Putting Humans in Control of Machine Learning

I’m the Machine Learning Engineer here at Kinzen. Working in a small team with a big mission means wearing many hats, but broadly, my role is to surface insights from our data that will help improve the experience of being part of Kinzen.

Before joining the team in March, I had been working as a mathematician in academia. While being part of building a startup often seems a world apart from that environment, the core challenges and motivations that made a career in mathematics appeal to me are also what drew me to work here.

Fundamentally I am driven by three desires in what I want from my work. Firstly, I want to understand and solve interesting and challenging problems. Secondly, I enjoy sharing my work by finding ways of explaining and presenting the problems I’m working on and receiving and learning from feedback. Lastly, I am driven by a desire to contribute to a wider community, both within my profession and outside of it — to feel that I am helping people and having an impact on making the world a better place.

We are committed to putting the user in control of their experience and this means being able to explain what our algorithms are doing and why we’ve made the choices we have around data collection, processing and analysis. Explaining what a machine learning algorithm is doing and how it reaches its conclusions is a notoriously challenging problem, but it’s a challenge we’re determined to take on and one I’m excited to face.

Our goal is always to give the user more control, and we will never use an opaque algorithm to take away user options.

Our membership model means we are not dependent on advertising revenue, and we have no vested interests in recommending one source over another. This means we are not forced to optimise our systems for getting the most clicks, or shares, regardless of the quality of the content.

We are free to build a system designed only to serve the needs and intentions of our community. Where we implement algorithms for suggesting sources or topics to follow, we will be clear what data we are using and what outcomes we are optimising for.

Although GDPR has put data collection firmly back into the public consciousness and forced data collectors to justify their reasons, too often in the past it has been taken for granted that if data can be collected, it should be. Many sites and services are still collecting vast amounts of data from their users with little change other than a more complex opt-in/out process that many users will never explore. We’re determined to do better and will ask our community to hold us to account.

Our decisions at all levels are discussed in a multidisciplinary team, ensuring we have to present our work in a way that is accessible to people from different backgrounds and with varied skill-sets. We strongly believe that this helps us navigate the traps of unintended consequences and tunnel-vision that far too often can lead projects, especially those using machine learning, into unexpected and undesired territory.

There is no question that Kinzen’s mission tackles challenging and interesting problems. Aside from the technical challenges involved in our data driven work which require a lot of thought and problem-solving from our whole team, from data collection, labelling and storage through to surfacing useful insights for our members, Kinzen is attempting to address some of the biggest issues facing society today.

We’re under no illusions about the scale of that task. We want to create a news platform that is designed for the user, with trust, transparency and control at the heart of the experience. It’s going to be a tough journey ahead, but I’m excited about where we’re going.