 Yn oedd yn digwydd, yn rhoi dwylo mwyn sy'n deunydd ym M Perm Prid ymryd II erioedr, ond felly mae'n byw meddwl, oherwydd mae'n holl ffagwcle, yn rhan o'n rhan o'r methau myth 720p. Yn oedd gan amlwg Femol, mae'n dweud bod y pethau sydd wedi gweld am hyn sy'n rhan o'r gweithfyrdd ..facilitated by smart phones, social networking... ..the creation of very large quantities of social data... ..and the computing power to work with that. So, actually, there's only... ..this pattern of behaviour you've only been around for a decade or so... ..and the term is still in evolution. So, I should give a rather sort of high level top. Well, basically, the intuition's this... ..that we have an idea that, you know, his society with machines... ..we have society which is the sort of the fleshy pastel shade soft bits... ..and then there's the silver, black plastic and metal bits... ..and we can kind of...that's the big division. And the idea is actually you could possibly start... ..starting on more meaningfully doing some vertical characterisation... ..of these agencies and sort of imply that this, this and this... ..each in their different ways have different kinds of agency. So, what we're kind of thinking of... ..we're thinking of what does human behaviour look like... ..in the context of networking technology... ..working at very large scales, a lot of distributed agency... ..self-organising groups, a lot of communication between people... ..but also between the technology, the technologies and people... ..and people mediated by technology, et cetera, et cetera, et cetera... ..with huge quantities of data. So, the term itself... ..you find the term in literature going back... ..Norman Mailer, for example, American novelist and Looney... ..writes about social machines in very disparaging ways. But the social machine research comes... ..there's sort of a quote from Tim Berners-Lee's book, Weaving the Web. It says, real life is and must be full of all kinds of social constraints... ..the very processes from which society arises. Computers can help if we use them to create abstract social machines on the web... ..processes in which people do the creative work... ..and the machine does the administration. That's quite an optimistic view. We get quite a lot of social machines in which... ..the machine does the creative work and the people do the boring bits. And then later on he says the stage is set for an evolutionary growth... ..of new social engines and he's talking about new forms of social process. So, that's a kind of 1999 prediction of the sorts of behaviour... ..that we're starting to see more recently. So, if you can imagine a graph... ..this graph actually works with it. You can do almost any research topic you like with this. But if you imagine that we get more and more people involved in computing... ..and more and more machines involved... ..this graph looks like this. So, in the bottom corner, there's conventional computation... ..the sort of stuff we use to people just working with laptops. As we get more people involved, more machines involved... ..then we get distributed computation grids, things like that. If we have not many machines but lots of people... ..then we start to look at things like social networks... ..and then social machines kind of start appearing in the top right corner. So, you might have scientific computation over there... ..but when you get more people involved, you start getting big data... ..and get it up the side as more machines involved... ..you start getting these kind of issues. And the general trajectory is just up into the top corner... ..and things like artificial intelligence, machine learning... ..and the internet of things which we haven't talked about today... ..which will be a big driver of all this kind of research... ..kind of pushing the large number of machines... ..and the large number of people involved further and further. So, these are the kinds of problems. So, when you think about not many machines involved... ..then you have your sort of crowdsourcing... ..or co-creation like Wikipedia, social networking. When we get into the top right, we start to look at things... ..like social responses to emergencies, crime, transport and so on. So, you start dealing with... ..or at least trying to address certain social problems. So, I'm not going to define social machines... ..there's a number of reasons I'm not going to do that... ..and this is trying to summarise it. I think it's one of these concepts where you're in danger... ..of either casting the definitional net too widely... ..so you include processes that aren't really very interesting... ..or too narrowly, in which case you'd miss out on... ..some rather juicy examples that you'd want to include. And this is partly exacerbated because we've got really, really fast changes... ..in technology and practice. Of course, interlinked as they co-create each other. And it would be very easy to write a definition... ..that completely missed some new thing on the horizon. Before social networking emerged in the middle of the last decade... ..I don't know anyone that really predicted it would happen. It's very hard to define in terms of communication... ..because communication is mediated by the technology. So, it's very hard to understand