Notes from watching Bot Summit 2016 from my home
- The talks
- Darius Kazemi – Welcome
- Doug Dodds – Computer art at the V&A
- George Buckenham – Tools
- Esther Seyffarth – Bots, translation, and culture
- Michael Cook – Don’t be human
- Matthew Plummer-Fernandez – Making the unprintable printable
- Helen Hester – I’d rather be an iPhone than a woman
- Martin O’Leary – Epic botmaking
- Emma Winston – Performance of GraphicScoreBot
- Star Simpson – Building Personable Machines (Lessons in affect from physical bots)
- Ariana Giorgi – Journalism bots
- Dave Lublin – Bot TV
- Shardcore – What is it like to be a bot?
- Katie Rose Pipkin – Bots and the rise of digital folk art
- Darius Kazemi – It’s the Money / Outro
I clued into Bot Summit kind of late, as I saw my Twitter feed explode on Saturday. While I haven’t been into the bot scene for very long, my awareness of this community has been steadily growing thanks to my longstanding interest in procedurally generated content. I’ve also been making bots at work recently (although not full time), so my interest has become both personal and professional. Upon realizing that this was a really cool thing that was going on, I dedicated some time to give a detailed account of the content that was discussed. I hope that this will help more people experience what ended up being an incredible series of talks. Four hours is a lot of time to dedicate, so if this can help you decide what you want to watch, then great! My hope, though, is that you’ll read this and realize you just need to watch the whole thing :-)
I assembled as much of the supporting material that was mentioned in the talks as I could (as well as slides, when provided), with a bias towards material related directly to bots. So even if you want to watch the entire conference, this should still act as a handy companion. So far the conference is only available as one unbroken video, so I’ve added links to the start of each talk. Hugo was also kind enough to assemble more supporting material, which includes lists of Tweets about Bot Summit as well as IRC logs.
While I did my best to represent all talks as well as I could, some summaries vary in depth based on the amount of content presented or how easy I found the talk to summarize (which varied by how much I was able to see and hear and as my own energy ebbed and flowed). My sincere apologies if I misunderstood or misrepresented any of the talks. If you feel that I did so, please let me know, and I’ll do my best to correct myself!
Darius Kazemi – Welcome
5 minutes – video link
Our esteemed host Darius Kazemi welcomes the crowd and introduces the sponsors.
Doug Dodds – Computer art at the V&A
15 minutes – video link
Doug Dodds is a senior curator for the Victoria & Albert museum in London, whose involvement includes the digital collection they have. The V&A has an impressive digital art collection dating all the way back to 1968, which includes one of my favorite digital artists, Desmond Paul Henry.
Doug’s talk gives a very visual overview of the digital art collection at the V&A. This was clearly an interesting overview, but unfortunately the camera was not focused on the projector so if you want to be able to see any of the art that he was displaying, you’ll have to search for it yourself.
Worth checking out, however, is the museum’s database of their extensive collections, which contains over 600,000 images. The database is accessible through an API which led to the always interesting Victoria Albert Bot on Twitter.
George Buckenham – Tools
20 minutes – video link
George Buckenham is the creator of Cheap Bots Done Quick, a platform that lets you write Twitter bots using Kate Compton’s Tracery grammars – no programming required! CBDQ got its start when George tried to give a workshop on bot making, and realized that explaining the setup took way too long (and had nothing to do with making bots).
“Poets are better botmakers than programmers.” - @v21 #botsummit— BotWatch (@TheBotWatch) April 9, 2016
This talk was based around George’s experience with CBDQ. They shared a few design principles for tools for people who want to play with something (and not necessarily be an expert in it):
- Make it simple to get something working quickly – It’s easier for people to learn more complex things when they already have something working.
- Don’t start with a blank state – Give people something to start with (and that they can learn from)
- Composability is important – You don’t need to provide unlimited power when you can provide simple tools that can be composed in complex ways. For instance, CBDQ’s SVG support arose because SVGs are just text, which is Tracery’s medium, yet the two technologies were never designed to work together.
A fun talk for a fun tool, with some excellent advice!
Esther Seyffarth – Bots, translation, and culture
11 minutes – video link
Esther Seyffarth is a German computational linguist who gave a short talk examining how bots are shaped by the language that they use. Esther kindly provided slides of the presentation, plus the talk in written form!
