Newsroom technology and the future of journalism

Newsroom technology and the future of journalism



for joining me today for a discussion of newsroom technology and the future of journalism just to kick us off a little bit about me and what we're doing at Bloomberg I'm the global head of breaking news and markets coverage at Bloomberg where I've worked for 20 years in London Hong Kong and now in New York which is our headquarters and more recently I'm also responsible for the teams within the newsroom that build editorial tools and the really kind of build the technology that is supporting our journalism and that includes automation and it includes the AI strategy that we have in the newsroom so in that sense I'm a journalist with oversights of technology which is obviously a you might think a bit of a dangerous thing I'm definitely not an engineer so you'll have to forgive me a little bit for for not overly granular or technical examination of newsroom innovation but the idea really is to think about and discuss some of the trends that we're seeing some of the benefits of technology in journalism you know some discussion of the risks and and what the future holds and maybe some predictions I guess another disclaimer is I'm not I'm not here to especially brag about what Bloomberg is doing in this space I do feel that we're doing something pretty good and significant and I feel as a company we've got a special interest somewhat unique interest in the intersection of Technology and journalism and I'll get into the reasons for that but you know there's plenty of other organizations obviously doing fantastic work in this space Reuters the AP kourt's just some of them so just to step back and state the obvious a technology I think we all know as brought plenty of change and fear to journalism through the ages 1814 The Times of London started using a steam-powered press that effectively raised the amount of papers you could print by tenfold and cheap newspapers appeared everywhere as a result if we go back couple of decades or even a decade you know the internet was obviously the technology that was scaring newspapers scaring journalists the most and with some justification time has probably told that it wasn't necessarily the internet per se that was the problem it was more giving away the content for free and to some extent we've seen that trend reversed and newspapers have adapted now there's a new spectre haunting the industry AI or artificial intelligence you've all seen the headlines journalists will be replaced by robots the quality of writing goes down robots will only tell us the news we want to hear fake news will proliferate all of these things now we know broadly technology saves time and has changed the course of journalism and in some cases has and will affect jobs but are all these horror stories fair personally I think not and to show that I do want to look a little bit of why does technology matter to us and why does it matter to our audience a bloomberg we have 2,700 journalists and analysts and an array of platforms from the web to TV radio and of course our main subscription product which is something called the terminal which is used by financial professionals worldwide but we're we're also more than a news organization we're a technology company a data company and many other things so we're fortunate as journalists because we have support from departments of engineers departments of data specialists data scientists and product people as we try and do more with technology and news and it's really because of this DNA that we've actually been doing automation of content in one way or another since before I joined the company two decades ago but these days we reckon that about 1/3 of everything we do 1/3 of all our stories have some elements of automation there's a big reason why people pay for a terminal markets obviously fluctuate in the blink of an eye and news has to move just as quickly and at Bloomberg obviously we need to be first on headlines and news that are gonna have an effect on asset classes so think about bonds or currencies or stock markets commodities oil all of these things can have can can be immediately affected by news and our readers depend on us being accurate they depend on us being comprehensive but also first and we need to be first by split-second and if we're not our audience loses so what is AI artificial intelligence doing to help those efforts well there's a few things I would highlight the first thing is automation generally is very good at handling events which are repetitive so repeat events especially those things that involve data of some kind so what what a repeat events well you could say that sport results are sports reporting is something which is a repeat event for us critically we focused on corporate earnings when I first got to Bloomberg one of the things that we used to do were to handle a kind of avalanche of press releases from companies which came every day was we would have a desk of people and myself included would scan the press releases as they all came out in real time looking for the salient information looking for the numbers or the news that mattered most and then kind of pulling out the facts and the comments and turning them into rapid headlines and short stories that summarized the information and then of course we would build a narrative story from there these days with the help of automation we have a different system which we built in-house and we called it cyborg and what happens now is before the earnings from companies or before corporate disclosures come we train the Machine we train the system on what to look for so essentially you have journalists making editorial decisions about what might be most relevant from these releases and then training this system to extract the relevant data points send out immediate headlines if there's stuff that's market moving and a first version of the story to the wire and now we can do that in multiple languages simultaneously and we now cover thousands of corporate releases in multiple languages using this automated cyborg system