The Center for Data Innovation spoke with Peter Sarlin, co-founder and chief executive officer of Silo.AI, a private AI laboratory based in Helsinki, Finland, that builds AI solutions for various industries. Sarlin discussed where AI is most impactful, how businesses can make the most of its potential, and what the technology can achieve.
Eline Chivot: How did you get into AI, and can you describe what Silo.AI is about?
Peter Sarlin: Being driven by creating sustainable impact with new technologies, I have a long background in academia, particularly in applied machine learning. After the financial crisis back in 2009, the research group I was heading built an AI solution for central banks with the goal to safeguard financial stability globally with a machine learning based tool that predicted systemic financial risk. At the same time, we were involved in a large number of other high-impact industry projects.
The seed of Silo.AI was planted after we built this solution; many others were deployed into production and used globally. I saw that there was a need for more collaboration between academia and practice. Fast-forward a few years, my co-founders and I founded Silo.AI to build the next European flagship AI company. We wanted to democratize access to world-class AI expertise and tooling through serving our clients with unique capabilities of delivering AI-driven products. Ultimately, relying on our strong academic and technology background, we’re creating value in practical industry use cases through a unique understanding of the interaction between the human and the machine. This allows us to build AI for people for a world with safe human-centric AI that frees the human mind for meaningful work and empowers human creativity.
We managed to gather a team with world-class expertise, people who are leading academics in their research field. At the moment, the team consists of more than 70 AI experts, most with PhDs or otherwise long experience in AI-related R&D. We’re headquartered in Finland, with offices in London, UK and Palo Alto, US, but we’re currently expanding in the Nordics and wider Europe.
Chivot: Where do you believe AI can be particularly impactful?
Sarlin: Rather than artificial general intelligence or superintelligence, today’s AI is narrow and for specific, well-defined problems. Yet, through this, AI is nevertheless impacting most business processes, changing organizations, the way we work, and the society at large. And this significantly impacts the way value is created in organizations.
One of the most central technologies behind the uprise of AI is machine learning (ML), a field concerned with machines that learn without being explicitly programmed. Within ML, what we can see being applied successfully today is supervised learning, the task of teaching a machine with large volumes of labeled data. This is nothing else than a mapping function between “A” and “B,” or inputs and outputs, covering examples like cats in images, traffic signs in video, or translations between languages. Now, in the real world, there is typically one significant challenge: Where to get good enough large-volume labeled data.
We are currently creating most value within areas such as smart vehicles, ports, cities, and devices, covering industries such as automotive, maritime, heavy machinery, medical, and many others. This relies on being able to collect high-quality sensory data as a precise measurement of the world as it happens. In such environments and contexts, it’s most often also highly impactful to build AI-driven solutions close to sensors and devices. It’s needless to say that the other large impact lies in helping us, humans, digest in optimal ways the information flow that digital means provide us access to. This oftentimes involves building intelligent applications to personify, recommend, and filter information in optimal ways for the end user. In creating these types of solutions, we function as a trusted AI partner for our clients, in development projects that are core to their business, and accelerate it with the competitive advantage provided by AI.
Chivot: When is AI particularly suited for data-driven decision-making? Can you give an example of a use case which particularly reflects that?
Sarlin: In data-driven decision-making, we usually face situations with a wide range of data types need to be processed simultaneously in order to recognize patterns that eventually drive decisions. Oftentimes, machine learning and AI both provide opportunities and face challenges in such environments. This comes from machine learning mostly being suitable for narrow, well-defined and repetitive tasks, whereas humans excel in tasks that are complex and creative or relate to human relationships. Thus, often machine learning needs to become part of a workflow that is both repetitive and collects a lot of data, but in which people are also able to make use of the results, be it predictions or suggestions, in order to make better decisions.
A few examples of how AI helps people in their daily decision-making are cases where we improve the situational awareness or assist in laborious tasks with machine learning. The Finnish airline Finnair is one case where we worked together with the Finnair Operations Control Center to improve their visibility to and awareness of the overall flight situation at Helsinki-Vantaa airport. Our machine learning model makes predictions on the flights that are likely to get delayed due to weather or a number of other conditions, and hence improves the overall level of adjustment to the upcoming situation. In similar ways, we are also helping automotive, maritime, mining, and other industries with, for instance, assisted driving as a first step towards autonomous vehicles or visual quality control and predictive maintenance as a way towards fully automated production processes.
Chivot: How can companies use AI to develop new business models?
Sarlin: While at many levels investments, advances, and deployments of AI are in certain sectors lagging behind, Europe provides a unique environment for AI-driven solutions, transformation, and businesses. If we start by asking the question: “What is going to happen to industries during the AI revolution?,” on top of the known dynamics of platforms, ecosystems, and network effects, AI will leverage the business models of software to significantly accelerate their value creation. And most of this relies on having access to closed-loop data by collecting from every interaction inputs and outputs, or eventually predictions and outcomes, such as driving behavior and patterns, and production process measurements such as production quality and machine failures.
For many incumbents, this is an opportunity to create models with their data and close the loop with existing processes or customers. This implies thinking about your existing business with an AI mindset. A good example is the traditional explosive manufacturing company Orica that transitioned into a scalable service by building a digital service for explosion optimization by collecting feedback from every explosion. An example of incumbents with potential is the wider industrial sector, which is also the backbone of the European economy, accounting for 80 percent of Europe’s exports. In addition, as data is at the core of AI, Europe has been paving the way for data privacy. While setting certain obstacles to short-term gains in AI development, this provides a much more stable and sustainable basis for long-term success in wider use of AI.
Chivot: What is the state of the technology in some of the sectors you work with, how will it evolve, and what will it be able to accomplish in the future?
Sarlin: Unlike how it is generally perceived, AI is rarely about fully autonomous systems that work, learn, and evolve by themselves. In my view, today’s AI at its core is about executing narrow, repetitive tasks that help people do their jobs better.
All of this relies on advances in machine learning, the technology behind AI. ML is frequently divided into three types of learning: Supervised learning, which implies learning from examples (or labels); unsupervised learning, which implies learning with no examples; and reinforcement learning from rewards when taking actions in a given environment.
While the media has extensively covered advances in unsupervised and reinforcement learning, we easily forget that most of this is either research or applied to problems that are unrealistic in most real-world settings. As an example, Google DeepMind’s AlphaGo is a remarkable breakthrough in applying reinforcement learning to the game Go with superhuman performance, and also applying the same approach to other games such as chess, Atari, and more.
Wouldn’t it be impressive to see unsupervised learning on its own start to comprehend problems and solve one after another? Unfortunately, that is rarely what we can do in the real world. What we can see being applied successfully today is mostly supervised learning, the task of teaching a machine with large volumes of labeled data. This is nothing else than a mapping function. In the real world, we can identify cats in images, give personalized product recommendations, identify defaulting creditors, convert speech to text, and translate sentences to different languages—among many other examples.