One of the problems with creating the avatars for the DGR (Twitter) accounts is that they may look like someone. This is/was never the intent and something that I need to ponder over.
We've asked Market Research experimenter/Communication Planning Explorer and facilitator of brands narrative needs, John Griffiths to chip in with the DGR commentary and development. Here his first post with a couple of demands.
It really is a bit like that. The urge to play to combine simple elements to see what complicated behaviours emerge. And if some kind of Turing test is passed, will the bot be believable.
So here's my wishlist. Which probably reflects my having studied philosophy a long time ago. Bootstrapping and counterexample figure large.
1/ Social media geek. Twitter and blogdom are awash with them. So it shouldn't be too hard to mug up a bot which talks the talk. This should be straightforward enough from keywords. What I want to find out is whether the bot will be insightful, hilarious or a bore. And right now I really can't call it. So try and see what happens.
2/ Call me crass and commercial but the reality show which still grips the teen audience is The X Factor. And thar's money involved. So why not have an The X Factor groupie aged around 15 who watches the show and all the add-ons religiously, votes every week and still things the show is about talent. We could tune the bot towards a fave judge - I favour Cheryl though a conflict of interest between a critical Cheryl and a pet vocalist might also be worth a look. The point about this bot is that we don't just want an enthusiast for the show but one for whom opinions matter These 2 represent 2 extremes. One commercial - because if every TV show had its complement of fanbots well I'd say that's a monetisation strategy. And if this Demo Graphic Replicator thing is to work then we may as well take a risk and make Pygmalion. Easy to pick demographic splinters we enjoy patronising.
But how like us could we make a bot. And would we like it if we did? And could we recognise ourselves? The ultimate Turing test.
Marcus' excellent post about the nature and emergence of streamtelling, as opposed to the beginning-middle-and-end-dominated craft of storytelling, made me think about the role of back story in the formation of a character. Any writer who's ever developed a complex, rich and interesting character -whether it's for games, novels or film- has had the pleasure of mapping out the character's back story, i.e. its history, background and life leading up to the point where the story picks up on him/her.
As explained here, questions you should ask yourself when writing your character's back story are as specific as "What did your hero eat for breakfast? What does he carry in his wallet? What was the most traumatic experience of his life?". Detail matters. The more detail the back story contains, the fuller the character is likely to become, and the more interesting the resulting story.
The DGRs seem to be a perfect way to construct such back story. Their retweets give insight in their daily thoughts, actions and routine. They don't immediately reveal the major plotpoints or events in a character's life, but reveal their life little by little, thought by thought, retweet by retweet. By streamtelling -derived from the neologism "to streamtell"- these detailed dots, a writer can build an interesting back story to a character over time.
Great characters don't emerge overnight, they emerge over time, through detailed exploration of their background and lives. DGRs enable a writer to construct such back story through social intelligence.
I’m not going to try and attempt to map out the history of storytelling here because it would be long and I’m sure that there are hundreds of other people out there who are much more capable of taking on that task. This does have something to do with storytelling, that is to say that this has nothing to do with storytelling because I’m going to try and explain what is going on with the Sibling that I have been building for the Demo Graphic Replicator project – Mr Felix Freeman.
In order to explain what’s happening with Felix, I need to quickly outline what I think storytelling is, and why what’s happening with Felix is different.
For me, storytelling is a robust method of communicating a beginning, a middle and an end. That’s pretty easy isn’t it? It’s such a robust model of communicating something that it’s been around since humans have had a brain and could use a finger to paint a Mammoth on cave wall. Storytelling suits the way we think; “oh this is the middle and I hope there’s going to be a happy end” etc. Regardless of how complex a story is, how many different characters feature in the story, or however episodes/versions/seasons/books etc. etc it takes to tell the story, it will always have a beginning, middle and an end.
Felix doesn’t, and yet he’s a fictional character telling us something that looks remarkably like a story.
