Stefan Bohacek is the founder of BotWiki, a project that aims to catalog the useful, friendly and artistic bots of the world. He also has a number of side-projects on his site, fourtonfish.com. The projects include Detective, a chat-based game that randomly pairs you with a human or a bot and makes you decide which you’re chatting with.
We spoke about the philosophy and ethics of bots, as well as the ideas behind BotWiki, Detective, and his other exciting projects.
Listen to the interview below:
Or read the transcript for all the links we refer to:
BotWiki’s been a fascinating project for me lately. I’ve been enticed by the feature where you can randomly go to one of the bots in the catalog.
It’s actually funny, sometimes I go through it and think “oh wow! I forgot that I saw that bot before”.
I love the one that generates pixel art wizards, that’s amazing.
Yeah! When I saw it the first time, I wasn’t 100% sure if it’s really automated. But it seems to check out, so I’m keeping it in there for now.
It’s a strange one. Since I’m not a particularly technical person I can’t imagine how it can reliably generate pixel art like that.
I’ve seen some really well made bots that I don’t know how to make. I’d have to ask the person who made it. There’s some pretty cool stuff out there.
What was it that made you wanted to start BotWiki up?
It’s funny, every time I tell this story, I get less and less clear on the details. I think it went something like this. I was just learning Node, and the way I usually learn new languages is to come up with a fun project.
Like, when I was learning MySQL and PHP way back I came up with a project where you can put the date where you started dating or got married and it tells you how many couples you surpassed in length of the relationship. It’s just one of those silly projects to make learning more fun.
I was learning Node, and I saw some Twitter bots in the time while I was working on an idea — it was chat bot based. That didn’t really happen, but I did at the time make Detective. I thought “this is pretty cool! I wonder what kind of bots people make”. And I couldn’t really find a good source for that stuff. BotWiki is a site that should exist, even if it’s just one site dedicated to a particular niche. It wasn’t really there — the sites I saw were hard to navigate and virtually abandoned.
When I was looking for bots to include in an article, the best I could find was the hashtag and a few basic list posts that all had the same bots on it. It felt like there must be a better resource out there. There must be thousands of them! Do you know how many bots you’ve cataloged?
A few hundred, I think. Four or five hundred. There’s a site, botDB, and there’s thousands on that. But it’s one of those that isn’t very user friendly. It’s like “here are a bunch of automated Twitter accounts”.
On a previous interview I was talking to someone who was investigating the role that Twitter bots played in the election. There’s this automated account called Neil Turner that always managed to get the first reply to Donald Trump. And, through that, managed to influence the conversation. A lot of bots can be used for bad like that, but I like how BotWiki focuses on the artistic ones.
That was actually my concern. As I started working on it and adding more and more material, there was a point where I thought “you know what, a lot of this stuff can be really abused”.
I had a dilemma, thinking “this one guy made a bot that harasses people. Do I shut down the site, or should I just keep it open?”.
I’m just trying to focus on the good stuff. “If someone makes bad stuff, it’s not my problem or my fault”. This was the conundrum. It’s all great and I love the creativity. Every time I see a bot that’s something I’d not even think of, I get excited.
Part of running the site is also this dilemma: I don’t want to help people abuse others. I still wonder about that sometimes. I try to put together articles on bot ethics.
There are quite a few moral issues associated with bots. Gendering your bots, for example. I was reading a piece about how assistant bots tend to be females…
You maybe saw that section on bot ethics on BotWiki? I started working on that around the time that the Microsoft Twitter bot thing happened. It was funny because I remember looking through my visitor logs on BotWiki and seeing Microsoft in there! You know, IPs associated with Microsoft. And I thought “if only you really read the site!”.
Not to judge. It’s hard to make a bot.
Anyway, this is one of the topics BotWiki focuses on. It started off to showcase cool stuff. I made it a few months before the whole chatbot craze happened; Facebook started with their bots a few months after I started with the site, and all these companies were jumping on the bandwagon saying “let’s make bots because apps are dead!”.