whether a person is talking to a person... ..via a machine, or whether a person is talking to a machine... ..who's then sending instructions to another person. The structures are quite difficult to disentangle. So, I like to think about what methods are you going to use... ..to look at a social machine... ..because that will tell you the sorts of social machines... ..that you're going to be interested in. I'm interested in social machines... ..where there's a very high level of social activity... ..where there's a lot of person-to-person interaction. And I don't mind fudging the definition... ..to make it clear that that's the sort of thing I'm interested in. So, I'm just going to give you a few examples of the sorts of things... ..because, actually, the examples are much more instructive... ..than wiffling on about distribution, scale and technology. So, here's one example, Ushahidi. So, this came about in the context of... ..there was a disputed election in Kenya a few years ago... ..and there was a lot of violence afterwards. It was very difficult to understand how the violence was happening... ..and mapping the violence. And someone had a very good idea of producing an interactive Google Map... ..that you could upload photographs of the violence, too. And in the technology via your smartphone... ..it's a very high level of smartphone usage in Kenya. So, because of the metadata in the photographs that were being uploaded... ..you could then place those photographs on the map. So, you actually got a real-time dynamic map... ..of where the violence was happening in Kenya... ..and it was validated through being actual photographs. So, that's an example of how you couldn't do that with the people... ..with just with people. You couldn't do that just with machines. Nobody knew the entire picture, but everybody knew a part of it. So, the trick was how to aggregate all those things... ..to produce the map of the Kenyan violence. And that was about 2009, I think, the violence. That was kind of an early example. Here's another example. This is Cajun Navy. So, there were floods in Louisiana a few years ago... ..and someone set up a Facebook group of boat owners... ..who would react to reports of people being stranded in the floods... ..and they would be able to do that. So, they coordinated a rescue operation entirely bottom-up. This was nothing to do with the authorities entirely bottom-up... ..across Louisiana. And that's been replicated two or three times... ..in other flooding incidents in the US. Here's a slightly more sinister one. This is Blue Servo. This is large numbers of people who are concerned... ..about immigration in the US... ..looking at CCTV photographs at the Mexican border. At the US Mexican border. So, there's no wall here, as you will see. I don't know why that hasn't been built yet... ..but this has been marked up as, oh look... ..there's these infrared images... ..which is interesting and possibly disturbing. Some social machines are utterly lunatic. This is 4chan think tank. So, this is the Boston Marathon bombing... ..and a load of 4chan users got together... ..to really examine the photographs of the crowd... ..to determine who the bomber was. A lot of people thought it was this chap here... ..with the baseball cap, because here he is. He's got a bag. And now look, he hasn't got a bag. And oh look, everyone's watching the race... ..and he's looking over here. So, it must be him. Of course, it wasn't him at all. It was a completely mad idea. But that's another example of... ..you couldn't do that with people... ..and you couldn't do that with machines... ..but you could do that with the combination. Here's a slightly more sanguine example. This is Zooniverse. This is Citizen Science. This is the idea that many, many people at scale... ..can do certainly classificatory jobs within science... ..far better than scientists themselves could do. Zooniverse is a platform for doing this... ..run by the University of Oxford... ..and they run many, many projects on this. And this is Snapshot Serengeti. The idea is that people... ..they have hundreds of thousands of volunteers doing this. Quite happy to look at photographs taken... ..by cameras set up on the Serengeti... ..and they identify the animals within that. This is the interface for that. So, there are lots and lots of social science projects... ..a lot of astronomy stuff... ..a lot of stuff... ..decoding old letters. So, handwritten letters in the 18th century... ..very hard for a computer to read... ..but people can do it quite well... ..and getting the interaction. I'll next slide, actually. Well, the AI is quite often social rather than machine. And what I'm trying to get over is... ..or what the talk will try and get over is... ..there's actually not much of a distinction between the two. So, but this is a slide of the Zooniverse platform. This is how it works. And this structure is very important. So, what I'm saying is you have a load of data... ..which in the case of Snapshot Serengeti... ..is the photographs from the Serengeti National Park. Your scientists, your actual zoologists... ..will decide what data gets put onto these Zooniverse platforms... ..which sends that data to volunteers. Now, the volunteers of whom, as I say, there are