1 00:00:02,250 --> 00:00:06,930 This has been a rich day already. I will try to keep it quite short. 2 00:00:06,930 --> 00:00:14,520 So there is time for one or two questions, perhaps, and then for one for those who are interested in wine rather than questions. 3 00:00:14,520 --> 00:00:18,720 So I will briefly talk to sort of three aspects of the work that we do at the Reuters 4 00:00:18,720 --> 00:00:24,360 Institute for the Study of Journalism that concerns the role of A.I. in the area of news. 5 00:00:24,360 --> 00:00:35,220 The first is how AI is used on news. The second one is how he is used by news organisations, and the third one is how A.I. features in the news. 6 00:00:35,220 --> 00:00:42,660 And you might think that it's strange that someone who directs an institute that is committed to exploring the future of journalism. 7 00:00:42,660 --> 00:00:50,670 I think a profession that people perhaps associate bit more with sort of ink stained hats or, 8 00:00:50,670 --> 00:00:57,540 you know, stately August old broadcast us is interested in the eye and works in the eye. 9 00:00:57,540 --> 00:01:01,470 But because our focus is on the future, and as Gina pointed out, 10 00:01:01,470 --> 00:01:05,550 sometimes I think we focus a bit on the distant issues without recognising the 11 00:01:05,550 --> 00:01:09,630 way in which analogies are already reshaping the present and the near future. 12 00:01:09,630 --> 00:01:13,920 I will draw attention to a couple of things that are already happening around A.I. 13 00:01:13,920 --> 00:01:18,780 that that is quite central to journalism and the way in which journalism can. 14 00:01:18,780 --> 00:01:25,950 To the best of its imperfect ability in different countries around the world, try to play the different roles that we may associate with it. 15 00:01:25,950 --> 00:01:31,560 Inform us by making the invisible world visible to the citizen of the modern state. 16 00:01:31,560 --> 00:01:38,400 Help to portray the contending forces of our time so we can think about where we stand on the issues of the day and confront us, 17 00:01:38,400 --> 00:01:43,260 if you will, with experience to go beyond our own immediate lived experience. 18 00:01:43,260 --> 00:01:46,380 So where are we with A.I. in this? 19 00:01:46,380 --> 00:01:56,580 Well, the first thing I want to point out is that arguably the biggest impact of A.I. as it exists today in practise in this case, 20 00:01:56,580 --> 00:01:59,580 you know, mostly various forms of machine learning, 21 00:01:59,580 --> 00:02:07,650 but also some forms of deep learning and neural networks is applied to news by other companies that do not 22 00:02:07,650 --> 00:02:13,650 themselves provide news and is quite central to the way in which most of us navigate news and public information. 23 00:02:13,650 --> 00:02:21,720 I'm talking, of course, about the role of platform companies that operate search services like Google Search or social 24 00:02:21,720 --> 00:02:30,180 media platforms like Facebook that by now are the most widely used ways of accessing online news. 25 00:02:30,180 --> 00:02:36,270 So other word is interest to it. Every year, we run a big survey that write a sensitive of digital news report where in 2019 26 00:02:36,270 --> 00:02:40,080 we surveyed online news services in 38 different markets across the world, 27 00:02:40,080 --> 00:02:44,040 and we and most of these people use a wide variety of different ways of getting news online. 28 00:02:44,040 --> 00:02:48,420 But when we asked the follow up question about what is their main way of accessing news online, 29 00:02:48,420 --> 00:02:55,800 only 29 percent identify going directly to the website or app with a news publisher as their main way of getting news online, 30 00:02:55,800 --> 00:03:01,680 whereas more than two thirds identify various forms of site or access via search or social, 31 00:03:01,680 --> 00:03:05,340 or also other forms of site doors like aggregators, for example. 32 00:03:05,340 --> 00:03:14,490 And in 2019, for the first time, we had actually had a majority 53 percent of our respondents identifying various forms of algorithmic selection, 33 00:03:14,490 --> 00:03:19,770 search social media aggregators as their main way of accessing news online. 34 00:03:19,770 --> 00:03:26,160 So this is A.I. that is used on news content to filter it as all of us are voting with 35 00:03:26,160 --> 00:03:32,490 our feet in terms of what are the channels we rely on to discover the news content. 