and as a result our coverage is a lot broader than it ever was before the other thing you can do once you have a system like that in place is you can train it to keep looking for more data so you can expand it beyond just the top-line numbers all the key earnings numbers that we're originally doing you can look you can have it look for a secondary data in the release and you can use this system we call cyborg for more than just earnings you can use it for economic data because again that's a structured data that comes in you can use it for any really corporate announcement companies day-in day-out are announcing things to the market that go way beyond earnings they announce M&A they announce dividends buybacks bond sales really anything that comes in in a in a predictable format you can then train this machine to extract so briefly on data extraction you have to obviously have it happen in milliseconds and you have to have it happen with precision the documents may come in different formats you may get it in PDF you might get it in HTML you might have it on the web it might be text it might be tables it could be press release typical press release or an SEC filing or a court document or a financial disclosure and obviously the text comes with natural variations and what you have to do is build statistical models that then learn from the variety that we we encounter every day so that's cyborg but what about other repetitive tasks that we're trying to tackle well well obviously in journalism we once you start on this track you realize there's there's a lot of repetitive tasks that we're trying to do every day among them calling into conference calls transcribing interviews a Bloomberg we write a lot of daily markets reports on the footsie or sterling other asset classes translation which I mentioned before or you know when it comes to analysis of huge troves of data if you think about something like the Panama papers the New York Times last week actually had an article where it detailed how its reporters went through about 900 pages of legal documentation related to the Michael Cohen case and they did it in 10 minutes using this technology so technology is obviously helping save time and speed up coverage if you think about transcription of tape interviews obviously there's a lot of time taken every day in doing that but a machine can do that pretty much immediately in our world again if we think back to the earnings example we deal with a lot of briefing calls with companies so you have reporters darling into briefing calls to hear what the chief executive has to say when he or she is commenting on their earnings but obviously if you if you're doing it the old-fashioned way you know a one reporter can only dial into one call at a time you know if you have the right system you have the right machine the Machine can listen to unlimited numbers of conference calls simultaneously so again if you have a journalist train the model on what to look for or what keywords to search for then the Machine can either alert the journalist that something has been said that's relevant or we could have the Machine produce something small a short story on what was said obviously again the value is immense when we you know we're in the business of trying to bring transparency to financial markets and if we can use these systems to suddenly listen to thousands of conference calls that maybe we didn't have the resources to do previously then you're adding a lot of value immediately the other thing I mentioned is translation the advances in translation are very exciting to us and I think the entire industry and if you have like we do reporters in Tokyo or Istanbul and they're covering the economy or they're covering politics of they're covering companies and they're doing in their local language in Japanese or Turkish then you can have much more of an instant translation into English you're bringing all other languages Spanish Italian you can that you can then bring a lot more of this coverage to an international audience what our cyborg system obviously allows us to do as well is you can prepare templates in local languages so that when for instance if you had a US company disclosure an SMS P 500 company that was disclosing in English but you've already instructed the model to extract the information and produce a first story there's no reason why you can't do that in any language you want it's just a question of building the template and training the model and obviously for us as well the data angle is is really critical Automation is good at detecting discrepancies or deviations from a standard set of data and because we are ultimately a data company and we have a lot of data at our fingertips we built alerts that can tell the reporter when something is happening that is unique or different so the machine could say for instance okay the GD you know have you noticed that the GDP GDP in Germany is rising five times faster than the UK since brexit happened do you want to write a story on that and so we're finding these kind of triggers where the Machine identifies correlations or interesting pieces of data and alerts journalists very useful indeed if you imagine again you know if the Machine notices the directors of a company like BMW are selling shares in that in that company then again we can have a trigger and alert to the news room or if there have been a record number of hot days in a city like Rome which might obviously have an effect like on energy prices or movie ticket sales you again you can have the Machine kind of pinpoint these correlations and alert the journalists then there's what I would call the noise factor which I think we can all agree is not only a problem for journalists but a problem for readers as well you know we live and breathe this hyperaccelerated news cycle we have news breaking in all different shapes and sizes we are from all different sources social media press releases the web and so on so how do you kind of find what's relevant from this gigantic fire hose of information well one thing that we do and we have built is a system that the