As a part of Innovation Center Denmark’s ongoing series on applied artificial intelligence within the financial industry, we are bringing you the stories of some of the most interesting start-ups working on the cutting edge of applied AI.
In today’s blog post, we bring you the story of Ayasdi and its Swedish founder and Stanford Professor emeritus, Gunnar Carlsson, which currently have secured almost $100 million in venture funding.
Introduction from Senior Advisor Anders Christjansen
In the interview (provided in full below), I had the pleasure of talking with Mr. Carlsson about his journey from being a highly regarded academic to founding a start-up which is now one of the most promising within the field of AI. During the interview, we had the chance to touch upon the early days of Ayasdi, where it took several pivots to find the right fit between technology and market.
I probed for Mr. Carlsson’s perspective on the increased focus that companies as well as the public has lately placed on AI. While the arguments for utilizing AI are compelling, he argues that the implementation of AI systems is a decision, which requires consideration of the specific setting, problem and customer that it is intended to help.
... the goal should be to deal with problems that humans do not do well or where the aid of an artificial intelligence unrestricted by human biases and pre-set ideas can bring forward alternative results and solutions.
In the realm of implementing AI solutions, the goal should be to deal with problems that humans do not do well or where the aid of an artificial intelligence unrestricted by human biases and pre-set ideas can bring forward alternative results and solutions. Furthermore, to decrease mistrust, ease adoption in highly regulated fields such as banking and public services, Mr. Carlsson considers an AI system’s ability to justify and explain to humans the nature of the results and decisions that it is making, a vital consideration in the design process.
Many tasks will still be dependent on utilizing the strengths of a combined human-computer skillset, where AI’s enable human workers do what they do best.
When it comes to imagining how AI may affect future organizations, and the labor market, Mr. Carlsson is of the conviction that many tasks will still be dependent on utilizing the strengths of a combined human-computer skillset, where AI’s enable human workers do what they do best. It is apparent that we are still in an infant phase of applying AI, and that saying how, and to what extent, it will impact our society is hard to predict. There is however little doubt, that Mr. Carlsson believes that it is going to affect every asset of every sector in the society of tomorrow, and that while there will be some trial and error, the earlier that we do those trials, the faster we use AI to drive a positive change in the world.
Hi Gunnar. Thank you for taking the time to talk to us. We've already studied your impressive background at both Stanford and Ayasdi, but could you give me a short introduction of yourself and your company?
Certainly. I have a PhD in Mathematics from Stanford, where I also dedicated the first 25 years of my career researching so-called algebraic topology – one of the more abstract concepts even in pure mathematics. I was however always curious about how to apply my research practically.
In the beginning of the millennium, Gunnar eventually had a finding that showed potential in helping interpreting and making sense of big data sets. The project initially gained support from the National Science Foundation, and later enrolled in a 5-year project with the U.S. DARPA initiative. It was during that period what was to become Ayasdi was founded.
During that time, we worked with some things that we really thought deserved commercialization, so together with a former student and an old Stanford colleague we began to spin up a company. As the company became more and more established I came to the realization that one can really only do one thing at a time. Therefore, I’m now a Professor Emeritus at Stanford, which is basically a nice way of saying ‘retired’, and working full-time on Ayasdi.
Ayasdi’s founders (from left to right): Gurjeet Singh, Gunnar Carlsson, and Harlan Sexton
How come that you, as an AI company, have chosen to focus on public healthcare and finance particularly, and how do you see this changing over time as the technology matures and the industry develops?
When we first started our research at Stanford, it was for the sake of pure science. However, as time passed we began to think of practical applications. At first, we considered how it could be applied in big pharma, but also recognized that the technology was really quite broad - horizontal some might say - and in reality applicable to any context where people deal with large and complex data sets (‘big data’) .
While it may be obvious to people from the business world, it took us some time to realize that we needed to narrow our focus - become more vertical if you will - to a number of well-defined business areas. Rather than focusing on general life sciences, we decided that we could do some good by moving into healthcare, but also saw a large need for our competences in financial services. That was more or less how we ended up working within those two verticals.
Now, as we’ve gotten older and more experienced and brought in new people, we’re slowly but surely coming back to some government roots. What happened with Ayasdi was that we didn’t just want to sell a data analytics tool, but rather focus on building applications and enable others to do the same.
We’ve heard about AI within Healthcare and Fintech, but how do you see the prospect of AI in Public Administration, and do you see it evolve?
I do. Absolutely. As said, our technology is not exclusive to certain sectors, so our work in healthcare could easily be translated into the public sector. In fact, I'd say we're wide open and pushing in all possible directions - also on the federal side.