“there are no ‘normal’ words”—@ojahnn on language and worldview in computer-generated language #botsummit— Allison Parrish (@aparrish) April 9, 2016
Esther’s talk touched on cross-cultural and linguistic issues arising from idioms and the signification of words, and took a quick look at the kinds of personalities bots are being given and how they reflect their creators. They ended with a call for more diversity, linguistic and otherwise, in the bots that people make. Hear, hear!
Michael Cook – Don’t be human
14 minutes – video link
The tireless Micheal Cook (who organizes the awesome Procjam and whose Saturday Papers are the best thing for people who like to know what’s going on in the world of procedural generation in games) gave an excellent talk about being careful with people’s expectations.
“I've had a conversation with @wikisext but it can't understand me in a personal level.” @mtrc #botsummit— Corinna Gardner (@always__curious) April 9, 2016
Micheal relates their experiences with their bot, Angelina, who itself is a creator of content (namely games). They stress the importance of managing the expectations of the person who is interacting with a bot, and how the more human you try to make something the higher their expectations will be. Microsoft’s recent Tay missteps are a good example of how people were significantly more disappointed than if their expectations had been lower. Micheal’s advice is a bit more measured than their prescriptive title: Be aware when you’re raising expectations because when you do, you’ll find you need to obey a different set of rules. Definitely worth a listen!
Matthew Plummer-Fernandez – Making the unprintable printable
13 minutes – video link
Matthew Plummer-Fernandez is an artist who explores socio-technological interactions using technology. They gave a quick presentation of their collaboration Shiv Integer, which leverages the collaborative 3D modeling platform of Thingiverse to create Katamari-inspired creations generated by glomming other people’s things together.
Helen Hester – I’d rather be an iPhone than a woman
19 minutes – video link
Helen Hester presented the idea that bots which strive to act as personal assistants (e.g. Siri) have recently veered into becoming feminized.
Women's work was culturally invisible until machines took it over -@HelenHester #botsummit— BotWatch (@TheBotWatch) April 9, 2016
They show us that this hasn’t always been the case, as seen in Apple’s 1987 vision of the future. Helen contrasts Siri’s characteristics with the male gendered bot that is styled as a professional research assistant in this video and concludes that Siri’s role is closer to the mother character in the video than the bot’s.
Helen then relates to us Ian Botost’s concept of Hyperemployment, wherein technology has not saved us from doing labour, but has instead forced people to take responsibility things that were previously managed by other paid workers (correspondence, scheduling, etc.) They go on to relate that many women of the past were simply expected to manage such affairs along with their other jobs of managing the household and potentially working for pay as well. Helen concludes that this technology has really just brought the work that women have been expected to manage all along into the consciousness of men.
By their own admission Helen had much more to say, but was pressed for time. It’s fortunate for us that they gave a much longer talk on the subject, last year. If you enjoyed their Bot Summit talk, you should definitely check this one out!
Martin O’Leary – Epic botmaking
23 minutes – video link
Martin O’Leary is a glaciologist (a profession that I didn’t even realize existed!) who has also made lots of bots, like Bot-ston (which I think is my new favorite bot). While Micheal’s talk told us not to make our bots act like humans, Martin tries to answer the question, “How?” Slides of the presentation can be viewed on Dropbox.
“Tay was not pretending to be human. Microsoft was pretending that Tay was an AI pretending to be human” #botSUMMIT— BooDoo (@BooDooPerson) April 9, 2016
First, though, they explains why this is even an issue. After all, rarely do we make things (paintings, cars, books) where there is any question of whether or not that thing is a human. The issue is inherent to the medium: bots by definition communicate through means normally reserved for humans. It’s this that skews people’s expectations of the interactions they have with bots. The question then becomes, “How do we manage people’s expectations?”
