:00:00. > :00:00.Coming up shortly on BBC News, we will have new Swatch but first it is
:00:00. > :00:32.time for Click. -- news watch. Summer is on the way and, well,
:00:33. > :00:36.it wouldn't be a British summer without a visit to a good
:00:37. > :00:38.old fashioned festival. Known as the Town of Books,
:00:39. > :00:47.Hay-on-Wye, in Wales, It's a literary mecca,
:00:48. > :00:53.an annual gathering of artists, authors, Daleks and,
:00:54. > :00:58.yep, even Royals. It's even been called the Woodstock
:00:59. > :01:02.of the Mind by none other than former US President Bill
:01:03. > :01:04.Clinton. This year it's the 30th Hay Festival
:01:05. > :01:09.and the line-up is pretty stellar. Well, for the second year in a row,
:01:10. > :01:13.we've been invited to share some of our favourite experiences
:01:14. > :01:16.and show off some really good tech, all in front of a real,
:01:17. > :01:23.live audience of actual people. A packed tent waited,
:01:24. > :01:30.all that we had to do was wow them! We have robots falling over,
:01:31. > :01:36.experiments in haptic feedback and demos in binaural sound,
:01:37. > :01:41.but that was nothing compared to the climax -
:01:42. > :01:44.a Click-created wavy, shouty game built using
:01:45. > :01:50.artificial intelligence. In the meantime, it can't have have
:01:51. > :01:53.escaped your attention that around the UK things are getting
:01:54. > :01:56.a touch political. As the general election looms, those
:01:57. > :01:59.politicians are using increasingly sophisticated techniques in order
:02:00. > :02:09.to learn more about us. The advertising reach of Facebook
:02:10. > :02:12.has long been an open secret, but now it's something the political
:02:13. > :02:19.parties are getting in on too. In fact, both the Trump campaign
:02:20. > :02:22.and the Leave.EU groups credited Facebook as being a vital part
:02:23. > :02:26.of their electioneering. We know that the personal details
:02:27. > :02:30.that you give to social networks allow them to send you relevant,
:02:31. > :02:33.targeted content, and it goes much deeper than just
:02:34. > :02:37.your basic demographics. There are now data analytics
:02:38. > :02:43.companies claiming to be able to micro-target and micro-tweak
:02:44. > :02:45.messages for individual readers, If you know the personality
:02:46. > :02:51.of the people you're targeting, you can nuance your messaging
:02:52. > :02:54.to resonate more effectively What's also emerging is that
:02:55. > :03:02.political parties have been using this data to reach potential
:03:03. > :03:07.voters, on a very granular level. So who is being targeted
:03:08. > :03:10.on Facebook and how? Well, until now, there's been
:03:11. > :03:16.nothing around to analyse any of this, but the snap general
:03:17. > :03:18.election galvanised Louis Knight-Webb and Sam Jeffers
:03:19. > :03:21.to develop Who Targets Me, a plug-in to tell each of us how
:03:22. > :03:27.we're being targeted. When you install the plug-in
:03:28. > :03:29.for the first time, it asks for your age,
:03:30. > :03:32.your gender and your location, and then it starts scouring your
:03:33. > :03:35.Facebook feed looking for adverts So once you've installed
:03:36. > :03:38.the plug-in, it works in the background to extract
:03:39. > :03:41.the whole advert that So it pulls out the headline,
:03:42. > :03:46.the subtitle, any related videos, We also get the reaction -
:03:47. > :03:51.so how many likes, how many comments, how many shares -
:03:52. > :03:54.so we can see which messages Are they particularly
:03:55. > :03:57.clandestine messages, are they slightly subversive,
:03:58. > :03:59.are they even fake news? But how do data companies get
:04:00. > :04:02.the information in the first place? A lot of the quizzes
:04:03. > :04:05.you fill out on Facebook or, you know, you open a survey,
:04:06. > :04:12.it asks your Facebook Sometimes you'll notice that there's
:04:13. > :04:17.a lot of permissions attached and as soon as you click yes,