tackles social media in this way and it distills the millions of tweets published every day into something that is much more reasonable to monitor and search if you're looking for relevant news and we use machine learning to filter out spam we use kind of name recognition algorithms and then we classify relevant topics and this will cut through the noise and help us kind of capture news that we think is relevant to the financial community and you can apply those kind of models to anything really from terrorist attacks industrial strikes refinery outages all these things that tend to have an effect on financial markets we can then filter out the noise faster and you can find the news more efficiently and find the stuff that matters and the other thing I would touch on briefly is technology helps us reformat content once upon a time you know all news stories obviously look the same or similar or if you wanted to read for instance about what the New York Times was saying about Margaret Thatcher back in the day you would read the whole story you would read the whole piece beginning middle and end now technology is helping us and others put news into a variety of different formats and I think certainly at Bloomberg and you can see it everywhere this is encouraging a lot of experimentation with formats obviously we have blogs we have more video journalism charts are a very popular thing with our audience for obvious reasons but all these new forms of innovative storytelling are being helped by technology now with finding of course with technology automation there are risks and there are limitations so I was going to spend a little bit of time talking about those the first one the first one I've mentioned is this idea of nuance and as I said before when we were talking about structured data and the things that the machine relies upon the machine generally is not very good at nuance and you know we obviously deal with nuance day-in day-out when we're evaluating news I mean if a press release that we get or a breaking story is vague or if we have people in the news saying one thing and meaning another which obviously it's something that we come across all the time the machine is not very good at reading between these lines or certainly not yet very good at reading between these lines extraction and AI has evolved a lot and it's come to a place where we can take complicated press releases difficult to understand press releases it can decide whether the theme is positive or negative or neutral and the challenge for this type of stuff is the it's always been the fact that the most valuable information sometimes is qualitative rather than numerical and then so obviously you need a human to need to read and interpret it increasingly you can train a machine to make some of those judgments especially around sentiment but that is very much enough in our minds this is this is very much a work in progress it's far from perfect the other point I'd make is AI really is not a good substitute for new judgment and you need standards and journalists around the automation process in order to really match the standards that you have in place for manual stories one of the things that we really insist upon and I think our audience would insist upon it too is that this full transparency around where there's something has been produced fully by full automation or part automation or a human journalist or whether or not it's been vetted by a journalist and so there are all these different types of stories with varying degrees of human and technical input and I think what's it incumbent on us to be transparent about how this stuff is produced another thing I would talk about again we were just we were just talking a little bit about noise and spam you know in this kind of technology arms race you know it's sometimes when you it's easy to get carried away with ideas around automation and kind of produce content just because you can and so if you kind of develop the technology the the enables production of content one of the things you have to have in place is a really good system for evaluating the value of that content obviously what you don't want to do is just add to the noise with a lot of automated stories that you're doing really for the sake of it or doing to kind of show off your technology again you have to come back to the idea of news judgment why are we writing this story and is it valuable for the reader does it serve a purpose or are we just spamming the reader so when we start to build out as a story type we have to question everything what will wake up the computer to write either write a story or alert a reporter what's the trigger how accurate is the data and what do we have to do to make sure that the data is good enough at this point we have hundreds of automated stories that run every day we have a high record for accuracy but how to avoid errors is obviously critical in our experience in this space we know there are a couple of areas where we need to be really vigilant first of all data we need to ensure and again this is stating the obvious but but we have to keep coming back to it you have to make sure the data is clean and reliable you have to make sure that what you're dealing with as a source is completely trustworthy and it's also critical when you're dealing with extraction rules that they're written in a way that they don't fire erroneously if the source changes the way that it formats the information when you're dealing with ml models they have to be trained and tested time and time again templates have to be written and set up by editors with experience people also ask about bias in AI and when there is little transparency into the classification of models and how algos are built we have to be very proactive about maintaining tools you can mitigate this bias by being very vigilant about the quality of the input data and and have to make sure that the input data is not biased and this this can creep in when the source data changes over time and you need to make sure as I said that the ml models are regularly maintained and we need to monitor the accuracy