36 00:03:32,490 --> 00:03:38,040 And this, of course, leaves news organisations in a situation where they are in turn increasingly reliant 37 00:03:38,040 --> 00:03:43,500 on the search engines and social media that all of us as end users are embracing. 38 00:03:43,500 --> 00:03:48,360 And in turn, these rely on various forms of AI and machine learning amongst the many different 39 00:03:48,360 --> 00:03:53,400 forms of automation that they use to near instantaneously rank content. 40 00:03:53,400 --> 00:03:58,620 Whether it's in response to a search query or a response to the act of opening a 41 00:03:58,620 --> 00:04:05,130 social media app or in the response to the act of accessing a news aggregator. 42 00:04:05,130 --> 00:04:09,000 This is a AI down to news, so to speak, by other companies. 43 00:04:09,000 --> 00:04:11,880 What about the news industry itself? 44 00:04:11,880 --> 00:04:21,510 So a lot of the discussion has been around the sort of idea of so-called robot journalists and essentially the hope, 45 00:04:21,510 --> 00:04:25,650 perhaps amongst people concerned about the expense side and the fear amongst 46 00:04:25,650 --> 00:04:30,090 those who actually produce the journalism that all of us rely on as citizens, 47 00:04:30,090 --> 00:04:38,340 that they would be replaced by systems akin to the tech generation that Katrina showed us. 48 00:04:38,340 --> 00:04:44,190 Various forms of neural networks like to generate content automatically without human agency. 49 00:04:44,190 --> 00:04:48,120 In fact, as is often the case with digital technology, 50 00:04:48,120 --> 00:04:56,070 a lot of the initial embrace of A.I. in the news industry has been at the back end rather than the front end. 51 00:04:56,070 --> 00:05:02,200 It's been around issues of recommendations. This is a survey of digital news leaders and we just published. 52 00:05:02,200 --> 00:05:08,170 Earlier this month, more than 200 editors and chief CEOs and other leaders and news organisations from around 53 00:05:08,170 --> 00:05:12,670 the world writing the forms of A.I. they consider most important for their organisation. 54 00:05:12,670 --> 00:05:17,770 So various forms of automated recommendations and secondly various forms of commercial use, for example, 55 00:05:17,770 --> 00:05:24,730 to predict the propensity that you might sign up as a subscriber for news organisation and thus target you and market to you in particular ways. 56 00:05:24,730 --> 00:05:31,930 And only less so is the use of this sort of automated tagging introduction metadata and a light to display things more efficiently. 57 00:05:31,930 --> 00:05:38,710 And there's actually very little focus on news gathering and so-called robot journalism in the traditional areas of concern. 58 00:05:38,710 --> 00:05:46,190 From the point of view of many professional journalists and much more at the back end of data of targeting and of recommendations. 59 00:05:46,190 --> 00:05:50,080 And of course, this in fact, is not so different from what the platforms are doing to journalism. 60 00:05:50,080 --> 00:05:55,480 In fact, publishers are in many ways trying to follow the platforms and adopting to the best of their ability. 61 00:05:55,480 --> 00:06:03,160 Some of the same technologies and here, of course, I think you can see resonances to some of the points that Alan drew up in the opening presentation 62 00:06:03,160 --> 00:06:07,090 about the way in which an access effect or performance effect might lead to very different outcomes, 63 00:06:07,090 --> 00:06:09,940 depending on who in fact, can use these technologies. 64 00:06:09,940 --> 00:06:17,410 There are two major news organisations here in Oxford the Oxford Mail slash, Oxford Times and then the BBC Oxford. 65 00:06:17,410 --> 00:06:21,850 And let's just say that the two organisations will have access to very different forms of technology, 66 00:06:21,850 --> 00:06:26,650 as News Quest and the BBC have rather different resources available and different abilities to gather data. 67 00:06:26,650 --> 00:06:33,400 For example, this might play out in quite different ways, depending on how the technology evolves. 68 00:06:33,400 --> 00:06:39,760 The third area is the way in which AI features in news coverage itself. 