CogX, the festival of all things AI, returned to London last week, bringing with it speakers and companies from around the world to discuss and demonstrate how the potential of this technology is becoming reality.
AI is regularly touted as a major disruptive force in practically every industry. In many places that’s still just hype, but in media it’s becoming reality. From bots writing sports stories to algorithms choosing what audiences see, this technology is changing the way content is created and distributed.
With that in mind, CogX featured a panel this year with the intriguing title ‘AI: the threat or cure for media?’. Helmed by tech media stalwart Mike Butcher, Editor-at-large for TechCrunch, the panel was made up of Grace Boswood (COO, Design and Engineering at the BBC), Dan Gilbert (Director of Data for News UK), and Bibblio’s Mads Holmen. We were there to bring you some of the highlights of the discussion:
Data transformation in the media
Mike kicked things off by asking Dan about the rapid changes in the industry from his viewpoint at News UK. Was it fair to suggest that most media organisations have only recently woken up to the fact that they are sitting on vast troves of valuable data, and are they now trying to become tech companies?
Dan agreed that things have really begun to change in the last few years. When he got into journalism 5-6 years ago, things were very analytics focussed and the emphasis was on trying to work out who was buying what. By comparison, in the last 2-3 years it’s been much more about helping journalists use the data they have about content, which had previously been impossible.
Practical applications of AI in the newsroom
Mike was curious to what extent the use of algorithms and AI had actually supercharged content consumption. Grace’s immediate reply was that it wasn’t just a case of supercharging things: as recently as 12 months ago on iPlayer, when you watched an episode of Top Gear you were asked ‘do you want to watch this episode again’. Whilst that might work for your three year old child, adults aren’t so keen on watching the same thing over and over.
“12 months ago, I kid you not, on iPlayer, when you watched an episode of Top Gear we asked you ‘Do you want to watch this episode again?’”
Beyond the simple fixes like this, there’s also a need to make the content journey even more sophisticated, because if the BBC doesn’t then other people will, and audiences will choose their platform instead. Grace mentioned Netflix as a company that stands out as doing that very effectively.
She also pointed out that the BBC’s unusual status, with a public service mandate, meant that they had distinct aims when it came to working on content recommendation. On terrestrial television, the news would always follow Eastenders because it helped encourage the large audience that tuned in for that to take an interest in current affairs. It was the BBC’s aim to take this kind of thinking into algorithm development as well – how can recommender systems help encourage civic engagement, as well as clicks?
Dan said that at News UK, AI is now helping at all stages of the content life cycle, from editors and journalists planning what to cover to then trying to get more people to read it via recommendations. A fundamental change is that technology has appeared in the last couple of years which allows marketing approaches like audience segmentation to be applied to the newsroom. This has now advanced to the stage where AI is on the cusp of helping with e.g. headline writing, but there’s still definitely a spirit of experimentation.
“We’re trying lots of things, and it’s all pretty nuanced”
Skepticism in editorial
Mike asked about the problem of cultural pushback from editorial teams when this data-led approach began to emerge. Dan reiterated that things really have changed in the last few years: there’s been a dramatic shift and there’s now a lot of enthusiasm. This is partly because teams are now much more collaborative.
“it’s not data scientists, in a bunker, trying to solve problems, it’s data scientists trying to work with the editors and journalists”
Grace agreed that there has been a rapid cultural shift from data almost being viewed as an enemy to now very firmly being viewed as a friend. Content creators don’t like being directed, and there had been a definite mood of “Stop weaponizing data! We know our audiences!” That’s changed. There’s an acceptance that data can enable a more audience-centric perspective and has a role in connecting the various part of the BBC.
Limitations of AI in the media
Mike was curious where the panel saw things heading. Mads said that while it was clear that personalization had become the Holy Grail for many content businesses, there were increasing signs of pushback. Over-personalization, e.g. in recommendations, led to narrow horizons. As one of his friends recently confessed about logging into personalized social media, he’s often disappointed to see “my own, unimaginative self staring back at me”.
The fact is, even when content businesses have enough data to personalize services, using it to send someone on a binge, a la Youtube, will ultimately be unproductive if the user begins to have negative feelings about their usage habits.
“(Some algorithms) optimize to the data we have, but maybe not the right data”
Mads pointed out that whilst trying to ‘hack’ attention looks successful in the short-term for an ad business, at some point visitors are likely to want to grow as people, and they’re going to want more from you. Even Gen. Z’ers, who’ve never seen a non-algorithmic world, are growing increasingly aware of the problem of going down content rabbit holes. They will get sick of shallow experiences. With the algorithms we have at the moment , it’s hard to detect when that moment is getting close.
Mads’ belief is that this is the next frontier: how can we begin to really capture what people want and how does media fit into their lives to help them achieve that.
“(Current algorithms) don’t capture the emotional aspect of being a human and consuming content.”
Grace agreed, and added that algorithms don’t just personalize, they also optimize for the business model. Many of the big tech players businesses are built on driving time spent, and the way to drive time spent is to send a visitor into a filter bubble so they carry on consuming content. That’s the way many businesses have built their fortunes, but it’s a pretty questionable practice.
She was adamant that there is an opportunity for other businesses if they can create distinctive experiences which are, frankly, founded on different business principles. The BBC really sees a role for itself in that world, not being driven by time spent but by value added for users, and how people feel about the experience you’re giving them.
Buzzword Bingo: Blockchain!
With the current hype around blockchain and cryptocurrencies subsiding but very much still around, Mike asked whether those technologies would have role in deciding which content people do and don’t end up consuming. Will blockchain and crypto mean that algorithms incentivise readers to look at what they’re paid to read?
“Crypto, blockchain, data science, AI all get bundled into one thing…”
Dan and Grace were fairly skeptical about the possibility that blockchain and crypto would be serious influences on audiences media consumption, and Dan pointed out there’s a lot of confusion around how they belong together, or don’t…
Mads was more positive about the idea, but cautioned that for the moment there was little sign of adoption beyond a core of die-hard tech fans. Citing the crypto token BAT, that had been built to deliver micropayments to people in exchange for their attention, he pointed out that there’s an experience problem.
“Actually using these products isn’t always a very delightful experience. Until we can lower that barrier it’s not going to scale beyond a certain group.”
The evening ended with two contrasting takes on where AI and media is headed:
Dan emphasised that with the field developing so quickly, it can be naïve to make grand predictions about how things will change in the newsroom.
“As AI and machine learning are evolving and evolving, the way you introduce it never really stands still.”
Grace cautioned that getting journalists to really understand and engage with the new technology is still a work in progress, given that using data has only recently become par for the course. AI isn’t about to replace content creators just yet.
“There is the magic in editorial voices, and it’s very, very unlikely in the near term that those are going to get replaced.”