of our AI outputs constantly throughout this process you have to have the human gut check is what we're seeing logical does it make sense is the end result surprising is the machine generating something inaccurate and we see times and the value of this human judgments AI models will only be as good as the humans that structure and train them and it normally reflects the knowledge and the understanding of humans about certain topics so it's constantly a question of kind of retraining the AIS and testing for biases I'm gonna just do a couple of minutes on fate news because I think think news is something that that there's a hot topic and and obviously when it comes to technology it's interesting to think about how technology is either encouraging or helping defend from fate news and the first thing I would say about fake news is it didn't start with the internet you know humans have always tried to manipulate humans you can go back to the Trojan horse the Garden of Eden Octavian poisoned the minds of the people of Rome about Antony discovering a fake will where he left all his money to Egypt Shakespeare made a villain out of Richard the third I even learned this week when I was in London on April fools day I think it was Jacob Riis MOG said that there was a proclamation of the Privy Council in England against the spread of false information that was made in 1688 so fake news has been with us for a while it's obviously evolved thanks to AI and these days rather than kind of broadcasting one big lie you can narrow cast a lot of small ones and that obviously helps spread doubt so you have bad news machines that can come up with ways to manipulate good news machines and you have different types of fake news you have these deep fakes where photos and videos are altered and these things are increasingly a reality that we're seeing on social media technology obviously can help us spot fakes there's a variety of ways in which technology can do that but again I would suggest and certainly from what we're seeing there's no substitute for human judgement in this process hoax press releases and fake news again is something at our company that we've been dealing with for a long long time and technology can certainly put you in a better position to spot this stuff but again there's no substitute for human judgement so where does that leave us I would firstly suggest a hybrid future with our cyborg model it's worth stressing that we need to tell journalists to tell the Machine what to look for and set up templates we need humans to double-check this process to look for the surprises that the Machine doesn't find maybe in a corporate press release the chief executive resigns and it's buried on page 10 you know you you you typically need to be able to find that without with without the help of the Machine typically it would be up to the journalist to be able to spot that surprise and you can't afford to be wrong and so ultimately we're finding that the machines are technology these models are working hand in hand with the journalists there was a good article last week and the New York Times by Catherine Meyer and the headline was without humans AI can wreak havoc which I think sums it up as good as anything a Bloomberg we're lucky because we have a lot of market knowledge in the newsroom and we can leverage that to identify what's critical what needs to be designed in the machine learning model and what needs to be extracted and so you're kind of leveraging this expertise to gain the smartest ideas and use that in the technology so I think we see this partnership between human machine as some as one that is growing the second thing I would I would point to is this evolving newsroom idea I think I think as a result of some of these trends we're really seeing different types of newsroom jobs emerge again if we go back to the horror stories slide people were and still are very worried about the impact on jobs I think what we're seeing is teams and individuals and journalists enthusiastically now working on some of these AI models and bringing new skills to the newsroom involving themselves in the content and the strategy and how the technology intersects with the journalism and so I think you're seeing a kind of new form of newsroom role that lies at this intersection it and and we're certainly seeing at Bloomberg and I think if you look around you see it at other places as well newsrooms seem to be hiring with technology in mind and the JIT Wall Street Journal just a couple of weeks ago posted a slew of jobs that were related to innovation within the newsroom you know graduates that we're encountering that are coming out of journalism school increasingly speak the language of technology and are interested in being involved in it so and then the third thing that I would point to just to wrap up we see the obviously the impact of AI and automation in terms of the quality the speed we do it to remain competitive in some ways we do it out of a necessity because of our business the other thing that we feel is is a consequence is that the quality of some of the non technology journalism that we're producing is is increasing as a result and and so I would suggest that as the technology improves the traditional journalism also improves because if we increasingly have machines strategically have these models do more of the repetitive work that is involved in journalism then obviously it immediately raises the bar for your beat reporters who are covering companies or who recovery the economy or her recovering markets and and immediately they have more time hopefully an opportunity to do the things that they want to do break news interview newsmakers write analysis we think that journalists don't want to be robots they want to be out chasing scoops and so we feel that by doing some of this stuff it's enabling that to happen so and when I look at we've looked at some of the content that we're producing some of the scoops that we are getting some of the stories that we're covering now I feel like that time saving from automation is really helping us raise the bar in other areas