:04:18. > :04:20.all of your data is mined, and it's then sold on to data
:04:21. > :04:23.brokers who then, eventually, sell it to the political parties
:04:24. > :04:26.for use in their campaigns. Although Facebook says it doesn't
:04:27. > :04:28.sell our information on, data brokers can overlay any details
:04:29. > :04:32.they mine from the site with other datasets that they have on people
:04:33. > :04:36.based on their email addresses. The next step after that of course
:04:37. > :04:40.is to find similar users that are using Facebook and then target
:04:41. > :04:42.adverts, from that advertiser that supplied the email
:04:43. > :04:46.addresses, to those users. There are just some people that
:04:47. > :04:56.you don't find on Twitter. The very nature of the fact that
:04:57. > :04:59.I can't see your adverts, you can't see my adverts,
:05:00. > :05:02.means that this approach had to be It's a first of its kind anywhere
:05:03. > :05:11.in the world on this scale, giving us citizens some transparency
:05:12. > :05:13.into what we're being shown, Do you think that people wouldn't
:05:14. > :05:19.know that certain things are advert A lot of the time people
:05:20. > :05:27.are scrolling through Facebook and the adverts fit into this weird
:05:28. > :05:33.intersection of friend It's quite easy to miss
:05:34. > :05:36.the adverts on Facebook. So far, Who Targets Me
:05:37. > :05:41.has some 6,700 users in 620 constituencies,
:05:42. > :05:43.and it's rising as On the down side, it's only
:05:44. > :05:49.as good as the data it's managed to crowd-source,
:05:50. > :05:51.so it isn't necessarily representative, and it also doesn't
:05:52. > :05:56.work with mobile Facebook, So we're seeing
:05:57. > :05:59.a mixture of two things. We're seeing, firstly, A/B testing,
:06:00. > :06:02.which is where I try out two different messages
:06:03. > :06:07.with the same group. I see which one gets
:06:08. > :06:10.the best reaction and then We're also seeing targeting,
:06:11. > :06:16.which is where I pick a particular demographic of people,
:06:17. > :06:19.and then I send a message So, for example, it might be
:06:20. > :06:22.young people targeted The data from Who Targets Me is also
:06:23. > :06:27.being poured over by analysts One aspect of their research
:06:28. > :06:31.is collecting dark posts, ads which are here one day
:06:32. > :06:36.and gone the next. It gives us the ability to create
:06:37. > :06:40.a respository of those dark posts. So if promises are being made
:06:41. > :06:43.on Facebook, in ads which will disappear the day after you use
:06:44. > :06:47.them, we should be able to go back to those after the election,
:06:48. > :06:50.look at them, evaluate them and maybe discuss them
:06:51. > :06:54.in the cold light of day. And the irony is that,
:06:55. > :06:56.as we demand more transparency from public bodies, the whole basis
:06:57. > :06:59.of political propaganda could be on the brink
:07:00. > :07:07.of a revolutionary change. What's interesting, I think,
:07:08. > :07:09.about the new environment is the potential for using paid
:07:10. > :07:12.advertising and other techniques to create individual propaganda
:07:13. > :07:16.bubbles around individual voters. And that's not about controlling
:07:17. > :07:21.the market as a whole, but it's about using smart targeted
:07:22. > :07:25.which, in a sense, creates such a compelling and overarching
:07:26. > :07:27.information environment for individual people that that
:07:28. > :07:30.in some ways constrains what they do I think that's why some academic
:07:31. > :07:37.commentators and others are beginning to think some of this
:07:38. > :07:41.is a bit spooky. But politicians aren't the only ones
:07:42. > :07:45.with Facebook on their minds. The social network was one of many
:07:46. > :07:49.topics on the very large brain of national treasure and tech geek
:07:50. > :07:55.Stephen Fry. I met up with him after he gave
:07:56. > :07:59.a lecture at the Hay Festival highlighting how he thinks the world
:08:00. > :08:02.is being changed by social The very current conversation