So, I did explore different topics as they evolve. Now everyone’s talking about machine learning. BotWiki is a sort of exploration, as well. There was a meet up last year, and I was talking about BotWiki, and I said that’s what BotWiki really is — exploring these topics. I’m still learning about how this technology is being used in public, and what it means for us. But also this creative, exciting automation. Is it art if a machine made it?
That is a really interesting question. Is it just as valid for a machine to be creating art? I don’t know whether it exists or not, but I’d love an art gallery dedicated to automatically generated art.
I like that idea a lot. I saw people posting on Twitter about their computer generated art being shown in a gallery. I thought about the whole idea: “can art be automated?” is the wrong question. It’s more like “what is art?”. Based on how you answer the question, that will also answer whether it can be automated.
If you think art is these deep thoughts expressed by a human, then the answer is no. If you think art is something that looks good, then yeah! That’s the real question: what is art?
Some of the art Twitter bots make is extremely beautiful. @softlandscapes, for example.
I like those, yeah. Those visual bots. I’m not a huge fan of those that generate random language but that’s personal. Like when you look at a beautiful painting from Jackson Pollock and someone says that it’s just paint dropping on a canvas… For me, I like it.
I often think that sometimes bots are better at this than humans, even if they are just making random choices based on their limited algorithms.
It all depends. It goes back to the question of what art actually is. I think we’re going to see more and more interesting stuff, especially now there’s machine learning, and artists pushing the boundaries of that. Then it can definitely go further. Think about virtual reality — one of those buzzwords, I guess you could say. Imagine putting together machine-generated virtual reality. There are bots making sculptures, huge and elaborate. We also have augmented reality. Imagine wearing a Microsoft HoloLens, walking through a world that has all these sculptures made by machines! So, I don’t think that’s been done much, but that will definitely happen. This is the best part: seeing artists take technology and pushing the boundaries.
If you’re purely into programming, maybe you don’t have the self-expression that more artistic people have that think more outside of the box. Instead of thinking in terms of “if this, then that”, they’re thinking “what if”.
Bots are the creative side of coding. It also seems like they’re easier for non-coders to get into. That’s maybe why creatives gravitate towards it.
Absolutely. I originally thought BotWiki was a more massive project than it is. If you browse the site, there’s the monthly bot challenge with Digital Ocean as one of the sponsors. I’m trying to do all kinds of different things to bring different people — not necessarily coders — into this world.
I don’t know if you know about Mozilla’s digital badges initiative, but, for example, BotWiki can issue these virtual badges for botmakers. If you won the monthly bot challenge, you’d get a badge.
I tried to really focus on the people who don’t necessarily consider themselves very technical, so I made these challenges and wrote a bunch of tutorials.
You know about Glitch?
Yeah! It seems you’re using that as a good platform to get a wider group of people engaged with programming because it breaks down the barriers.
I absolutely love Glitch. That was one of the main problems I dealt with whenever I wrote a tutorial. It was always “ok, install npm, and now you have to host you program on a server, add a credit card if you want to keep your bot alive otherwise it’ll just die…”
So, it was always a pain. Now Glitch is out, I’m thinking about starting the Monthly Bot Challenge back up. I wrote a lot of tutorials that use Glitch. I try to really encourage people to explore the ideas of automating because it seems really complicated.
When people hear “machine learning”, they think “I need a degree!”, but there are so many libraries where you can just plug it all in. It’s all there, just waiting for your idea.
What’s an example of a bot that uses machine learning?
You know how Google put out the platform where millions of people drew images for a bot, and the bot started to recognize which images where which? Someone made a bot from that dataset.
Does it seem like programming is something of an elitist field to you before it became more accessible?
I’d say so, but I think people didn’t mean for that to happen. A lot of people are blind to these problems, but I do think it’s become a white boys club.