say more of a more of our journalists have time to do the things that they want to do and obviously you know we really hope that trend continues thank you [Applause] yeah questions I don't know how much time we have but would you like to okay some really interesting things you were talking about about one thing that left out to me was talking about transcription service and any journalist who's had to transcribe interviews will know how time-consuming that is right there are some I've been looking at some solutions but they cost and I just wondered any chance of Bloomberg making that service available to all journalists wouldn't it be wonderful it would and actually I think it's something that you might expect you know we're already doing this and we're doing it it's scale but actually we I feel like we and others are really just getting going in this area as well it's something that we are ambitious about and it's something that we're kind of evaluating at the moment but we we do do a lot of Technology in-house and we do obviously do a lot of the projects that I've spoken about in-house that when it comes to the transcription services actually it's something that we I think like others look at vendors and see what's out there before making decision at this point we're not locked into any one technology or working with any one kind of specific vendor but it's not something that we've built the technology ourselves and are doing a scale it's something that I think we and everyone else is still figuring out but I agree if it was free for everyone it would be great go ahead Magnusson saying from the Finnish Broadcasting Company when we when you talked about the factors what the AI searches for you talked about minimizing errors being factual and eliminating fake news but what about the ethics of a story say for an example that the program notices that people are more and more withdrawing money from a bank and that's a story in itself it's true it's actual and spectral but the ethics bit if you report that people are withdrawing money you could create a maybe a bigger crisis if you understand how do you take that into factor the ethics of the situation yeah I think I think you absolutely do have to consider it I think I would take two things that are important about this one is the transparency side I feel the and we worked very close we have a standards and ethics team within the newsroom and we work very closely with them to identify and decide upon the right level of transparency around these stories I think our audience has always demanded a fairly high level of transparency when it comes to how we saw stories and how we produce stories so the first thing that we really wanted to do was make sure that if a story has been produced by machine automated semi automated prepared by a human journalist and then finished off by the machine there are all these kind of variations we try and be as clear as possible in each of those categories so that you know our readers can can make an assessment our readers can can be aware of how the stuff is produced the the second thing I would say is that in every you know some of the stuff that I've talked about he's obviously very data-driven it's obviously that they tend to be very two stories they tend to be very reactive stories so far they're using things like press releases what we don't do typically because you know it's obviously inviting a lot of risk is try and employ technology if we're dealing with very sensitive stories or if you're dealing with stories that are legally sensitive ethically sensitive complex then those are the types of things that obviously they don't lend themselves very readily to this type of process or this type of technology so I think you have to kind of part of the part of the challenge with this always and we've kind of worked through it like everyone else is figuring out what are this what is this stuff that the machine is really good at and what is the type of content that lends itself to this type of process and there is plenty in that category but there's also plenty in the other category where the technology is not there that there are undue risks that there are ethical considerations that we need to be more wary of thank you very much that was super interesting a couple questions did you kind of consciously decided to have an AI strategy for the newsroom or this has kind of grown out of needs to help the newsroom as such and question number two you've built it in-house how long did it take and did you kind of need to prioritize this over something else that didn't get done within the newsroom I think as Bloomberg being what it is and as I said at the beginning we we are kind of living at the intersection of a lot of these trends technology data news I think that we you know always we're trying to find ways to use technology to assist with journalism we're very lucky really as journalists there because we have the support of these engineering teams and and we we in some ways are a technology company so I think we've always kind of had an eye on how do you kind of have the best tools how do you have machine assisted workflow and all the rest of it so I think that we always kind of have had made it a priority using AI and I think I think obviously what's happened is you know you can do it and do that in very basic ways and as I said we were we were actually doing some level of automated markets reports 20 20 years ago you can do it in very basic ways and then as the technology kind of evolves you can do more and more sophisticated stuff so I think we've kind of grown from there the other thing I would say in terms of motivation I mean obviously it's it's kind of great do it it's exciting it kind of helps us helps journalists have more time it does all it does all these things the other thing is permit from a competitive point of view we kind of have to do it because of who we are because to go back to where I was at the beginning you we can't afford to be late or wrong and so with that in mind when it comes to anything that's market sensitive we have to constantly be thinking