Martin proposes a framework based on Bertolt Brecht’s Epic Theatre – a formal set of principals that Brecht felt would lead to good theatre, which he defined as, “A form of theatre that promotes rational self-reflection and a critical view of the action.” Martin calls their framework Epic Botmaking and defines it as, “A set of techniques which promote a realistic view of bots and a bot-like view of reality.” To Martin, Epic Botmaking (and theatre) is based on the following principals:
- Theatricalism, the principal that the audience should never forget they are in a theatre. The bot corollary is that, “The audience should never forget they are interacting with a machine.” To do so we can:
- Expose the artifice – Make it clear that bots are not magic, that they are just a set of rules. Open-sourcing is not enough, because only a small subset of people can understand code. However, almost anyone can understand an algorithm that’s explained well. Darius does an excellent job of this for their Sorting Bot.
- Be explicit, not implicit – Explain where things are coming from. Wikisext makes it clear that it’s using Wikihow as its base for material, which goes a long way in showing that there’s no hidden intelligence lurking behind the curtain.
- Create a conflict of the elements – Offer information that does not supplement, but conflicts with the main body, with the aim of helping the audience more critically examine the elements that are present. An example of this is Yesterday News which tweets news headlines accompanied with seemingly unrelated satellite imagery of the earth.
- Verfremdungseffekt, roughly the “weirdifying-effect”. By making the audience feel detached from the event, we can promote self-reflection. Tools for doing so include:
- Using a fixed form – Use the same form over and over again, which exposes the underlying structure. Two Headlines has been telling the same form of joke 20,000 times!
- Pre-announcing: Let people know what’s going to happen before it does. That way, they can focus more on the details, rather than the event itself. everyword is tweeting every word of the English language. You always know what’s coming next!
- Present the unbelievable: Doing so can make people question their assumptions about similar things. manwhohasitall is not a bot, but exploits this form well by tweeting phrases sexist towards women, recontextualized to be about men.
- Historicization: Focus on details, not the universal concepts at play. censusAmericans tweets details from the US census, which gives us an interesting look at the stories going on behind the aggregate numbers.
Honestly, it doesn’t get much better than using mid-century German theatre theory to inform the choices we make about bots. A great talk that’s well worth the listen!
Emma Winston – Performance of GraphicScoreBot
7 minutes – video link
Emma Winston gave a live musical performance of their interpretation of a graphical score generated by GraphicScoreBot. GraphicScoreBot is a bot running on Cheap Bots Done Quick, which generates Graphic Scores. As a music theory keener, I was very pleased to be introduced to this interesting alternative form of notation. Nice performance, Emma!
Star Simpson – Building Personable Machines (Lessons in affect from physical bots)
18 minutes – video link
Star Simpson comes from the world of building actual physical robots. They’ve done research into building machines that interact with people in sociable ways, and their presentation shared the lessons they learned from that experience.
“Why can't we build a robot with as much personality as Tom Waits?” #botSUMMIT asks the tough questions.— BooDoo (@BooDooPerson) April 9, 2016
Star began with a brief survey of some important robots in the history of expressive robots. These robots often had facial features, and seemed capable of emoting. Some of the interesting lessons from these projects included:
- People had a hard time trusting robots (even if that was the goal of the robot!)
- People liked interacting robots with personalities closest to their own
- You could learn about a person by providing multiple paths that all lead to the same result. When choosing between “Hi”, “Good to see you”, and “Okay” as a way of advancing through an interface, people would subconsciously choose whatever matched their internal dialog.
Star defined bots as software with personality, and posited that things with personality are expressive, generative, and unexpected. The excellent Yellow Drum Machine (“This robot’s mission in life is to find flat surfaces and jam out on them”) was provided as an example of a robot that embodies these qualities:
- Expressive: It evokes emotion in the viewer
- Generative: It does something new every time, using feedback from itself
- Unexpected: It responds to its environment to produce new results
If you’ve never seen the Yellow Drum Machine, you’ve been missing out. Star did a good job of bridging the gap between the physical and the digital!
Ariana Giorgi – Journalism bots
15 minutes – video link
Ariana Giorgi is a data reporter who gave a fascinating overview of the way that bots are being used in newsrooms. Their talk was sponsored in part by OpenNews, which is a really cool sounding project. Ariana kindly provided a link to their slides.
Journalists can use bots to receive alerts, deliver news, generate data-heavy stories - @ArianaNGiorgi on journalism & bots #botsummit— BotWatch (@TheBotWatch) April 9, 2016
This talk opened my eyes to the fact that people are using bots in their professional (non-programming!) lives every day. Ariana broke down the ways that bots are being used, and provided great examples:
- Alerts: Reporters need to get information fast. It’s important for news agencies to be able to report breaking news.
