of the past. So that's, that's a real concern about, you know, these automated decision making systems is oftentimes they just reinforce previous inequalities or, you know, frankly, like racist systems that have evolved.
17:33 Daniel Whitenack? Yeah, in that report you mentioned, if I recall, there was an inappropriate bias against African Americans as a result of that. Correct me if I'm wrong, but I think that's what the result was.
17:44 Speaker 02 If anybody's interested, you can just look up ProPublica. The report is called "Machine Bias", and it basically took two, two people that were eligible for bail. And the system basically predicts whether one person is going to reoffend or not. And it, two people with a very, with similar crimes, an African American woman and a Caucasian man. And while the man actually had much more serious previous offenses, the African American woman was denied her bail because she was assessed to be a higher risk than the man. So now, why is that? And is that because African Americans in the United States
18:25 Speaker 02 are incarcerated at a higher level than Caucasian people? That's true. Does that mean that that woman, that specific woman, is more likely to reoffend than that man? No, it does not. But the system assessed them this way because it's looking at this, a history in the United States that unfortunately incarcerates black people at a much higher level than it does white people. And so then, what happened in this case is the man got bail, and he reoffended, and she didn't get bail, was, you know, left in jail for a longer period of time. And so, that's just sort of like doubling down on some of the problems that we, that we see in these systems. So, in this case, it's kind of like reinforcing problems of the past.
19:09 Daniel Whitenack? Yeah, and we recently had a talk with Lindsay Zulaga, who works in, in hiring, and we were talking about some of the biases there as well. And some of these things come up very subtly. You know, she was talking about just the fact, you know, that you have fewer women applicants, for example, in software engineering positions. And so, thus, you have less data. You know, if only one woman applies for the position in software engineering, and the AI, for whatever reason, determines that that wasn't a good candidate, then it can generalize to all women applying. And software engineering, right? And so these things come up and they have they have a huge impact on people's real lives. And I know that we're going to talk here in a second about some of the awards that Mozilla
19:54 Daniel Whitenack? has given out recently, but I know that those are focused around AI's impact on society. Are these the types of impacts on society that you're that you're imagining in terms of maybe a biased AI, you know, giving certain people a privilege or whatever it is? Or in general, how do you kind of see the biggest impacts of AI that AI is having now? And maybe in the in the near future?
20:18 Speaker 02 Yeah, that's exactly right. This is basically the way that we framed it is projects that use this kind of media advocacy to highlight some of these unintended consequences of of artificial intelligence and and places where we want to be thoughtful about how we apply it. So we're we're awarding media makers who are kind of exploring these topics in a way that, you know, laypeople can understand that unlike the three of us don't think about this stuff obsessively every day, it's like for the first time. You know, like when you talk to people about AI, I'm sure you both get this all the time. You were like to talk to somebody at a Christmas party, they'd be like, oh, you mean like Terminator? And like, you know, like, you're like, you know, like problems in the future with AI.
21:02 Speaker 02 You mean like that the robots are coming for all of us?
21:04 Daniel Whitenack? You mean they're not?
21:06 Speaker 02 Well, maybe. And there is some real stuff happening right now in the world that we want to be thoughtful of and that actually can be either course corrected or, you know, with intention and thoughtful design and like the proper application of ethics can maybe turn out all right. We don't have to worry about a future, you know, robot invasion. So it's these questions of bias, but it's also, you know, I mean, I could talk about some of them. There's one that I'm really excited about. It's called Stealing Your Feelings by a really fun and funny engineer slash comedian, which is a fun mix that you don't really get.
21:43 Daniel Whitenack? It's a perfect combo.
21:44 Speaker 02 I know, right? So so Noah has created this project that looks at facial recognition systems and specifically patents that snap. Chat has recently filed to be able to do facial recognition on like groups of people. So what he's gonna do is use the webcam of your computer to in real time analyze what your face is doing while you're watching the film. So that he can, you know, either shock you or make you laugh or make you surprised or make you angry. And the film can react to your emotions while you're watching it. And that sort of is the perfect example of what we want to do because an audience that watches it. It's not like you're just telling them that cameras can, you know, detect their feelings.
22:29 Speaker 02 You're showing them like their feelings will be detected in real time and that will change the movie. So that's one, you know, and it's like, it's not, that's not necessarily biased, but it's a, it's an issue and would be.
22:38 Chris Benson? A little bit creepy.