Google recently stated that they are now moving from being "mobile first" to "AI first". How do you see AI and the role that it plays in companies? Do you see it as having a primary or secondary role, and how does it impact companies?
To put it very simply, when it comes to the priority that individual companies put on AI, I see it as a question of being opportunistic.
My view is that there are certain things, or situations, where it's important to be able to have AI. Not only to have AI automate tasks, but one that is also able to justify itself and explain the nature of the decisions that it's making.
To qualify that argument, I’d like to point to the regulatory issues that exists in the financial world, where very large and complex machine learning models have been developed for different things. The issue is that these things are often a black box - one doesn't understand how they come to a particular output. So while it does represent a significant hurdle for regulators to accept it, it’s obvious that there’s a need for it, so what one has to do is build something that is suited for that particular task. A part of that is saying that "Yes, I need to do some things automatically, but I need to have them be justifiable, diagnosable, explainable and so on to humans."
In other situations, a simple case could be call center routing – pretty much an automatic process - to route a call - so there one can go more automatic. The main point is that one has to be flexible on this and understand what level of automatization one can go to. Ask yourself; What is the right level for the particular customer and particular task at hand?
Thinking in the lines of the impact that your technology may have, how do you see it impact your partners in terms of organizational structure, hiring practices, individual work flows and distribution of certain employees?
Overall, there will certainly be changes to some processes. However, what I think one should keep in mind is how it's going to allow people to focus on the things that they do best. That's a broad general statement, but to provide an example I could use banks employing a significant amount of people in anti-money laundering; A great deal of the job is analyzing large sets of data, transactions and so forth. However, 98 % or so of their findings turn out to be false positives. That's not a happy situation for neither employee nor employer. Therefore, by introducing an AI that can dramatically cut down on those false positives, and let the employees deal with the things that people do well, that will give them a better work experience and the bank less overhead.
In the end, it’s a spectrum. There are definitely tasks that machines can fully replace, but there's a lot of other things where the combination between man and machine is better suited.
Ayasdi provided this roadmap to help Mercy health system identify best practices in its hospital setting.
So it seems like you’re making a fairly clear distinction between what AI can do and what humans can do, in that AIs for example doesn’t have the decision-making skill that humans do, but are superior in detecting false positives and such? Do you think that the future goal should be for AI to take over the human side as well?
I didn’t mean to suggest that machines can’t do the decision making at all; actually I think it’s very clear that they can to some extent. Moreover, I think we will probably end up growing that level beyond anything that any of us can foresee right now.
We will find that we can start to trust them more towards making medical diagnoses and so on. However, in the end, it’s going to come as our experience with them strengthens. What that means is that if one wants to rapidly develop those kind of things, and prototype and test it to sort of see what do we trust and what don't we trust. Now we can't make these decisions beforehand. We kind of have to see how the technology works. My answer would be that we hope the machines will be able to take over some of the decision-making, but we have to be aware of it being a big deal, and that we have to talk to regular people as well as technologists asking not only pure technology questions.
What are the properties of an artificial intelligence system?
For us, that boils down to around 5 big things. It needs to be able to do unsupervised analysis really well. Narrow AI will kind of replace routine tasks, such as driving cars, manning call centers and that type of jobs. More broadly speaking of the concept, AI is supposed to deal with problems that humans don’t do well, or where there is no human template for how to handle them. When speaking of AI doing data analytic unsupervised, I’m referring to the fact that we can rely on the AI not to work out of biases and pre-set ideas about how things should be done. Additionally, you need to be able to do prediction and classification, as well as do justification, which is a big one. That's the one that I didn't understand how big it was until we got into this. Justification is about explanation of results. Diagnosing things when they go wrong. One tends to think of these computerized systems as something that works, but as with everything else they break down at times, and especially over time. So you need to be able to diagnose problems with them. Finally, you'd need it to act, or give recommendations for action, and ultimately be able to learn as well.
Therefore, what we're trying to do is hit all those sweet spots, or the actual realization of the applications, and try to build them; develop intuitive tools that allow people to build things that has those five properties very quickly. That's what we're about.
Thanks to Gunnar Carlsson from Ayasdi for sitting down and giving us some insight on his journey and vision. Our mission here at Innovation Center Denmark is to connect stakeholders in Denmark with those in Silicon Valley. If you, or your company, are interested in exploring how you may utilize applied artificial intelligence in your business, do not hesitate to contact us. And remember; the future is already here, it is just not evenly distributed.
Check out Ayasdi on their own website
Contact Senior Advisor Anders Christjansen for inquiries about our work here at Innovation Center Denmark and applied artificial intelligence in the financial sector.