The fundamental difference between Felix Freeman as character and any other fictional character is that his content, the fabric of Felix, isn’t being described by me as the author but is being coerced at random by a constant stream of demographic chitter-chatter. At heart Felix is an Ag8.com Demo Graphic Replicator, a twitter bot that grabs the tweets of people that make the most “emotional” sense to him. I’ve fed Felix with emotions and key words and Felix does the rest. The DGR code, becomes Felix's soul and this is how I put him together (you may want to watch this in full screen mode):
I have to admit that when I agreed to get involved in the DGR project I had preconceived ideas about who Felix was going to be because, at that time I was thinking in terms of storytelling. It seemed impossible to me that something could come about without heavily steering the bot but I was soon proven wrong. He just gets on with it and I’ve found myself confused by the stuff that I’ve been doing with Felix. The real breakthrough came with a Felix blog post (Felix still lets me write things now and then) called “Longfellow” in which Felix describes a bad night’s sleep. That post was directly influenced by a tweet that caught my eye as quite visual, so much so that I tried to find the street discussed in it and one thing led to another and we suddenly had something that was decidedly eerie and very, very real. We had a dream that you could play with:
Felix was beginning to live, but not in terms of a story. I'd already been playing with simple video sketches to try and get a feel for his personality:
But I wanted the stream of retweets to build him and not necessarily me - not in the classical sense of author. I then tried to shape him physically/digitally and created a very simple animation of the Longfellow text which worked, and I suddenly realised that I had probably not seen this kind of thing before.
It was only when David asked me to consider using an actor that I really started to panic. They had already opened up Felix to a broader public, invited people to voice over the animation and having quickly filmed myself in black and white and integrating the google map sequence of the Longfellow dream, we had our first 40 seconds of Felix film, a film that started as a tweet, which became of blog post, which then went on to become a script, then an animation and finally the film. There are three films in total, each one using one of the voices kindly submitted to be used in the animation. Here they are:
I’ll leave it to David to talk about what this potentially means for traditional storytellers (whether that be film maker, writers, game developers, advertising agencies, music, theatre, television, radio) although I’m utterly aware of what this means my self. I think they can just explain things better. But they did ask me, “what is this? What are you doing and how are you doing it? Describe it”, which I found quite difficult to do. On Saturday, however, whilst buying peanuts in a Supermarket it occurred to me that I’m not storytelling but streamtelling.
Streamtelling allows me to be guided by the character which is made up of what David calls vectors. These dots make up the emotional outline of Felix and it’s my responsibility to join them in order to make a characteristic silhouette. If we have more dots we get a sharper outline and the character becomes richer. The brilliant thing is that there are so many ways to do this and all of them influence Felix and his story. Thanks to this clever little thing that David put together, we can see, for example, what Felix is thinking right now:
So, this is streamtelling. I think it's very exciting. I'd love to hear what you think about all this.
Hello. David has very kindly invited me to post on the DGR blog, so hello to you all. As David wrote yesterday I’ve been busy building on my sibling character, Felix Freeman. Yesterday we asked for your help in finding a voice for Felix and so far we’ve had 3 voices in. The clip below is a very rough cut of Chris Hurst’s voiceover of this Felix post (which features a neat little interactive dream that you can play with), a post that was pushed along by this tweet. There will be proper edits of all the voiceovers appearing here very soon.
This is really exciting stuff and if you’d like to get involved with Felix please check out this post (just make sure I can actually download the voice over).
Artificial intelligence is the simulation of human decision making. Artificial affectivity is the simulation of human emotion. Emotion has historically been underrated in artificial intelligence. Without emotion, otherwise rational human beings cannot make decisions. Tellingly, the classic test for artificial intelligence, the Turing Test, is a test of software's performance in conversation rather than at number crunching.
Joseph Weizenbaum's "Eliza" system of the late 1960s featured a simulated Rogerian psychoanalyst ("Doctor", which most people just call Eliza). Kenneth Colby's early 1970s "Parry" simulated the conversation of a paranoid schizophrenic. Parry had a simulation of beliefs and mental state, but Doctor just echoed back the words of its user. Both manage to build an emotional rapport with their conversation partners until they make a mistake and break the user's suspension of disbelief.
Modelling human personalities and emotions as hierarchies of related concepts has a long history in psychology. But the various proposed taxonomies disagree even on what a basic emotion is, so there is a lot of work remaining to be done before we can describe emotion in a robust manner. This hasn't prevented these models being used outside of psychology.