22:40 Speaker 02 A little bit creepy.
22:40 Chris Benson? Good use of masking tape on your laptop.
22:43 Speaker 02 Exactly. Might, might lead to, you know, duct tape flying off the shelves. There's a fun one called a week with Wanda that basically simulates a, an AI that tries to be helpful in your life. But goes off the rails and starts to suggest things to you that you may or may not want. This one's gonna be kind of all done with, it's like an episodic serialized email exchange with you. So the AI might one week be like, oh, I noticed that, you know, you wanted to spend more time with quality friends. So I went ahead and deleted like half of your Facebook friends because you don't talk to them anymore. So the idea like that this AI is being like a little bit too keen and too helpful.
23:28 Speaker 02 There's one called survival of the best fit, which does address a lot of these, the issues that, that you were mentioning, Daniel, about biases in workplace hiring. So they want to like show that by simulating a job application process. Another one that I'm like super excited about is called the Training Commission, which is a, it's basically like a work of creative fiction. Stay with me here. It's a speculative fiction from the future that is looking, that is, takes place when a truth and reconciliation committee is struck to see what happened with an artificial intelligence.
24:09 Speaker 02 That basically something cataclysmic happened in, in society. So this AI, we don't exactly know what happened. The, the AI either. I don't know. Did it, did it, did it assume that it needed to fix some part of the way that humanity was structured? Did it, what exactly happened? So we have to piece together what happened in this story by the events that are told to us about this truth and reconciliation condition. I'm murdering this one. Ingrid is going to be really sad to hear my explanation. And then there's, there's another great one that looks called What Do You See? That examines the difference between what an image recognition system sees and what a human sees. So it uses
24:54 Speaker 02 a lot of these like edge cases of where AI systems can't understand, like when people, when it shows a picture of somebody wearing a mask, for example. This one, Mate Me or Eat Me is going to be all about bias in dating apps. And it's basically like a game where you swipe right or left to choose which monster you want to date. And then it'll sort of show in that process how quickly bias and reinforcement can lead to discrimination within those sort of dating systems.
25:23 Daniel Whitenack? So Brett, I'm thoroughly interested and intrigued by all these projects that you just mentioned. I kind of have like a general question about all of these, given that, you know, Chris and I are kind of like AI practitioners, I guess you would call us. I'm really interested to hear your perspective on why you think that in this effort to explain kind of like how AI works and expose like some of these things like bias and other things, why it's so important to involve creative people, maybe, you know, artists and like you were mentioning comedians, filmmakers, writers, why it's essential that we kind of involve those people in helping us tell that story.
26:03 Speaker 02 I think it's really important to involve creative people because often their job is to give a language to things that we don't know how to talk about yet. You know, like, if you think about, again, if you think about your friends that aren't obsessing over these topics, when they sort of encounter something that we might call machine learning or artificial intelligence, where it's sort of a decision is taken that they didn't understand, they might just call it Creepy. They don't really know, they don't have like the contours of where, of how they should feel about these systems and where they might affect them.
26:45 Speaker 02 So it's kind of like, one of the roles of creative people is to sort of map out that landscape and, and, and also to actually map it out emotionally. So
26:57 Chris Benson? Would this be kind of in the, in the, in the idea of design thinking in terms of applying that methodology to this creatively?
27:04 Speaker 02 That's a good way of thinking about it. You mean like
27:06 Chris Benson? Kind of going and analyzing, you know, what you're trying to get to almost from scratch? And then figuring out how this fits in? Is that where you're going with it?
27:15 Speaker 02 Yeah. So it's interesting. We, we, we try to think about the impact that these projects will have with their intended audience. And we work with the people that we award in this way to really get them thinking about that. Who is this for? What, what change do you want to see in that person? So you might just say, it's like, oh, it's for, like, we talked to Noah about this and he was like, oh, it's for millennials and I want them to feel angry. It's like, okay, so that's going to shape how you create that work, you know, the, the platforms that it goes out onto, the sort of references that would be included in it, the tone, the length, all of that. And so it's important for creative people to be leading the charge
27:57 Speaker 02 in that because that's their job and their work is to create a reaction in people. And so we kind of need that in these sort of early days of kind of creating a multidisciplinary approach to responsibly building these AI systems because we need to know what people think about AI in their lives and what people think about machine learning. And especially if we, we, we know some of the right policy interventions and some of the right design questions that needed to be asked. We need to quickly help the public catch up to where people at the leading edge of this stuff have been thinking. So if you think about like all the, you know, the questions of bias, there are like really concrete proposals
28:39 Speaker 02 of how you can design these systems, but they're not gonna, they're not gonna get traction if the public doesn't understand them or understand the urgency to them.