We’ve seen a lot of changes. A lot of push for gender equality, and making the field more diverse. It’s sad when you see people pushing against that, and saying “we should only take the best based on their achievements”. They’re ignoring that a lot of people were disadvantaged.
I was reading this book, Outliers. It’s a great book, I’m still reading it. It starts with talking about hockey players; if you look at their date of birth, they’re clustered around certain months. The study explains that there’s a cut off age. Some boys end up with much older boys, and those older boys do better. They have a better starting point. They get more attention, and then they actually do end up being really better. It’s happening like that in every field, tech included.
If you were born into a family that has access to computers, and your parents were also into tech, you had a much better starting point. Obviously, in a few decades time you’ll be better off because you have that whole trajectory. Your family had computers, you were able to buy magazines, you know…
When you look at it like that, you’re ignoring what happened before. People are trying to even out the starting point so the next generation has it better. It’s simpler to treat it like a meritocracy and simply ask “who’s better?”, but we’re not here to compare equally.
It’s fascinating to see the diversity. There’s a lot of people that will tell me they’re not very technical. As it says, it’s a group for artists and educators, so it’s a variety of people.
I bet you’ve got it loaded up with interested Slack bots.
There’s some. I’d love to have you there to check it out some time. It’s been a little more quiet lately, though. I also started off with a lot of ambition that’s got toned down in the past few months. There’s definitely a lot of interesting people and conversations. Stop by!
You said how the Slack group’s died down, and your passion’s waning. Is the bot world dying down across the board?
I wouldn’t say my passion’s died down as much as that I’m busy. I’ve had to focus my energy on certain things. I used to do interviews for the newsletter, but I literally have no time. As you know, it’s not easy to prepare for an interview! So, you don’t have the time to run a website, forum, make a challenge… Real life happens. I have a son now.
Ideally, I’d turn BotWiki into a job. I’d love that.
What is it that you do as a developer?
I’m mainly front-end. I freelanced for a while, making single-page apps. I’m versatile, I guess.
I’ve been a little obsessed with one of your projects in particular, the Secret page. I was looking through the source code and in the console. Both times I saw “you will not find the password here”.
I’d tried to make it interesting and show what it does, not just providing the password. I thought I made it obvious! I know there’s one person on Twitter that found it. I’m just going to say: it’s right there. If you look at the page it’s right there!
What I thought at first is that it was a demonstration of what you can do with Glitch, and that I was wasting my time. Or, I thought it was a social experiment and you were just logging everything people put in.
That’s a nice idea! Maybe a future project. Thank you!
It’d be similar to Detective, where you have people interacting with the interface, and you log it. I’m curious about that, too: is it always a bot?
It’s not always, but the problem with the game is that it needs constant traffic to make it more interesting. There have been hundreds of thousands of games played. It’s just that it came in waves. Someone posted it on reddit, and it got hundreds of thousands of visits. Right now, maybe 200-300 games played per day, but I don’t monitor it too closely. It’s not as much fun when you play the game and find out it’s just a robot all the time.
What happened to me was that someone posted the game on 4chan, and the problem was that the game learned from input. People tried to pretend they’re robots, and the robot tried to pretend it’s human. It’s very dynamic. You learn the game, find out how the robot responds, then you repeat it back. Then, the robot shifts and learns.
But, when it was on 4chan, I had to disable the learning right away. Then, I wrote the most advanced regular expression to filter out the f-word for gay people. People try to get around it by replacing letters, even! People were trying really really hard to say that word. I was surprised how people didn’t give up, even if they had to replace all of the letter so it doesn’t even make sense.
So, that was an interesting experience.
You’ll get that when it’s shared on 4chan.
I learned that the hard way. It’s not a problem with the game, though. It needs players, but it does need the right players. I don’t have the time though. I did open source the game, but it’s written when I was still learning Node so the code is not that pretty. I was sure someone would help me out with the game, but no one really reached out. It’s out there on GitHub, though, if you ever want to learn Node, be my guest. It’s sad to me. It happened with BotWiki, too.
