The War for Our Attention
The ASU GSV Summit is an annual conference for the learning and talent innovation community. The 2017 edition was held in Salt Lake City, Utah. Mads Holmen had the opportunity to speak on innovation in education and share what Bibblio is working on.
Here’s a transcript of his talk, The War for Attention, where he considers the attention economy, the role of algorithms and how to create trustworthy user experiences:
The battle is raging
“Hey, my name’s Mads Holmen from Bibblio. Today I’ll be talking about Bibblio, but also about you. About us, the users of the web.
“At present, there’s a war for attention. With an increasing amount of information, inevitably there’s a decrease in the amount of attention. Herbert Simon recognized this 46 years ago.
“We now spend more than 10 billion hours a week on platforms like Netflix, YouTube and Facebook. Not all of this activity is bad for us, of course, but some of it is.
“Actually, quite a lot of it is.
“As these platforms are funded by advertising, their incentive is to get us to spend as much time as possible using them, even if it’s not productive. One of their tactics is to serve you bite-sized distractions that I call ‘sugary content’. It’s stuff that’s very clickable, easy to consume and leaves you still feeling hungry.
What’s not on the menu
“These days, algorithms decide what we see and what we don’t. That raises a few very important questions. As Tristan Harris observed, the important thing isn’t what’s on the menu — it’s what’s not. When you look at Facebook or Instagram, it’s important that you consider the world you’re missing out on, as a result of their algorithms.
“More than 80% of time spent on these platforms comes from algorithmic recommendations. These days, algorithms are our editors. In a way, the developers who built the algorithms decide the world that we should see. Netflix performed an academic study last year in which they estimated their recommendations are worth more than a billion dollars a year, as they helped reduce churn. If you’re a subscription business, this is a key thing to consider.
A different approach
“Bibblio is a recommendation engine. We figure out what content people should see, but approach it in a different way. For example, I’ll show you one of our partners, The Day. I’m a big fan of The Day. They do current affairs for schools. In the UK they reach about a million students and teachers. They’ve recently launched in the US, together with Follett. It’s called The Day USA.
“Bibblio is the system behind the scenes that decides which related articles teachers and students should see. To give you another quick example, this platform’s called Coursedot. They introduce IT training courses to professionals. They raised €400K from Eleven Ventures and Unicredit, with customers across 50 countries, including HPE and Knowlogy. You’re looking at a Microsoft Azure course here, with other related courses to discover just below it.
“So we sit on the recommendation side. Now, discovery on the web generally starts with Google search. We then open a Wikipedia article, and thirty minutes later, we find ourselves seven clicks down a delightful rabbit hole. That journey consisted of recommendations. Those links you clicked on weren’t in your original search.
“Let’s consider the ingredients of a good recommendation. Those nutritious ingredients, as opposed to the sugary ones.
“Firstly, you can look at the content itself. Is the recommendation relevant? Is it related? Are there some underlying concepts that relate to each other? We try to figure this out by doing a lot of natural language processing. We express those relationships in a vast data graph.
“Then there’s the behavior. Once people start interacting with the system we can learn a lot. The second you go live, all hell breaks loose! Users add an important subjective perspective and they are all quite different. The way we handle this separates us from platforms like Facebook. Most ad-driven systems will optimize towards clicks and popularity. The internet of today is based on popularity, not on trust. At Bibblio we want to change this. We not only look at every click, but we also look at the entire session. We work on replicating successful journeys through content to other users.
“Thirdly, it’s about introducing a form of personalization to the user. It’s one of those terms like AI and machine learning that people throw around a lot. We use it very carefully. I think personalization is great when applied responsibly. It’s quite easy to figure out what people like and give them more and more of it. Our Lead Data Scientist, Dr. Mahbub Gani, sums it up quite beautifully:
“Most recommender systems are a reflection of who we’ve been, not who we aspire to become.”
“Many recommenders don’t take chances, because in an ad-driven economy it doesn’t make any sense.
Exploitation vs exploration
“I’d sum up the challenge for recommender systems as ‘exploitation vs exploration’. You might say this sounds dramatic. These words are not mine. This polarization has been mentioned in research papers for some time.
“Exploitation uses what we know about the user to drive performance. Exploration considers the user’s values and how they have changed. For instance, if there are new preferences or other routes for them to discover. Where most systems use exploitative methods, we are building the opposite. It will still look at optimizing engagement and satisfaction, but it does so in a careful and responsible way.
“A big part of solving this puzzle is machine learning. There’s a lovely and exciting cycle going on. The more users are active within the system, the more we learn.
“Quality and trust are key themes for Bibblio. One of the things we do is conduct qualitative surveys. This is a control set of 2,500 recommendations. Users interacted with this set and rated how good it was. In the last year our average score has increased from 52% to 73%. This is a great result. Our system is always learning, and to keep it that way it’s important to measure and benchmark our output quality.
“For a publisher to take advantage of quality recommendations, they are turning to Bibblio to help deliver them. We are an API-driven B2B platform. You plug into the system and we automatically read all of your content and understand how it interrelates. Start generating recommendations immediately and your users will benefit from a trustworthy and nutritious diet of discovery.”
Bibblio is a content recommendation platform that helps content businesses and publishers deliver more relevant and engaging discovery experiences to their users. Visit us on Twitter, LinkedIn and Facebook.
More juicy posts by Bibblio:
From content to digital experiences
How recommender systems make their suggestions
The future comes highly recommended
Advertising vs Subscription
60 YouTube Channels that will make you smarter