69 00:06:39,760 --> 00:06:46,300 I think Gina issued a call, if you will, for a public debate, not just around the ethics of A.I. 70 00:06:46,300 --> 00:06:51,430 What do we consider to be good or bad forms of this technology or use of this technology, but also the politics of it? 71 00:06:51,430 --> 00:06:58,060 How do we make collectively binding sessions again, a theme that Alan start from the beginning of the day as well? 72 00:06:58,060 --> 00:07:03,880 I suppose I would say this as someone who both works on and cares deeply about professional journalism, 73 00:07:03,880 --> 00:07:12,580 but because most of us have precious little personal insight into A.I. or the wider implications beyond our 74 00:07:12,580 --> 00:07:19,330 experience as users where we mean it may often be invisible to us that these technologies are relied upon are used. 75 00:07:19,330 --> 00:07:27,310 I think we rely, at least in part on the ability of journalism to inform us about the ways in which these technologies are used. 76 00:07:27,310 --> 00:07:30,490 Who uses them for what and with what life implications. 77 00:07:30,490 --> 00:07:36,760 If we are in fact a form of you as citizens on how we think this technology ought to be governed 78 00:07:36,760 --> 00:07:40,030 and what we think of the ways in which is being put to use by different organisations, 79 00:07:40,030 --> 00:07:47,140 including, of course, importantly, our own governments to do a variety of things that we may not find equally agreeable. 80 00:07:47,140 --> 00:07:53,110 So how is this in fact covered in the news? Well, I think most of you will have if you have an interest in this area, 81 00:07:53,110 --> 00:07:58,900 will come across articles that are illustrated with a still from various Terminator movies. 82 00:07:58,900 --> 00:08:06,550 The sort of the fear of killer robots. And I think it's quite a resonant framing, if you will, of the issue. 83 00:08:06,550 --> 00:08:11,950 And when we set out to try to understand how AI is covered in the UK news media. 84 00:08:11,950 --> 00:08:14,860 Two years ago, at the beginning of a project, 85 00:08:14,860 --> 00:08:24,460 I suppose our hypothesis was that that kind of sort of slightly sensationalist scaremongering would be quite prominent in the volume of coverage. 86 00:08:24,460 --> 00:08:28,570 In fact, we found a rather different texture of the coverage, 87 00:08:28,570 --> 00:08:35,740 which is that when we look at what the sources are and the locations that occupation news coverage of A.I. in the UK, 88 00:08:35,740 --> 00:08:40,660 it is overwhelmingly driven by the industry itself. 89 00:08:40,660 --> 00:08:48,040 So more than half, in fact, 60 percent of all the articles we identified that dealt with A.I. in the content. 90 00:08:48,040 --> 00:08:52,060 Now, as of eight months of coverage in six mainstream news outlets here in the UK, 91 00:08:52,060 --> 00:08:56,290 60 percent were occasioned by various forms of release, press release, 92 00:08:56,290 --> 00:09:03,790 product launches and the like by companies commercial for profit companies that develop various forms of A.I. technology, 93 00:09:03,790 --> 00:09:08,110 very little academic research, very little government, almost no civil society or others. 94 00:09:08,110 --> 00:09:12,640 And in fact, last year we did a follow up studies on this to better understand the role of academics. 95 00:09:12,640 --> 00:09:17,230 It turns out that amongst the academics, more than two thirds in fact have very strong industry affiliations, 96 00:09:17,230 --> 00:09:20,710 not simply in the way that they collaborate with some of these companies, which we do too. 97 00:09:20,710 --> 00:09:24,280 At the institute, as many entities in Oxford and elsewhere do. 98 00:09:24,280 --> 00:09:29,560 But in that they work primarily for industry and have only part time on nominal academic affiliations. 99 00:09:29,560 --> 00:09:38,140 So in fact, what is here listed academic research turns out on closer inspection to be largely be more industry sources driving the coverage. 100 00:09:38,140 --> 00:09:44,110 Now, this is not always a bad thing, but I think we can sort of see the risk for a certain conflict of interest, if you, 101 00:09:44,110 --> 00:09:50,680 if you will, in a coverage that is largely driven by sources from the industry itself, a for profit commercial industry. 102 00:09:50,680 --> 00:09:57,700 And I think if we step back from China early to look at what this sort of what analogies to the situation might be, 103 00:09:57,700 --> 00:10:01,690 I think we can think of sort of two things from the not too distant past the way in. 104 00:10:01,690 --> 00:10:07,630 To which the finance industry has been covered historically, including in the run up to the financial crisis. 