You may have done a personality type test during a job interview, or for fun on the Web. Writers can buy books describing personality traits types to flesh out the characters in their stories. The pen-and-paper role-playing games that followed Dungeons & Dragons in the 1970s eventually added systems for modelling personality and the life stories that influence it. Few adopted scientific models although an article in "Challenge" magazine did so rigorously.
Rosalid Picard's 2000 book "Affective Computing" presented examples of emotional models and rule-based computer simulation of them to both parse human emotions and represent computer output using them. The Sims, the most popular computer game of all time in the early 2000s, contained a simple emotional simulation for each game character.
A switch from rule-based expert systems to internet-data-processing statistical methods can be seen in affective computing as well as in artificial intelligence. The best example of this is Jonathan Harris and Sep Kamvar's project "We Feel Fine" which extracts statements of the form "I feel X" from blog posts along with information about who has made this stament, when and where. This affective sample of the blogosphere is then made available for other software to use through an API.
The DGRs use We Feel Fine's API to set their inner emotional state. They search for users that match their demographic information, then (if they find a match) they choose words as tags that represent how they express or act on that emotion. The DGRs are not artificial intelligences, they do not make practical decisions on the basis of a disembodied and disinterested emulation of human intellect. They are artificial affectivities. To ask whether they actually feel is like asking whether a submarine can swim (to borrow Edsger Dijkstra's response to the question of whether computers can think). They are performative, functional equivalents to human emoting within the limited affective bandwidth of microblogging.
As a follow up to the sampler of audience responses, I thought this might be a useful way to show the feedback the Siblings are getting from across Twitter.
The FriendFeed is aggregating RSS feeds of Twitter searches for their names (@krankychloe, @Felix_freeman etc) in real time. You have to look at each posts footer to see who the responses are aimed at.
I'm tempted to do the same for Direct Messages that the Siblings receive - as the review of hardcore spam is quite interesting but I'm not sure if publicly posting DM automatically is socially acceptable? Doesn't seem to be anything in the Twitter TOS about that.
Transmedia is the aching hip scene for content producers - it's a way to develop stories across multiple platforms to 'engage' an audience on behalf of a sponsor.
The DGR system is different - it's driven by passive audiences (bloggers and microbloggers) and then synthesized into character, which then generates narrative.
I say narrative rather than storytelling as the later is a predefined tale, narrative is something that evolves, mutates, unfolds. Creating Transmedia narratives is also requires Free Culture principles because the cross platform spreading and scaling needs to have freedoms of use built into the design.
What is happening here is akin to Wikipedia, culture is being filtered into a piece of media just as every subject matter is being converted into a web page, which can then be reused by anyone due to the Creative Commons Attribute ShareAlike licence.
Marcus says,
"Although this little clip is a very rough sketch (that’s not Felix’s voice, I have a very different voice in my head for Felix) it helps to make Felix just that little bit more real. This is quite an exciting step. Suddenly the full potential of Felix as a character can be seen; you can see him in other contexts, like games or little short films and he is suddenly much more human that he was a twitter bot. He’s becoming the most interesting character I’ve created to date."
Felix in video form is not a million miles away from 80's cult figure Max Headroom - embodying satire within simulacra, portraying the locus of the loquacious.
Which is useful for all kinds of commercial arts endeavors when trying to catch the waves of culture - because it's the real thing.
When Happiness Factory arrived on our screens (your mobile, *tubes, teleboxes, ipods et al), the premise was to extend it beyond the 30 second advert - a universe of characters awaiting to be discovered, played with, promoted and loved. Lead by masters of 'finding the brand's voice' - Weiden and Kennedy, we're seeing socialized media unfolding as immersive storytelling. As beautiful as it is, it's not 'opening happiness' in a way that culture operates well. Culture is emergent, immersion is a static spectacle.
The dominance of a centralized editorial is something that transmedia producers are questioning - "What does a transmedia audienceship look like?" is a good starter. It questions the canonical authority of storytelling because the influence of the audiences' opinion is a reason for cultural production in the first place. Just as Twitter plays with the emotional disposition of marketing, voice and participation, the mechanistic voice of an authors creation through reproduction of their work opens the oppotunity for the reader to supplant their own voice within the characters profile. If the reader 'completes' the characterization, how far should that go?