- Google Alerts is arguably a sort of bot, customized to deliver you information you’re interested in. Reporters who are working on a particular area of information can use it to stay on top of the most recent stories.
- Sqoop: A similar service based specifically off of automated government-published data. Used to monitor patent, security, and court filings.
- Quakebot: A bot used by the LA Times to generate automatic news stories from real-time geological monitoring data. This lets people check the news the minute they feel something to get a sense of the context of what happened, without needing to wait for a human to compose something.
- Story generation: Quakebot also falls into this category. These are bots that can take some raw aggregate data and turn it into a more digestible piece targeted at a particular segment.
- Narrative Science: Is a company that works with clients to generate content relevant to a particular segment based on a (potentially) large amount of data. Examples given included generated reports that give details about the performance of individual school districts, and automated quarterly earnings reports.
- Automated Insights: Another such company that also generates quarterly earnings reports for the Associated Press.
- Chat bots
- BBC Weather Bot: Is a chat bot that lives on twitter, than tells people the weather in their region, when prompted.
- Slack bots: Of course newsrooms also use the now ubiquitous Slack bots. Ariana singled out the Dolores Landingham Bot, which helps to onboard new hires.
- Botkit: Speaking of Slack bots, Botkit is a platform for building them!
- Carebot : A bot released by NPR, that reports to journalists about analytics for the engagement of their stories. You can read more about it in this blog post.
Ariana was quick to reassure that these bots were not costing any jobs, but freeing reporters to work on things in greater depth.
This talk really delivered at providing convincing examples of the real-world utility of the bot medium! It was interesting to see both the tools that are being used professionally, as well as seeing how the strength of bots are being leveraged in a targeted way.
Dave Lublin – Bot TV
23 minutes – video link
Dave Lublin created TV Comment Bot, a bot that watches TV and comments on it. TV Comment Bot exists in a limited form on Twitter, but was originally a live installation called TV Helper that now lives in Dave’s home. Their talk was about bots and the future of television.
“The inventor can no longer understand their own machine” #botsummit— George Buckenham (@v21) April 9, 2016
Dave was inspired to make TV Comment Bot by videoinstrumentalism, which is, “The performed art of remixing audio visual materials into a live video feed” – analogous to live-coding or a solo musical performance. They also enjoyed the idea of Novice art blogger, which is a bot which sort of tries to describe art, but not accurately – instead it gives its own slant to things.
Dave started out with a description of how TV Comment bot works. It uses the hardware setup illustrated above to get the TV feed into a computer for processing. Once the computer has the image, it uses a (unspecified I believe) deep belief network to get nouns related to the image. The nouns then go through a custom framework (similar to Word Net and Concept Net) to accumulate synonyms and related words. Finally, these words are inserted into themed “Madlibs” style templates to create a phrase which is overlaid onto the image.
Dave was also kind enough to share some of the experiments that were attempted in the development of TV Comment Bot, but not used:
- Text recognition – Was found to be too slow.
- Speech to text – TV shows are not what these systems were designed for. They have too much noise going on, with music playing, sound effects, and people talking over each other.
- Face recognition/emotion analysis – Dave felt that it didn’t add the sort of information they were looking for
An important part of this talk was about moderation, and how it can be very much non-trivial. Indeed, TV Comment Bot was designed to be moderated, and only runs when Dave is around to monitor it. They realized this was necessary as they found the bot kept creating content that could be interpreted as tasteless or rude.
The moderation issue was a result of two factors:
- Since the content of TV can’t be well predicted, there is no easy way to limit the set of inputs to the bot. Even if its dictionary is limited, pairing certain words with certain images (which the bot does not really understand) can lead to unsavoury subtext.
- The bot has no real grasp on language, let alone cultural implications. Dave pointed to ELIZA, possibly the best known early bot, as an example of this. When people first learned about ELIZA, they felt like it would lead to a be huge revolution in AI. Yet its author, Joseph Weizenbaum, realized that it was proof to the contrary: he concluded that there was no general solution to understanding language, that context was always necessary.