ahead of our competitors as to how we can be first accurate and all of these things and so if we're going to be you know if we can our subscribers demand that and to do that you kind of need the technology so some of this some of these projects absolutely a kind of born of necessity as much as anything else a lot of your automation at the moment is related to text how do you see the future of audio and video packages being automated and produced by machines for example your tick-tock stories and Twitter do you see that being produced by machines in maybe near future I think I think when it comes to tick-tock and they kind of obviously we're doing a lot of video journalism and when it comes to our tick-tock product which is obviously something which has grown rapidly over social media I think the benefits so far in terms of technology has really been about first of all being able to sift through you know the kind of the reams of content that's on social media and kind of figuring out the stuff that is important to us and the other benefit obviously which I touched on has been how do we use technology to spot stuff that is fake or misinformation and how can that kind of support those efforts and I think those have been the two kind of really meaningful things that that of the technologies help support tick tock and video journalism I think obviously you know the the product when it comes to kind of the production of video at this point you know obviously technology involved but really with the with the type of video that we're producing it's very much the case it's going to be human endeavor rather than something that's automated you mentioned that journalists write templates and then templates then produce the content have you tried to do all also automation in in text generation like machine learning based natural language generation or that kind of things Thanks we we do use natural language language processing in some projects I think when it comes to and you can use NLP in a number of ways something on our on our actual interface as well there that we've used NLP as a way for to enable our audience to eat more easily find information and find the stories they're looking for so it kind of assists discoverability when it comes to production we still feel that while the initial you can certainly have the machine produce initial stories it's really really important in to our success to have journalists who understand the companies to think about and this changes all the time the the kind of what our priorities for a specific company or a specific news event and I and and we feel pretty strongly that you want that that in in terms of preparation you want journalists reporters editors really training these models in both what aspects to look for what are the key salient information that should the Machine be searching for how should it be formatted or displayed and I thought I in our experience so far that level of preparation human preparation really means better results by the machine at the at the end of the process Hey look I was wondering to what extent AI has in fact then replaced um you know so various journalists so to what extent you've then or maybe they've left as well or to what extent you've provided sort of bridging trainings and what sort of trainings I mean we as I said we feel we've certainly we keep growing and so for us the it's never really been about replacing journalists with technology that was never a motivation that was never a reason for doing it and and we've continued to kind of grow the newsroom and we continue to recruit now I think that what the what it's what it's really as I said allowed us to do is think about those time savings that you gain from it think about some of the efficiencies allow you know enable reporters to kind of spend more time on really kind of value-added journalism so I think it's it's enabled people in our newsroom to kind of elevate what they're doing and spend less time on some of this repetitive work and I think obviously as I said at the beginning I think everyone you know over the last few years if you go back a few years I think there was a lot of anxiety around it but I think my feeling is that people have recognized more than actually you know the the AI is helping and it's it is taking on some of these more onerous asks and and allowing them as I say to kind of get out and kind of break news and write more meaningful stories I think when it's interesting but we do provide training to people as we've kind of grown the groups of people who were involved in this stuff in the newsroom we have come up with different ways of training people who want to be involved we figured out ways where you know if there are reporters who editors who want to spend some you know some time kind of seconded to these groups so that they can learn more we've done things like that and we've done specialist kind of training classes for people who are interested and I think you know unsurprisingly the interest level is kind of definitely increased a new kind of first of all you'll met with perhaps a little bit of skepticism and then over time I think people have become far more interested in it but again the the other interesting thing that we've noticed is that when you're talking to graduates and and you know we have a very successful internship program and when we're talking when we're looking at kind of some of the amazing graduates that are coming into Bloomberg via this internship program you know you you know one in that group needs an explanation of why this is interesting or why this is important or or do they want to be involved it's kind of we feel that the kind of the next generation of journalists who are arriving at Bloomberg are extremely involved want to be involved want to have the skills want to be able to combine the skills with their traditional journalism and and very much kind of come in with that mentality which obviously is great okay I think we've run out of time but thank you very much [Applause]

Author:

Leave a Reply

Your email address will not be published. Required fields are marked *