105 00:10:07,630 --> 00:10:12,160 In a way that's heavily reliant on industry sources with technical insight and expertise, 106 00:10:12,160 --> 00:10:18,220 which make some good sources but also potential conflict of interest. And, of course, even more adjacent to the end, in some cases, 107 00:10:18,220 --> 00:10:25,330 even dealing with the same very same companies the way in which say, social media has been covered in the tents, 108 00:10:25,330 --> 00:10:30,760 if you will, again, in a way that was heavily heavily reliant on industry sources and often quite, 109 00:10:30,760 --> 00:10:33,940 if you will, quite positive about the potential of new technologies. 110 00:10:33,940 --> 00:10:38,410 And I think we have seen belatedly and not so attuned to the sort of the wider range of 111 00:10:38,410 --> 00:10:44,080 possible outcomes and consequences of the large scale deployment of new technologies. 112 00:10:44,080 --> 00:10:52,960 So where does that leave us? I want to leave with a couple of questions that we can discuss, either in this room or delightful refreshments outside. 113 00:10:52,960 --> 00:10:55,750 So I think there are some questions that this raised, at least for me, 114 00:10:55,750 --> 00:11:01,360 and in the work we do at the Writers Institute around the growing role of A.I. news. 115 00:11:01,360 --> 00:11:08,890 I think the first one is how do we ensure accountability, intelligibility and transparency when now is used on news? 116 00:11:08,890 --> 00:11:12,760 I think fundamentally for us as users, but more importantly, perhaps as us, 117 00:11:12,760 --> 00:11:16,780 as citizens and indeed for many of the participating organisations for the public service 118 00:11:16,780 --> 00:11:21,220 media like the BBC or for profit ones like News Quest that owns the Oxford Mail, 119 00:11:21,220 --> 00:11:26,590 it is extremely hard near impossible to actually understand what the [INAUDIBLE] happens. 120 00:11:26,590 --> 00:11:31,540 When news is filtered through, for example, search or social, 121 00:11:31,540 --> 00:11:36,940 and I think it's a question whether that is sort of a state of affair we can be content with. 122 00:11:36,940 --> 00:11:43,090 Secondly, when we think about how the news industry itself leverage the power of various forms of A.I., 123 00:11:43,090 --> 00:11:48,880 how might they avoid the already well known and documented risks of, for example, 124 00:11:48,880 --> 00:11:55,000 rampant discrimination in the way in which we deal with gender and race again, an issue that Gina Flack for us, 125 00:11:55,000 --> 00:12:02,830 it is well known that these systems, unless very carefully calibrated and constantly monitor, can reinforce and amplify existing inequalities. 126 00:12:02,830 --> 00:12:08,950 Is that really something that the news industry should be complicit with, in particular knowing that these risks are very real? 127 00:12:08,950 --> 00:12:15,220 And secondly, of course, concentration. There are many dynamics at play in our media environment right now that lead to sort of winner takes most 128 00:12:15,220 --> 00:12:20,950 markets where a few companies succeed beyond their wildest dreams and most other companies struggle. 129 00:12:20,950 --> 00:12:28,780 And I think this issue of how A.I. develops and whether it becomes more accessible or more effective, 130 00:12:28,780 --> 00:12:33,920 if you will, is a pretty critical question for what the future of the news industry will look like. 131 00:12:33,920 --> 00:12:40,450 Are these technologies that will only be really powerful and efficient to deploy if you are a very large company with lots of data, 132 00:12:40,450 --> 00:12:42,730 lots of engineers and lots of money? 133 00:12:42,730 --> 00:12:49,870 Or are these also tools that might empower smaller organisations, non-profit organisations, local news organisations and the like? 134 00:12:49,870 --> 00:12:57,550 So issues around discrimination, concentration and of course, finally for the journalism itself? 135 00:12:57,550 --> 00:13:03,550 How can journalism rise to the challenge of covering A.I. in a way that allow for an informed debate, 136 00:13:03,550 --> 00:13:12,700 not just in rooms like this at universities like this, but in public over the ethics and politics and future of A.I. with that? 137 00:13:12,700 --> 00:13:16,288 Thank you very much.