Like any good TV show, Dave has made a number of TV Comment Bot spin offs:
- Star Trek TV Bot
- MTV Comment Bot
- Faces On TV Bot (just faces)
- West Wing TV Bot
- In development: Karaoke mode, which sings the comments to music
The talk finished of with a demonstration of what it’s like to watch TV through TV Comment Bot. Throughout the talk, Dave dropped so many interesting nuggets of information that I found I couldn’t include them all here. If you’re at all interested, you’ll really have to watch the whole thing!
Shardcore – What is it like to be a bot?
21 minutes – video link
Eric Drass a.k.a. Shardcore is an artist who’s made a tonne of bots! factbot, the lying bot, may be my second new favorite bot. Eric’s talk was a thought experiment exploring what it means to be and experience the consciousness of a bot.
“How many of you here reckon you have an immutable soul?” (No hands go up.) “Ooh, tough crowd.” #botsummit— Matt Webb (@genmon) April 9, 2016
The title of the talk is riffing off of famous paper about consciousness by Thomas Nagel, What Is it Like to Be a Bat? I went through this talk taking a bunch of notes, and in the end I decided that all I should say is this: If you are interested in the philosophical implications of bot consciousness (and who isn’t?), just watch this talk. It’s dense and subtle, and there’s no way my summary would do it justice.
Katie Rose Pipkin – Bots and the rise of digital folk art
13 minutes – video link
Katie Rose Pipkin is an artist who’s made quite a few Twitter bots, including the very popular ⋆✵tiny star fields✵⋆ and the excellent moth generator (co-authored by Laren Schmidt), as well as many more arts. Katie’s talk situated bots in the landscape of digital folk art.
“the inside of a thing changes the outside of a thing even if you try to hide the inside” -@katierosepipkin #botsummit— Darius Kazemi in NYC (@tinysubversions) April 9, 2016
Folk art exists in contrast to fine art – it’s accessible, non-commercial (while it may support individuals, it does not support institutions), and it uses readily available materials. Katie explains how folk art is ingrained into the fabric of the Internet. Not only is the Internet full of accessible material (open APIs, libraries, data), but it was also simply not seen as a place for “serious” art for much of its history. Despite increased commercialization of the Internet, plenty of online spaces still remain highly accessible. Indeed, bots are one of the media of digital folk art! They’re egalitarian, often not for galleries or commercial purposes, and there are open tools to help almost anyone make a bot. This is pretty crazy!
Katie went on to talk about their most popular bot, tiny star fields, which they wrote as a novice programmer. It was 200 lines of code with no loops (because they didn’t know about loops), and it read from a text file which contained star characters because they didn’t realize that they could just stick strings in the source code itself. While it’s since been re-written to be about 20 lines, Katie brought up the point that this naive approach led to some unintended behaviour that added to the personality of the bot. The field of programming has traditionally been interested in making the implementation invisible, but this can rarely be completely achieved. And indeed, in the case of folk art, perhaps a quirky implementation is even a highly desirable property!
Katie ended the talk by pointing out that Twitter has become a fairly saturated platform for art bots, yet the skill-sets used for making them – using code to make things quickly, and scalable – are widely applicable. They urge us to use these tools in our own spaces, even if that means branching away from the familiar. This was a super well composed talk, filled with smart, concise commentary that definitely makes my summary pale in comparison!
Darius Kazemi – It’s the Money / Outro
1 minute – video link
Our gracious host gives thanks to the speakers, the sponsors, and the attendees. Thanks to everyone!
I was super impressed by the content of all of the talks that were given! I really do urge everyone to watch any and all of the talks that interest you, because I they were packed with great information (a lot of which I left out).
It was fascinating to see the many different ways that people approach and view bots. There was no one reality presented, but a wonderful array of facets. The high quality and delivery across the board honestly surprised me. It was clear that these were a passionate, thoughtful bunch. Listening to all these smart people together made me sad that I didn’t attend…
Which probably means I need to go to next year’s bot summit! All this excellent content has got me pondering about making a talk of my own… Hit me up, Darius ;-)
My thanks to all the speakers and organizers for putting on such a great event.
· ✫ * . * .— ⋆✵tiny star fields✵⋆ (@tiny_star_field) April 9, 2016
✧ ⋆ ✦
· ✺ · ·
· * ⋆
. . * ✦