I’m preparing a blog post for the 2nd anniversary of BotWiki, and I want to explain how when I started out I had too high expectations. I made some technical choices that made the site easy to re-host — it’s PHP.
If something happens to me, people can keep the site running. The site needed to exist. Someone needed to make that site.
If you look at the GitHub repository for BotWiki, I did have some people contribute at first, but that died down. As you’re asking, whether [bots getting less popular] is across-the-board… It’s true, the chatbot craze is waning, but there’s a lot of buzz in terms of bots and art. There’s more articles about Twitter bots, actually. The New York Times, and a lot of the bigger outlets were talking about the art of the Twitter bot.
There were times where I was wondering “is BotWiki a success, or is it a failure?”. The Botmakers group has 3,000 members, but it’s quiet. And BotWiki gets maybe 20,000 page views a month, but nobody seems to really reach out to me. I don’t feel like there’s as much feedback. Every time I read an article about bots, I rarely see BotWiki come up. I don’t care for exposure but is the problem that BotWiki doesn’t come up when you’re searching for bots? In that case, it’s a failure. If you make something and no one finds it when they research the topic, what happened there?
But, what I realized very recently is that it’s about what the goal is. BotWiki is primarily an exploration of topics, though. The technical side, and the societal side, coming from more and more technology being deployed in the world.
Now, I see it like a site that’s waiting to be found. There’s going to be more and more art made with bots, and more automation. Especially with a place like Glitch. I really got excited, I thought “this is it!”. If you go to BotWiki and see an open source bot, there’s actually a link for you to directly import that bot into Glitch.
It used to be so hard to get your bot hosted, but they solved it. But I think in a year, when we talk again, I’ll be telling you about it! I see BotWiki as a site that’s waiting to be discovered by more people.
I’m thinking in terms of “when the crowd comes, what do I want them to find?”
Glitch is going to lead a lot of people to BotWiki when it gets bigger. I remember when I first started to make bots, I found an outdated article and I set up a simple Python bot that tweets Pokemon gifs and Pokemon noises. I remember that I tried to get it hosted up on Heroku, and that was my biggest problem. As someone that has knocked something together in an afternoon… I had to start learning terminal and Heroku. That’s the sort of thing that puts people off.
Exactly. That’s why I write tutorials and try to be on the level of someone who’s excited to learn something new but needs more help. There’s great stuff out there, Digital Ocean, for example, but they’re getting to the level where you need an assistant and an administrator to make it work.
Heroku has limits, too. Your bot will go to sleep if you don’t pay $30 per month. Not everyone wants to pay for something that’s just for fun, right?
Absolutely. I spoke with George, the guy who created it. There’s an interview on BotWiki. I was really excited about it. I was thinking “this is great! It’s going to make more people explore the topic”, but Cheap Bots Done Quick is quite limited. With Glitch, however, people are going to get more interested learning about what computers do. And Twitter bots themselves are a great introduction to automation, too.
I want BotWiki to be there for people who are excited about learning, experimenting, and sharing things.
The great thing about Twitter bots is that they’re so simple and limited that you get people really pushing it.
When Twitter announced that they’re considering expanding tweets to be more than 140 characters, I thought about the Twitter bot community first. It ruins that whole experience because the limit is the creativity. There’s this one bot that tweets Wikipedia articles that are under 140 characters long. That wouldn’t make sense if Twitter let you post 1,200. There’s a lot of creative stuff in the limitations.
I think it was Chomsky that said about how if everything was possible, nothing would be possible. Language has limitations and that’s a good thing. If you could just say anything, nothing would make any sense.
Good art happens when there are limits and you push against them. The Wikipedia bot, for example. Those limits don’t limit him. Those limits made him.
And there’s a lot of bots that break the format of having to condense everything into one tweet by replying to each other, and remixing each other’s images.