So, if you fancy trying a little experiment for us, have a go at recording yourself reading the script of the film Marcus made - the one at the top of this blog post.
Soundcloud seems like a good service to upload the audio to and Marcus will edit in your voice to the original video. Make sure you apply this Creative Commons licence to your voice recording so that he can do this all nice and legally.
It will be fascinating to see where this goes. Leave a message in the comments below if you want to play along.
Another test subject, complete with over 500 Tweets. Lots of them were generated with test data, but this Sibling has a fresh outlook now. Go say hi to igguggogg.
# All fields are compulsory
# The twitter user name and password twitter_username: igguggogg twitter_password:
# Doesn't change. make sure it has imageid and feeling an a limit of at least 20 wff_search: http://api.wefeelfine.org:8080/ShowFeelings?returnfields=imageid%2Cfeeling&limit=50&gender=1&display=xml&city=hamburg&agerange=20
When I first decided to put the bots live, I was convinced that they would be met with human resistance, ignored, dismissed as spam, or worse, banned.
Part of the process for making normalizing them was to give them friends to follow and faces to present.
What I wasn't expecting was gratitude. Below is a sampler of Twitter users who are appreciating the RTweeting, which is not so much of a surprise - everyone likes to be ReTweeted, don't they?
Now, I'm not saying the DGRs are passing any Turing Test, but they do seem to be socially acceptable.
The Turing test is a proposal for a test of a machine's ability to demonstrate intelligence. It proceeds as follows: a human judge engages in a natural language conversation with one human and one machine, each of which tries to appear human. All participants are placed in isolated locations. If the judge cannot reliably tell the machine from the human, the machine is said to have passed the test. In order to test the machine's intelligence rather than its ability to render words into audio, the conversation is limited to a text-only channel such as a computer keyboard and screen.
The DGRs are not really participating in a conversation, they're cutting across. Which does seem to be creating (a positive, light hearted) confusion with some people.
I don't think the DGRs should ever be misleading, that is, deny that they are not human, but they do need to be respectful (non pestering), courteous (responsive) and useful (informative & timely).
It's the response article which is tricky - the bots have no sense of meaning, yet, so a casual "you're welcome" or an 'emoticon' reply to any @ messaging, runs the risk of being inappropriate.
One approach to this would be that DGRs only ever ask questions, never answer them. This may sound counter productive, but at their conception, they are research agents not search agents.
What I mean is that they are not harvesting, but relating, to live information - producing messaging that produces identity of 'their' character. They are, perhaps, (Method)actor models.
Bearing in mind, that the DGRs need to exist without a predefined 'script'...
The DGR builds up a persona through small amounts of information - at present - by using Retweets. It's about maximizing a character's development through the generation of the smallest amount of addition - specifically - the 'RT' prefix.
Bit by bit they begin to create a character's possible motives, direction and purpose.
Our visiting artist, Marcus, has begun to join these bits by building out a blog within the character's persona. Using the Retweets as the material to inform blog posts, he begins to give the back story, the mythology, the meaning behind the Retweeting.
Marcus is doing what any reader of the character's Twitter feed is doing - filling in the gaps between the tweets - joining the dots - making sense of the facts, as a fiction. As Rob pointed out, the DGR is a tool for storytellers, using probability and online structured data.
To set up the character, Marcus used this cool mind mapping application, MindMeister to map out the emotion to action relationships we use in the character definition file. Here's an embedded version of the map, which you can explore. It's a live file, so you're likely to see Marcus' changes over time.
Further, Marcus wanted to make a video through the eyes of Felix, to develop the persona, with a view to updating the characters parameters. By using Google Maps (Street View), you get a sense of familiarity with a sense of unease: this is the uncanny valley of experience, like the eeriness of the avatars that needs to be fiercely wrestled with at some point.
Here's the video journey of Felix traveling around his place of work in Liverpool Street, London, Music by Airborne Sound.
Bearing in mind that we'd like to play with AR somewhere in the future, this video gives a glimpse of what the DGRs could produce by themselves.