28:49 Chris Benson? So where does that take you? So if you, if you start getting a handle on a public perception of these technologies and how they're affecting their life, what do you as an artist go do with that? How do you take that new information and do something productive with it?
29:04 Speaker 02 I think it's, you know, it's not necessarily about, let me see, I want to dig into that one. I think this work is the most effective when the makers really understand the change they want to see in a, in a member of the audience. So if you, you know, going back to that example of Noah, if you feel like you want to anger this person, then what's, what can you expect that that person would be willing to do once you've achieved that emotion? Is it like, oh, you want them to delete Snapchat? Or you want them to write to their member of Congress? Or you want them to share it with a friend? Or you want them to, you know, complete a quick survey?
29:44 Speaker 02 If you're able to incite the curiosity or the emotion that you intend, you sort of have a little bit of a window where you can get people to do things.
29:54 Chris Benson? So if they have, at this point, a sense of awareness and some perception of that, that AI is involved, you know, going back to Noah's project where, you know, the cameras being used to do emotion detection based on, on facial reactions. Is that, does the awareness itself, in other words, if you are the viewer of that show and you're, you're looking at your laptop or TV with a camera on it, does that awareness change the reality that you're in? In other words, if I'm a viewer and I don't know that AI is being used in this and the, the movie is reacting, how is that different from if I'm a savvy person regarding AI? And I know that's happening with the camera that's looking at me right now and I'm still experiencing, how does that change the reality that the, that the person is engaged in?
30:37 Speaker 02 You mean like, how is it different if they just are told that versus if they like feel it in their bones? That, that, because the-
30:45 Chris Benson? We've kind of debated the idea of the public being aware of, you know, in quote, kind of how AI works and why is that important? And what, what is the meaning of our life when you have an educated person in the audience and they're aware of their experience being shaped by AI versus someone that's not? Where does it matter or does it not matter at all?
31:06 Speaker 02 So I think it matters. It's interesting, I was just in London at the Mozilla Festival and inside the tube, there was all these ads for- Programs that would teach young people how to code. And the way that they were trying to get kids to do this was like presenting it as magic. It was like, you have this magic wand. And if you know the spells, you can cast them and you can, you know, make the world whatever you want. And I think that's a terrible approach because it's not magic. It's actually humans that make really specific and concrete decisions that lead to really specific outcomes. So to answer your question, I think that what the opportunity that we have to show people how these systems are built is to realize that like, none of this stuff is a foregone conclusion.
31:51 Speaker 02 If we don't like the way that these systems make us feel, or we don't like the effects that they have on some of the more vulnerable members of society, there's an opportunity to change that. And when you see how it is working with your data or with, you know, an algorithm that's kind of like presented to you and you see like, oh, I get it. It takes these three things and compares them and then says, oh, okay, these two are alike and this one isn't alike. And you can sort of see how that can lead to things like confirmation bias. Or, you know, you can see the system, then you're much more likely to sort of say like, oh, this can be changed. This, we just, these people are just doing it wrong.
32:32 Speaker 02 Or, you know, like in the same way that like, if you're building a bridge, this is how you build it. And so that it won't fall down. And so we need to, you know, add those principles to the ways in which you design these AI systems. That like, oh yeah, you can't use data like that because it's, because it's clearly biased. Or you just give people the clear understanding that none of this is to be taken for granted. It's all up for design.
32:59 Daniel Whitenack? Yeah, I really glad you brought that up because frequently I have this moment when I am teaching like corporate workshops to people that haven't done machine learning or AI before. There's this moment and I, and I literally, I see it in their eyes where it's almost like a disappointment because they think that they were going to like learn something magical and level up and be wizards. But they really just find out that, you know, machine learning and AI, it's It's actually kind of a set of well-defined functions that you execute in code. And it's really just kind of a way of combining those in a certain process. And it's not like, you know, you sprinkle fairy dust over your computer and then the magical AI comes about. So I'm really glad that you brought that up.
33:43 Daniel Whitenack? I think it's an important thing, even for technical people to realize that this isn't, this isn't kind of a magical thing that is outside of our control, but it really is. It does have design behind it.