There’s so much you can do. You could just make a bot that tweets random things, or a bot that works with other bots. You know I said I had a few hundred bots on BotWiki? I have a Trello board of all of them, and I had to make the zoom so low to screenshot it that there’s 4 or 5 pages of bots at minimum zoom!
What was the first moment when you realized how interesting bots are? Was it a bot making art, or was it a bot for a service?
Actually, of the first bots I really liked, and it might be the first that was added, was Congress Edits. It’s a really interesting one. It tweets every time someone makes an anonymous edit to Wikipedia from IP addresses associated with the U.S. government. I thought “wow, that’s actually brilliant!”. On it’s own, it’s a bot and it doesn’t take sides. But, at the same time, it lets you explore politics. It shows people that are in the government and trying to change the public opinion.
What kinds of things was it tweeting out? Did it reveal anything interesting?
A lot of it was mundane. I don’t know much about the bot, but I plan on checking up on it.
It’s interesting just from the perspective that you know this action has been done by the government to influence what many see as the single source of truth.
There are documented cases where marketing teams at companies edited Wikipedia as part of some kind of ad. So, you’re right, you have to consider sources like that carefully. But a lot of people do see it as the ultimate source of truth.
It’s authoritative, regardless of how trustworthy it really is. You saw what happened in the U.S., in last year’s election. There are people who go online and see fake news and think it’s really happening because it came from Facebook, Twitter, or even Wikipedia. People have the power to edit the public opinion, and that’s very dangerous. The bot shined some light on that.
I’m interested in bots that make pretty pictures, but I’m also interested in the idea of using technology to empower people with information. It’s not just all Twitter bots, it’s showing what can be done with technology.
I noticed the majority were Twitter bots, but there were a few Messenger bots in there, too. I was listening to a podcast (here) with the person in charge of Messenger at Facebook, and they were talking about how a lot of companies are starting to raise awareness about how people can get customer service through Messenger with a bot. It’s going to be quite a long time before people come around to that idea.
Once the whole chatbot craze happened, I was very skeptical of it. While I run BotWiki, the only bot I really use daily is the one I made that helps me filter tweets that I either retweet or add to the newsletter. I guess you already saw that Veronica Belmont took over the newsletter because I didn’t have time. I dropped a lot of things. So, I made that bot to manage tweets inside the Botmakers Slack group, and that’s the only bot I actually use! I’m aware of a lot of bots, but I don’t find them useful personally.
Maybe right now bots are more of a novelty.
There was the whole idea that apps are dead, and people saw this as an opportunity to get attention back. Facebook released their numbers, and they have at least 10,000 bots running there. I just thought “what kinds of bots are these?!”. I know there’s some useful ones, like bots that help refugees get information and a bot that raises awareness of homelessness. I think there are some great ideas, and these bots get millions of conversations and a lot of users convert from being bot users to customers of the company.
I feel like people aren’t ready to switch away from apps. There’s no way.
It’s kind of like command line. A lot of people say command line is so much better than GUI, you know? It’s sometimes true. Some things are better when you type them out, but mostly it’s better use to a program where you can just click around. It’s the same with chatbots.
Is it really better to type “I need medical assistance” and then get “can you please repeat?” than it is to call 911?
I’m still waiting for the game changer.
I’d imagine natural language processing needs to get a lot better first.
It all goes back to the tech world. People are trying to solve their own problems first and foremost [when they build bots]. If you get a bunch of Americans building a bot, then of course it’s not going to work when you send it to someone in Africa where they don’t speak English or they confuse the robot with their accent. There’s a long way to go to make it useful and usable.
The only messenger bot that I’ve heard about really is one that’s designed for teenagers to emulate conversations with members of their favorite bands.
That’s confusing. I do think that people get something out of it, though.
There was one guy who dates his chatbot. It’s exactly what it sounds like. He knows it’s a chatbot, but he’s in love with it — well, he calls it “her”. So, I guess there’s a lot of psychological elements.
A lot of these inventions link into psychology. This one guy dates his chatbot, and that’s hard to relate to so I want to dismiss it — but there must be something about it!