hundreds of thousands... ..so you've got very, very large-scale interactions... ..do lots of things. They try and classify these objects... ..but they also talk amongst themselves. So, that's the top right there. And they have lots of discussions and they find anomalies. They worry about particular types of concepts... ..and that's quite informative for the scientists. They also talk directly to the scientists themselves. So, there are loads of conversations going on. I mean, there are many millions of conversations going on. This data goes back then to the... ..their classifications go back to the Zooniverse platform... ..which then goes into a kind of AI module here... ..which is at the bottom right. So does all the stuff from the talk. So you can get a load of sentiment analysis... ..you can get a load of stuff about how particular concepts are evolving... ..what concepts are being talked about... ..what concepts are being controversial. And then the data science... ..you get a feedback to the volunteers. Basically, we'll adjust the data that the volunteers receive... ..viair the Zooniverse platform. And indeed, it can do other things like... ..you can also start to look for particular human behaviour patterns... ..so you can start to formulate hypotheses... ..about when a volunteer is getting bored... ..and is going to want to do something else. So then you'll send that person more interesting input... ..to keep their interest going... ..or possibly send them messages about exaltatory messages... ..to say you're doing a wonderful thing for science. So I don't know if that makes that clearer. They're all online. It's all the whole things online... ..so they just store the whole thing and pass it over. Yeah, so social activity happens... ..is it my... ..well, in fact, I'll just go on to this one. I want to draw a distinction between the platforms... ..on which discussions take place... ..and these are things like Twitter, Facebook and so on... ..and the social machines that take place on the platforms. So as another example of social machines... ..often talked of as Mechanical Turk. This is the Amazon's cheap labour crowdsourcing system. And you find that sociality breaks out... ..in all these things quite often. So Amazon's model insists that... ..the people using the volunteers in Mechanical Turk... ..in fact, they're not volunteers, they're paid... ..don't talk to each other. So they're trying to repress that social instinct. In fact, what then happens is you get a load of back channels... ..through other platforms. So people on the same task will talk to each other... ..by another route, which then Amazon can't capture. So the sociality of social machines... ..breaks out all over even if you try to suppress it. Zooniverse itself, the citizen science platform... ..is talking about actually encourages that. It actually encourages conversation talk. And it's original models in there. They've learned that by trial and error. They've learned over many examples and many projects... ..that actually if you provide a really good forum... ..for volunteers to talk both to each other... ..and to professional scientists... ..you get far better output. So that's a few examples of social machines... ..and this concept actually makes sense. So my colleague David DeRour of the Oxford Research Centre... ..who is the co-author of this slide set... ..teachers, in fact, at this very moment... ..is teaching a digital humanities course in Oxford. And what we find there is that... ..you don't need to do much explanation of the concept of social machines... ..for the students to get it. It does kind of make sense. A social machine built or designed by post-grads... ..based on an existing voluntary charitable network... ..in somewhere in Latin America... ..a group called the Madres and the Mothers... ..who basically try and feed homeless and poor people... ..using leftover food from restaurants. And they immediately thought, well, OK, we could build that for that... ..and that would co-ordinate the behaviour in various ways. And they were very quickly, in the course of like an hour's... ..training able to produce really quite complex designs... ..such as that one. How am I doing for time? I'm probably getting quite... ..five to seven minutes. OK, I will probably skip... ..I'll talk about... ..the metaphor is a metaphor of a machine. The term machine is obviously a metaphor. Those are the sort of metaphorical interesting things. The slides you can download from the conference system. So I won't talk about that. This is a slide about how social machines tend to group together. You tend to get ecosystems of social machines. But I just want to talk briefly about... ..so I'm not really talking about a method of something. It's sort of a perspective on a phenomenon. And the comparator I like to use is digital divides. It's very hard to pin down what a digital divide is. But actually it's a phenomenon that most of us see and can identify. And social machines is a similar kind of thing. So you need the kind of methods you want to use... ..to investigate social machines. You're looking at the same methods that you used to investigate crowds... ..or crowdsourcing network interactions... ..and indeed any kind of data or information processing. So it's a