33:55 Chris Benson? You know, Daniel, I think you're totally misleading your workshops, though, because you walk into the classroom with the wizard's hat and the robes on and everything. I'm just saying, I think you're setting it, I think you're setting them up.
34:05 Daniel Whitenack? Yeah, well, Brett is wanting creative people to be involved, so I do my part. So I was wondering, you know, from that perspective, maybe Brett, you know, there's this group of like practitioners, like Chris and I, there's kind of researchers. And then there's maybe artists or designers or filmmakers or whatever it is that can help tell this story. How best do you think, or what opportunities are there for like practitioners like us or other people to maybe lend a hand in telling this story along with creators or designers? Because I'm imagining like these projects that you're funding, maybe the lead person is like a filmmaker or whoever it is, but like you were talking about with the feeling recognition thing,
34:46 Daniel Whitenack? there is a technical element to that that, you know, has to be built and figured out. So how can we as practitioners kind of engage with creative people to tell these stories?
34:57 Speaker 02 So actually, all of these projects do incorporate some type of AI within the actual creative approach. So they all actually have to have a real algorithm that does the sort of creative piece within it. So almost all of these, these folks that we're supporting are kind of these hybrid folks that are creative people and have an ability to tinker in engineering. Having said that, it's like, there's one that's really fun project. It's called Do Not Draw a Penis. And this is basically a comment on the Sort of algorithmic censorship process that you see in systems like YouTube or other kind of user-generated content systems,
35:42 Speaker 02 because all those systems are now applying machine learning to ensure that nothing bad happens on these platforms. They also-
35:52 Daniel Whitenack? Right, like nudity detection and certain things.
35:54 Speaker 02 They also, it has to be noted, employ thousands of people, unfortunately, to have to look at a lot of those images so that we don't have to. And they also employ some of these systems. Now, this particular project is basically, it presents you with a blank canvas, invites you to draw whatever you want, and knowing the internet, some people are gonna draw, guess what? Penises. And so the program will basically say, like, detect that and tell them like, "Hey, you know, you shouldn't be drawing that. This is a safe-for-work project." And kind of like turn their penis into a flower or, you know, a tree. That project uses sort of object recognition libraries
36:39 Speaker 02 for projects like Google Quick Draw. But of course, there is no penis detection within that system. So they've had to create their own huge library of, you guessed it, thousands and thousands of crudely drawn penises collected from the wide reaches of the internet. So my point here is, and I have one, is that a lot of these projects use existing libraries and existing approaches. So the more of that stuff that can be open source, the more creative people can be. They can sort of innovate on the content layer, if you will, of the project, rather than having to like create the entire stack
37:22 Speaker 02 of artificial intelligence systems that are needed.
37:26 Chris Benson? That is a tool that would make a big impact right now in my family's life. I have, I have a first grader and they're at that stage. And we're very active about going in the classroom and participating. And they're all at that stage where penises or anything else you can think of, farts, you know, are funny and cool. And they're trying to figure that out in the world. And a tool like that would, would actually be a delightful thing at this moment in my life.
37:47 Speaker 02 Oh yeah, I have a kid in grade two and it's like you just give him a blank sheet of paper and he draws the poop emoji on everything. It's like, dude, enough with the poop emoji. But so, you know, but like imagine an AI system that, you know, and each time he tried to draw the poop emoji is like, ah, you're not allowed to do that. I don't think we want to live in that future either where like his pen is embedded with like a naughtiness filter that when it like goes onto any blank sheet of paper, it sort of censors what he's saying. So, but that's kind of the reality.
38:21 Chris Benson? That's a good point.
38:22 Speaker 02 Some of these systems are building, you know, if we, if we assume and agree that the future creative palette of our children is going to be the internet. We really want to make sure that it's, that, you know, we're not preventing them from expressing themselves.
38:37 Chris Benson? That's a great point.
38:38 Speaker 02 I don't want to see penises over everything. And I, you know, don't want to live in a world where, you know, like certain combinations of lines cannot exist on a blank canvas. Tough stuff. We're going to solve it from Mozilla. You heard it here first.
