Machines are learning. Are you?

Social media optimisation, meta tags, script writing, news reporting — artificial intelligence is rewriting the rules around skills once…

Social media optimisation, meta tags, script writing, news reporting — artificial intelligence is rewriting the rules around skills once uniquely ours. We met David Ingham, associate partner, media & entertainment at IBM, to discuss why it’s crunch time for human learning.

What does AI mean for media businesses and the way publishers create and distribute content?

Let’s say you’re pulling a story off the wire and it’s The Telegraph or The Times. You don’t want to just use that same copy that came off the wire. You want to add your tone to it, their style guide, whatever their readers are going to be interested in. You could train artificial intelligence to look at what you’ve done in the past, if you have enough data. It could look at the original wire story, look at how you rewrote it and potentially apply that to future pieces. Maybe you’re rewriting certain words, you’re using certain tone or emotion. That could be automated.

It’s not possible now but it’s within the realms of possibility that a machine could watch riots in the Middle East, understand what’s going on and write a story from that. There’s a roadmap to it.

If you don’t have all your data in a single place, or neatly aggregated, you can’t do anything. Treat your data as gold. It’s oil for that machine.

How is AI being used?

The New York Times has a bot on Slack that is constantly looking at the articles that are published to the website or to traditional print. Then they have algorithms for how certain content works on social media. If a travel article or a food article comes in, it will compare them to patterns achieved in past articles. If a food article was written at 9am, it might hold it til 12.15 for Twitter, 12.21 for Facebook, because data says it performed best at lunchtime. It goes into a work queue for the social editor. Today there’s still a human that takes the recommendations the machine is giving them and decides when to push things out, but that could be completely automated. The machine might then learn that it should have pushed the Facebook post out at 12.01, because another newspaper got more likes at that time. It’s a feedback loop. You have to be creating those touch points so that the machine can learn automatically. Life continues to change, people continue to change and unless the algorithm understands, it’ll be worthless as soon as you put it in.

If you zoom out, all media organisations are doing something and it’s in very small pockets. There’s a very experimental, agile approach to it. They’re finding a use case to prove out the technology that applies to their content, to their culture, to their people and they’re doing it. We’re not talking about big dollars or pounds here.

How is feedback given to make algorithms intelligent?

It could be likes, it could be people clicking through. It’s probably data that you’re already capturing. It just needs to be fed back into the algorithm and interpreted in some way to say ‘is this good?’ or ‘what does this mean?’. The algorithm will do a lot for you and will give you a lot of insights, but if you’re not maintaining it and looking at the feedback loops to interpret what the data means, it’s going to get out of control. So there’s still work for humans to do in the process.

If I know how content is going to perform, how well it will do, how many people will read it or watch it, then that’s input to editorial. It will reduce costs and the media is under pressure to reduce costs. But I think it’s more interesting to say that it can decrease your time to market, or just increase the amount of content that you’re producing. When I talk to clients about a distributed media strategy, going to their website, their apps, to all social media and affiliate partners, everybody says “I can’t do everything. I have four people over there doing Snapchat-specific content and two people doing Facebook-specific content. I can’t get to every output.” But if you can automate that for the output points, then you could get to every single output point very quickly. It’s rule-based stuff. Then you can get to more eyeballs.

Where do the limitations lie in machine learning currently? How do humans have the upperhand?

Machine learning is completely backwards looking. You train it based on things that have happened in the past. I like to think that humans don’t work that way completely. We can anticipate things, or we can just pick something out of leftfield and have inspiration. Machines, at least right now, are very rigid. ‘Have I learned this in the past? This worked 87% of the time so I’m going to do that the next time.’ Whereas sometimes the thing that worked 1% of the time is the thing that will work going forward. For me that’s the one saving grace. You can’t replicate that. At least not now.

How will the role of apps evolve with AI?

I think that goes down to human behaviour, personalisation and recommendation. AI will look at how I have used an app or website or social media in the past and anticipate how I will use it in the future. If it’s completely seamless then it’s perfect, if I think it’s my experience.

Why should a media business consider AI?

You can’t take a back seat because you lose so much time just understanding the technology. Find little projects you can do for experience, then the next time they’re much easier. If you sit on the sidelines and just wait for it all to shake out, you lose that experience and then when you eventually dive into it, either you have to pay for consultants or hire people who’ve done it before. In media now you don’t have time. You need to launch products and play catch up immediately. Do you have six months to get up to speed with something? No. You have to seize the opportunity at that point, so I do think there’s some urgency to at least test AI out and find the right use case for it.

If you were going to advise people to do one thing in 2017, would it be this?

Yes. In 2015 I would have advised people that 2016 was the year for virtual reality. You need to do something in the VR space, whether it’s news organisations or entertainment. That continues to chug along. In 2017 I think it will be AI, machine learning.

How would you advise people to start?

Find the use case that works for you. Get on with it and start small. Be an early adopter, not a follower, to get the experience. Actively search for and create new data sets. If you don’t have data — and that could be scraping social data, comments, customer data, their history of what they’ve consumed on your site, or what they’ve bought from you — then you can’t train an algorithm to do anything with it. If you don’t have all your data in a single place, or neatly aggregated, you can’t do anything. Treat your data as gold. It’s oil for that machine.

I’ve been making the same argument in technology for 15 years. It’s to free up time to do more value-added stuff. I think this time it’s scarier. This time you’re training a brain, who could potentially replace everything that you do.