These sketches are not so much extensions of the character, but tools to design the character further. The crossover between using these tools to tell a fiction and to design a fiction is where the DGRs get really interesting because the characters are almostunproduct. Their persona flows between design and production as easily as information passes from one web service to another, adding a little more value to the viewers experience every time it's gets passed through a web service with zero overhead costs to the core system.
I say almost, as they are not really generating any new raw information themselves - they're parasitical (but non-physical) spimes at the moment producing metadata. We're still learning / deciding how best to implement their own, self styled, messaging - and for what reason.
But for now, they do seem to be generating droplets of persona every time they Retweet - so they are adding something to something, just by producing simple communications broadcasts.
We are also learning from the bots themselves. But it's a long process, because the bots are live on Twitter and we have to be respectful to the people they Retweet and be careful not to speed up the DGRs Tweeting rate too high. Also the patterns of Retweets are complex to decipher for narrative coherency.
This is where Marcus' sketches come in so useful - it's a form of testing to see where we can maximize the depth of the character's communications whilst keeping the logic of the system to a minimum. Unlike Conway's Game of Life, the simple rules of the DGR are influenced by the streams of social conversation (Blog posts & Tweets), adding infinite variables to our simple rule sets. This should make the character's persona rich, flowing and open.
Marcus' choice of design tools helps explore this context of the character and DGR development because his use of live web tools follow the development rather than inform it. Design as documentation, you might say.
Rob, the lead engineer on The Demo Graphic Replicator, has gladly stepped forward to explain some more about the reasoning behind the design. It's a pleasure working with this kind of thinking. Over to Rob...
Stories have structure, and that structure can be represented and manipulated by computers. In the mid-19th Century Ada Lovelace only imagined generating musical compositions using programs, not stories. But as early as the 1960s programs were being written to use Vladimir Propp's system for representing existing Russian fairy stories from his book "Morphology Of The Folktale" to generate descriptions of new stories.
The Artificial Intelligence programming of the 20th Century relied for the most part, like Propp's notation for folktales, on discovering and encoding rules. Narrative generation systems followed this lead, and tended to produce series of events that lack story arcs, character development or other high-level features of stories. The state of the art by the end of the 1980s is described in the book Possible Worlds, Artificial Intelligence and Narrative Theory by Marie-Laure Ryan. At best the systems described produce Aesopian fables without the moral; narrative without narrative.
In the 21st century, the rise of Collective Intelligence means that statistical methods based on information gathered from the activity large numbers of people have replaced rule-based approaches derived from expert opinion for many tasks. This isn't a new development for artificial intelligence or for generative narrative, Markov chains and other statistical methods have long been used to generate the text of short stories. What is new is the availability of vast amounts of structured text on the Internet for those statistical methods to be used on.
DGRs are story characters represented as lists of words that represent their personality, interests and activities. Each time the DGR is run it chooses some of those words randomly to represent its current interior state. It searches Twitter for tweets containing any of those words. Then it retweets the first found tweet to express its own current situation in the story.
The tweets may contradict each other over time, but at a demographic level they will tend to match the character. A significant amount of the retweeted tweets will match the DGR's personality and situation, and incongruous retweets can help to shake up the story. The DGR is a probabilistic character in a probabilistic narrative.
The breakthough with DGRs is to recognise that although some retweets won't fit the character the average of the retweets will be "good enough". The reader can fill in the gaps and file off the rough edges, rendering the character imaginatively in their mind as they would with a scripted character in a story written with a pre-planned plot by a single human being. They give the storyteller a powerful new tool to work with, and they give the reader engaging new characters to imagine and follow.
Well, it's taken about a week, but one of the Siblings has finally drifted under the nose of a Social Media Rock Star, in this case, Guy Kawasaki.
Guy, if you've wandered over to this blog - Hi! - leave us a comment - you're opinion is valued here. Maybe the siblings could have Alltop.com pages? Would be awesome if you provided an API.
darrylmorris @JamesMonaghan I think the profile pic on this looks a lot like you!!! No offense!! :D xx - @igguggogg
JamesMonaghan @darrylmorris The profile pic on what?