38:57 Daniel Whitenack? Awesome. Awesome. And I think like going back to one thing that was mixed into that whole conversation was you mentioned open source. And I think one of the things that I got out of that was one of the ways that maybe we as AI practitioners can kind of help tell these stories and help, you know, make more transparent how AI works is actually to, to put our work out there in an, in an open way and to, uh, you know, create, you know, tools and documentation and pre-trained models and, and those sorts of things and put them out in the open. So creative people can use them and try to understand what they're doing and what they're capable of and that sort of thing. Would you agree?
39:38 Speaker 02 Yeah, absolutely. All this creative exploration that we sponsor at Mozilla, I mean, we encourage people to release their stuff under open licenses and in fact, sometimes require it. But it's so true that like you can just get so much further if you can, A, see what other people have done in this realm and then B, like, oh, there's a library for this exact thing that I want to do and you just go and get it. Yeah, absolutely. Totally agree with you. That's just, that's a thing that the community can really help with. And then, you know, like this other ways of like peer review or, you know, you know, just participating in systems like GitHub when people have questions or pull requests or stuff like that. Just be active and know that you could never anticipate that somebody
40:22 Speaker 02 is going to build like a facial recognition movie out of a library that you make. But sometimes those kind of weird uses that they didn't need to ask you for permission for are some of the most delightful and unexpected things that can happen on the internet. So do more of that.
40:39 Daniel Whitenack? Yeah, I think one kind of interesting piece of this that I'm kind of curious to see is as more creative people utilize a lot of the open stuff that's out there, I think that they're actually going to be able to kind of help us probe some of these implications in ways that like the practitioners haven't even thought of. And I'm thinking of like the chatbot that Microsoft released that, you know, turned into a Nazi in however many days.
41:01 Speaker 02 The Nazi, yeah.
41:01 Daniel Whitenack? Or whatever it is. Sometimes the practitioners, the researchers, since the cycle between research and releasing, you know, like model code and all of that on GitHub is so quick now, I think having people probe those questions and think about the implications is also important, you know, for the other side. So it's important for us to release things into open so that creative people can use it. But it's also important for us to look at what the creative people are doing with what we're releasing because it can help us, you know, shed some light on the implications of what we're actually doing.
41:34 Speaker 02 I agree.
41:36 Chris Benson? So where can we find out more about the projects that Mozilla is funding through this program?
41:42 Speaker 02 I think that we tend to help support these projects at their release. So if you're following us on Twitter, that's probably one of the easiest ways. We have a nice blog post up about the awardees that folks could probably find fairly easy. And we're anticipating that these will all be released sometime over the course of the next year. Everything is meant to be done by June, but they'll all have different release dates depending on the complexity of their project. So I think just keeping in touch with Mozilla. We do have a mailing list as well that we tend to send the stuff out on if folks are not subscribed to Too Many Mailing List. But I think the best way for people to keep in touch with
42:24 Speaker 02 these projects is just to follow us on Twitter, @Mozilla. We also have a mailing list if people are interested. And there's a blog post that should give links to the creators of the project. So poke around and see what else these folks have made if you're interested in this type of work.
42:40 Chris Benson? This has been a really cool conversation for us, I think, in terms of the intersection of creativity and different forms of art and communication as opposed to AI. We got into where we talk about ethics a lot, but I think we need to have even more conversations about the larger world, people outside of just AI in the traditional sense doing this. So thank you so much.
43:00 Speaker 02 Yeah, it was a pleasure. Thank you to both of you.
43:06 Speaker 00 All right. Thank you for tuning into this episode of Practical AI. If you enjoyed the show, do us a favor, go on iTunes and give us a rating. Go in your podcast app and favorite it. If you are on Twitter or a social network, share a link with a friend, whatever you gotta do. Share the show with a friend if you enjoyed it. And bandwidth for changelog is provided by Fastly. Learn more at fastly.com. And we catch our errors before users do here at changelog because of Rollbar. Check them out at rollbar.com/changelog. And we're hosted on Linode cloud servers. Head to linode.com/changelog. Check them out. Support the show. This episode is hosted by Daniel Whitenack and Chris Benson. Editing is done by Tim Smith. The music is by Breakmaster Cylinder. And you can find more shows just like this at changelog.com. When you go there, pop in your email address. Get our weekly
43:50 Speaker 00 email keeping you up to date with the news and podcasts for developers in your inbox every single week. Thanks for tuning in. We'll see you next week.
Previous Getting into data science and AI Next UBER and Intel’s Machine Learning platforms Page updated October 26, 2026 · transcript and summary are machine-generated