:08:03. > :08:05.is whether Facebook and platforms like them should actually
:08:06. > :08:10.be considered publishers? Should they take responsibility
:08:11. > :08:14.for what ends up on the site? They are aware there
:08:15. > :08:16.is a problem, a serious problem. If 80%, some people
:08:17. > :08:18.have said, is the... You know, in proportion of people
:08:19. > :08:22.who get their news from Facebook rather than from mainstream media,
:08:23. > :08:24.then surely it is incumbent upon someone who is providing 80%
:08:25. > :08:33.of their news sources to make sure that those news sources
:08:34. > :08:35.are not defamatory, blatant lies, propaganda,
:08:36. > :08:36.the wrong kind of, I would posit there that a publisher
:08:37. > :08:42.is responsible for all the people They are employed by that publisher
:08:43. > :08:50.and Facebook is clearly not that. So do we need a third
:08:51. > :08:54.definition, a third thing? I think there is a median sort
:08:55. > :09:00.of definition that it's not beyond the wit of lawyers
:09:01. > :09:08.of the right kind to find that. Your presentation was a warning that
:09:09. > :09:10.people should prepare for the changes that are coming,
:09:11. > :09:13.for example, artificial I said that it was a sort
:09:14. > :09:16.of transformation of You know, it's an obsolescence
:09:17. > :09:20.of certain types of job, but that doesn't mean forced
:09:21. > :09:23.redundancy of millions of workers. I mentioned one of the pleasing
:09:24. > :09:26.things about AI and robotics, and that is what's known
:09:27. > :09:28.as Moravec's paradox, that what we're incredibly bad at,
:09:29. > :09:32.as individuals, machines tend to be Complicated sums, rapid
:09:33. > :09:38.and incredible access of memory from a database of a kind
:09:39. > :09:42.that we could never do, sorting and swapping
:09:43. > :09:44.of information and cataloguing But things we can do
:09:45. > :09:47.without even thinking, like walk across the room or pick up
:09:48. > :09:51.a glass and have a sip of water, But that's fine, because we don't
:09:52. > :09:56.want them to do that for us. Where it gets difficult is in medium
:09:57. > :10:00.sort of service jobs, I think. Well, you could go the etymological
:10:01. > :10:05.route, and you could say Legere is read and inter, interleg,
:10:06. > :10:15.and that's pretty good, actually, Just being able to see connections
:10:16. > :10:23.in things and people are talking about the moment that we arrive
:10:24. > :10:26.at AGI, artificial general intelligence, and that's
:10:27. > :10:29.when the various types of pattern recognition, you know,
:10:30. > :10:31.numbers, data, you know, certain faces and things like that,
:10:32. > :10:33.they all come together so that they can be intelligent
:10:34. > :10:39.across these different things. If you've got an artificial
:10:40. > :10:41.intelligence that's good at that and another one that's good at that
:10:42. > :10:45.and another one that's good at that and you get enough of them,
:10:46. > :10:49.and then you put something on top that goes, "Oh, you want to know
:10:50. > :10:52.about a language problem, Surely, just a collection
:10:53. > :10:58.of specialists of intelligences under one umbrella is
:10:59. > :11:00.a generals intelligence. It doesn't have to be
:11:01. > :11:04.a breakthrough, it just has to be I think that's very likely
:11:05. > :11:14.to be the way it goes. It's reward is similar
:11:15. > :11:16.to our reward system It's tryptophan and serotonin
:11:17. > :11:20.and endorphins of various kinds that reward us and then we have a pain
:11:21. > :11:24.system to deter us, and there's nothing to stop us
:11:25. > :11:27.giving that to a machine. Stephen, thank you so
:11:28. > :11:29.much for your time. Thank you for having
:11:30. > :11:38.us at your place. That is it for this short carte of
:11:39. > :11:42.Leg. The full-length version is an eye player. You can follow us on
:11:43. > :11:45.Facebook for loads of extra content as well.