Tell us more about Abzu and why did you decide to found the firm?
I’ve been interested in high performance computing for a long time, since the mid 90s. Engaged in the early community around the operating system Linux, I was an early adopter at using it at the university to build clusters of computers. Gradually I grew into data analysis and artificial intelligence.
The idea to found Abzu occurred over the years, that there was an interesting way to mix some of the methods that we used when we ran a specific type of simulation for quantum fields and artificial intelligence—an idea that nobody has actually ever tried.
This idea has been bubbling in the back of my mind for 20, 25 years to come up with a smarter way to do symbolic artificial intelligence, which is a subset of the approaches in AI. But of course, it’s quite a bold idea to just start a company and build an AI technology from scratch that you then expect to compete with things like deep learning and various methods that have been around and been developed over several decades.
It’s a crazy thought that a Danish man with a small team should be able to come up with a fundamentally new AI. In 2017, 2018, as I was thinking about what to do after having decided to leave another startup company I was a part of, I figured it was the time to give this crazy idea from my youth a try.
Seven of us eventually established Abzu in March 2018, and we all come from a background of high performance computing AI research and engineering.
Why focus on life science?
We were trying to build a system that could most likely explain the data that you had collected but not be limited to to a certain form of that explanation, like a linear model for instance.
The gist of my idea is to deal with infinite spaces, more highly dimensional infinite space – quantum field theory. We transpose the data question into a quantum field theory-related question and we solve it with some of the same methods that you use to solve that problem in quantum field theory. We then transpose the answer back into the domain of the data and now you have the best or the most likely answer.
Actually, the output of a system is not the most likely answer. It is a list, an infinite list of potential answers, ordered by how likely they are. That’s one of the one of the very powerful things for research – the system says this is the most likely explanation. This second explanation is almost as likely, third explanation almost likely, and so on.
That allows the researcher to utilise his/her own background knowledge to guide and control the search for hypothesis in a way that is not possible with classical traditional machine learning. We quickly realised that one of the groups that would benefit a lot from this was scientists working in natural science and even the humanities.
We needed to choose a first market for this technology, and were doing some projects with pilot customers in different industries, not only in lifescience. It so happened that we came across a very promising and interesting project with a big Pharma company that eventually sealed our fate.
It worked so well and there was a lot of scientific questions about activity and toxicity that could be answered with our technology. We quickly moved over to target-selection questions like what are the targetable things in the body for the research of new drugs and understanding diseases.
ABZU is not your first company. Can you share more about the other companies that you have founded and your career track to date?
I was at the university where a friend of mine and I came up with an idea to build a piece of hardware to actually make these clusters of computers more powerful. We created a company called Unispeed in 1999 and found a good product-market fit.
Unfortunately due to the dot com bubble burst, our entire customer base disappeared overnight, and I eventually sold it to a French VC in 2004.
I did a couple of years of sabbaticals. I wrote a fiction book called the ‘Land of Forgotten Gods’, had a few children, took on a few consultancy jobs here and there. I ended up joining a consultancy firm here in Denmark that did software development, product development and management consulting.
I also started to work as a consultant with a lot of VC companies that led me to very interesting tech companies in the early stage. I ended up joining one of those companies as their CTO as employee number 5, and it grew to 270 people over the years.
In 2017, I felt that I’ve taken the company as far as I could in the direction that I wanted it to go. I decided to leave and take another sabbatical. That sabbatical was what eventually led to ABZU.
What drives you?
I’m a ridiculously curious person. I just love learning, studying and understanding.
That has been the the centrepiece in my work, in all my startup companies – this quest for knowledge, understanding and learning. What is interesting is if you can really invent something, build spacecrafts or AI technologies and thinking machines.
What is your view on failure?
Learning involves failure, it’s that simple.
It’s a necessary part of any kind of innovation group.
You try things, you will fail. If you don’t try things, you won’t fail, but then you will never become smart.
What has been some career highlights?
By far the biggest career highlight ever in my life was when the QLattice, our technology, was proven to work. We took a leap of faith. We created a company, raised a lot of money, worked for several years at the keyboard, in the basement, trying things out.
We started getting results to explain things with an accuracy that is comparable to how a Black Box Model would explain or predict the same thing. There are also Eureka moments where the technology delivers in new insight that is then proven to be true.
Who or what has shaped who you are?
My mother is very special to me because although she didn’t finish her education, she was always this very intellectually curious person. I would spend nights, talking about philosophy, ancient history and natural sciences with her.
I also have a very strong passion and hobby for ancient Sumerian culture, and have taught myself to read and write Sumerian cuneiform, amongst others. You can never run out of things to learn and study.
Some lessons learned over the years?
Firstly, stop being afraid of failure—it’s like being afraid of pain when you run. It doesn’t make sense.
Secondly, unless you really know what you’re doing, don’t do what I’m doing. Don’t create a company where the tech comes before the product or change case.
Technology looking for a problem to solve is way more risky than than problems looking for the technology to solve. If you want to maximise your likelihood of success as a startup, find a problem and match that.
Thirdly, don’t follow these stupid recommendations from others. One person’s truth isn’t another person’s absolute truth. If I followed my own previous advice, this new approach to AI wouldn’t have existed.
If what you care about is building a a successful company, then find a problem and solve it. If what you care about is something else, then try to see if you can incorporate that into a company, accepting that there is a higher risk of failure.
What’s next for you?
Next is the challenge of combining Black Box and White Box artificial intelligence.
I’m not sure I want to build a thinking machine, but I would really like to know why we are who we are, and why we think. It is the ultimate question, and in a certain sense, I think we’re getting closer to answering it.