kind of descriptive layer of granularity... ..that's between the individual and the web. It's not a micro description of what's going on. It's not a macro description. These are the kinds of methods that you would approach that kind of data with. But I think more interesting is... ..and this is something that is very nascent... ..how would you theorise social machines? And there's lots of ways of doing that. All social machines are... ..exclusory in some respects and inclusory in other respects. That's an interesting factor. Gender, ethnicity, all those well-known divides... ..the big discussion, Wikipedia I think is a classic social machine. There's a big discussion about gender on Wikipedia. It's a very largely male dominated system. And there's a lot of discussion about how that affects... ..the output of Wikipedia classification. Social construction, technology, social shaping of technology... ..clearly very useful, et cetera, et cetera. Social capital I think is a very interesting perspective. And the sort of Marxian perspective of what's the value... ..what's the appropriation of labour... ..and how much of that value is going to the platform... ..and how much to the people in the participants of the machine... ..is an interesting question. So the writing of intelligence... ..I'll just talk about intelligence very briefly. This is a very interesting example. This is a contest. This was, again, probably about 10 years ago, DARPA, the American Defence Research People, set up this thing called a red balloon challenge. So what they did, they put 10 weather balloons across the United States in these area... ..on announced areas. That's where they placed them. There were nine states of the USA right across from east to west. And then they... ..so they basically said... ..the contest was... ..which there was a prize of however many... ..a thousand pounds for the group who could find all 10 balloons... ..quick, fastest. And this was the winning... ..the winning team was the team from MIT. And this was their pitch. And they said, look, we're giving $2,000 per balloon to the first person to send us the correct coordinates. And $1,000 for the person who invited them. And $500 to who invited the inviter and so on and so on and so on. And then it kind of goes on to explain. And what... ..that worked well enough... ..so Sandy Penton from MIT. So the model of rewarding whoever finds the balloon but awarding whoever invited the inviter is called a query incentive network model developed by Kleinberg in 2005. And they began with a team of four. They ended up with 5,000 participants. And the whole thing took 10 hours. And in fact the runners-up, who used a similar method, had found nine balloons in that 10-hour period. Which is, again, you couldn't imagine any individual, any group of humans, whether they were four or 5,000, doing that without the technological affordances. But equally you couldn't imagine the technology achieving anything like that without 5,000 people going around looking for balloons and inviting their friends. The key thing is there were two incentives. One, you had an incentive to find and report the balloon, but then you're in danger, of course, of getting people free-riding, people in competition with each other. So then the second incentive to introduce new people into the network kind of nullified that question of rivalry. Probably running out there. There was just going to be a few examples of how social processes in the finance sector are being used to add additional value to the AI in some quantified context, but I won't talk through those. So I'll just wrap up. So social machines is a kind of a lens on human machine networks at scale. We tend to expect very rapid and unanticipated assembly. So we expect the unexpected, partly because of the creative subversive powers of human beings who will never do what you expect them to do, but you can probably say the same of machines as well. We were all completely surprised when this technology worked and our slides came up first time. In terms of intelligence, what's happening is that the intelligence of both the machine and the human are being augmented by the social processes. Data's a really big issue. These social machines got to create and they're going to consume lots of data, lots of interesting issues about who gets it. And just the ethics piece is very important. There are ethical pieces about how you design these things and how do you manage them and what are your ethical responsibilities there. Research is clearly, you're looking at social interactions so there are lots of research ethics issues. Privacy is a given privacy is a problem with all these things, as Mark has already alluded to. And then there's the issue about, you see a lot of things that we might call the crime area, ways of conning people out of their money so you can get what we might call antisocial machines. Not all these things do good. Things have started to appear on the research radar, just a couple of projects, which are both just finished. Sociom is sort of an EPSRC project that you might want to visit their website. And Smart Society is an EU project covering pretty much the same kind of ground. That says, please don't sue me if I've stolen your image. And thank you very much.