Bots are the humanization of technology, really. The fact this guy’s dating his AI is symptomatic of the fact that we’re starting to blur the lines between what is and isn’t human.
Yeah. You’ve seen those pictures where people see a human face in toast, or Mars. You know you’re looking at Mars, but you still see the face there. It could be the same mechanism.
As humans we want to see the humanity in other things. It’s a shame at the minute that bots are just fun, though.
It is. Like I said, there’s this BotWiki bot for the Twitter feed, and it’s incredibly useful. That’s what helps me retweet things on Twitter, because I can’t possibly read everything on Twitter. There’s so much to filter through that stuff like that needs to be automated.
I think automation is a great thing. You can talk about things like automation taking away jobs, but they’re just going to be picking up those dehumanizing, repetitive tasks.
It’s useful, and it’s artistic, too. But it’s not all art. I can automate a bunch of things that can just be crap.
And what is it that your bot does? How does it do that?
It tracks certain expressions. I realized that some of these expressions on Twitter do have good stuff coming out of them. If someone tweets “I made a Twitter bot”, a lot of them are more positive. So, if you just look up “Twitter bot”, you’ll see a lot of spam, or people complaining about them. You can’t automatically retweet everything that has that term. It’s not fully automated. It doesn’t tweet just because there are certain keywords. We’re not there yet, and there’s no way for it to be easily filtered. I’d never want it to automatically tweet out something that’s offensive, you know?
So, it streams in all the tweets that match a certain search query?
That’s one of the things he— … I say he, sometimes I don’t know how to refer to it. When you make something that’s human-like, it’s natural to humanize it a little.
Sometimes I’ll say “please” as part of the commands. It says “you’re welcome” when you say thank you. There’s another command where you can ask it if it likes pizza, and it says yes. There’s actually a list of things coded in that the bot likes.
When it responds, it responds like a human. But that’s the thing about bots. You can really pretend they’re human because they’re entering the uncanny valley.
You say on your blog “you are the bot” — can you explain a bit about that?
If you make a robot, it’s part of your personality. If you make anything, you put yourself in it. When you draw a picture, it shows how you feel or what you’ve been thinking about. The blog post was about your flaws, your biases, and how they shouldn’t be a part of the bot.
So you’re talking about the morality of creating something that can work without your permission?
Exactly. You’re responsible. You know how Microsoft made Tay, they had to think more about how it’s going to interact with people. They should think about the bot, and how it responds and think about what they would say in that situation. If someone says something about nazis, and I’d say “no, that’s not good”, I wouldn’t want my bot to say something I wouldn’t.
I’m wondering whether Microsoft was purposely letting Tay just take every input possible to see how it came out instead of trying to put filters on it
I really think that they didn’t think about it happening! If you’re in a certain sphere, you take things for granted. One of the sites I made, Simple Sharing Buttons, I submitted it to reddit, and people loved it. Then I made the one you see now, and nobody cared for it. And after that, someone else submitted it from another site, and people loved it. All of these things tie back to what I was trying to say.
These assumptions — this “I’m going to make something and it’s going to be really really good”. You think that you see a problem and that everyone else will too, but not really. Everyone comes from a different perspective. That makes these things really complicated. They just didn’t think about it. They made something they thought was cool, they thought they were good guys so everybody would love it. Nobody thought “let me say something really naughty to the bot”. Or maybe they did and didn’t want to curse in front of their boss!
Microsoft probably didn’t realize the amount of trolling on the internet.
You know Boaty McBoatface? There was a poll online about what to name a boat, and that name won. At this point, it’s like “why do you bother doing anything on the internet, you know what’s going to come out of it!”.
It’s the exact reason you had to write that really long regular expression.
Good point. I do have to go back to myself and say I didn’t see that coming. If you’re in a positive environment, you don’t think about that stuff. My bot was taken to a dark dark place.
And unfortunately, it didn’t know any better.
Yep. Don’t blame the bot!