00:00:06.919
Thank you so much for having me. I really enjoyed my first time in Sofia, and I have really enjoyed this conference so far.
00:00:17.880
Today, I'm going to talk about what I learned from building a Twitter art bot. This is what I built: it's called an Egg Spot. It's a bot that works together with the database of a museum in Amsterdam called the Rijksmuseum.
00:00:30.480
Usually, people have a demo time a little bit later in the talk, but I'm going to welcome you right now to try it out! What you do is tweet one word, and you can also try more words, but maybe start with one.
00:00:41.460
The bot uses that word as a search term in the Rijksmuseum collection and responds with an image. All the images I used for this presentation are from the Rijksmuseum.
00:00:56.250
I had a lot of fun looking for illustrations to represent certain things, and I was trying to think of a good way to use these beautiful old images to illustrate bots. This is one of the best images I came up with—it depicts humans imitating each other in various ways.
00:01:14.040
I was actually surprised when I looked for a definition of bots, as there aren't really any good ones out there. Here's one: a computer program that works automatically, especially one that searches for and finds information on the Internet. I also really liked this association from the dictionary that referred to it as a 'killer app.'
00:01:36.990
I started working on this project a little while ago, and there was a bit more of a sense of innocence around Twitter back then. It was sort of a fun and playful platform. However, it has since gone south, and Twitter bots have been in the news cycle for all the wrong reasons. This is a headline from just a few days ago, stating that bots, especially Twitter bots, have been involved in swaying elections and causing trouble.
00:02:04.430
We've all been annoyed by them on Twitter. One well-known example is a project that involved building an AI-based Twitter bot that would absorb human language and mimic the behavior of people who tweeted at it. While it started as an experiment to understand human interaction, it was hijacked by an army of trolls and quickly began spewing out racist and sexist comments.
00:02:43.660
This is one of the nicer tweets I could find from that project, which serves as a good example of how a Twitter bot can go awry. Like many AI systems, it ended up reproducing human behavior, which can be dangerous.
00:03:00.030
However, I was inspired by something else when I started building my bot. I have a background in art history and worked in museums for many years before becoming a developer. One thing that inspires me is the idea of sharing data freely within cultural institutions. For instance, the head of the digital collection at SF MoMA stated that by making data public and usable, the museum's API becomes a platform for open exploration.
00:03:30.570
This concept is part of a movement called OpenGLAM, which stands for Open Galleries, Libraries, Archives, and Museums. The idea is that since we've already partially funded these cultural institutions through our taxes, we should also be able to access and enjoy the incredible content they house.
00:03:53.560
Inspired by this, I wanted to create a project that dealt with some of this data. I knew early on that I wanted to work with the Rijksmuseum API. I explored a few museum APIs, but I chose this one because it was really well-documented.
00:04:05.250
So how did I build it? If anyone wants to create a Twitter bot, I encourage you to go ahead; it’s pretty easy, especially with Ruby. You just have to create a client to communicate with the API, and there is a fantastic Twitter gem built by former Mint or Mine member, Eric. Then, you need to set up keys in your client to interact with the APIs.
00:04:25.830
In my case, I’m working with both the Rijksmuseum API and the Twitter API. After setting up the keys, I implemented very basic methods—one called 'get image' and another called 'reply with image.' To prevent the bot from endlessly replying to itself, I made sure to limit its responses.
00:04:47.100
I really appreciate this project because developers often document their trials and errors through platforms like GitHub. In this case, the Twitter account itself has a nice log of its development process. The first tweet simply said 'hello,' and subsequent tweets demonstrated the progress, including some rough interactions with the API.
00:05:06.990
I left all these tweets on the timeline because I think they’re part of the story of creating the bot. I probably should have started with test-driven development, but instead, I dove in and started implementing features. My testing involved tweeting things like, 'Does it work now?' over and over as I iterated.
00:05:37.390
After initial success, I faced a significant issue. I had set a default value for instances when the bot couldn't find an image for the word used, which caused it to respond with tulips repeatedly. While this was amusing, it also highlighted a problem: the bot could only function with search terms provided by museum employees, so its flexibility was limited.
00:06:05.460
Recognizing this limitation, I wanted to enrich the bot's responses. Therefore, I created another method called 'random image' to ensure that when it couldn’t find the specific term, it would at least provide something interesting to keep users engaged.
00:06:26.520
I really wanted to talk about the challenges I faced during this project. This was a side project I did for fun, and I learned so much. Working on larger projects with established teams usually streamlines the setup, but here, numerous things went wrong, and those errors turned out to be the most valuable learning experiences.
00:06:51.600
One major misstep was accidentally exposing all my secret keys on GitHub, which was an eye-opening experience. It was a mistake I certainly won't repeat. I learned how to rectify it and discovered that GitHub has a way to remove sensitive information from a project's history. It's a complex process, but I eventually found a tool called Repo Cleaner, which simplified the process significantly.
00:07:14.060
Another fail occurred when I inadvertently caused the bot to enter an infinite loop. I tweeted excitedly that users could now engage with the bot to receive images from the museum, which led to some unintended consequences. The bot started responding endlessly to itself, creating an infinite feedback loop.
00:07:45.440
Within minutes, Twitter blocked the account. The bot was set up with a sleep timer, which prevented it from tweeting too quickly, but within just a few minutes, it still managed to issue around 25 tweets. This incident gave me a newfound respect for Twitter, as they acted swiftly to shut it down.
00:08:05.440
Luckily, the issue was resolved, which was another fun learning opportunity. This project was built during the Code and Cake event, and it was my first venture into combining my passion for art and cultural institutions with programming. It certainly won't be my last.