The Role Of Ai And Ml In Our Digital Future

Leonard Aukea, Head Of Machine Learning Engineering And Operations, Volvo Cars

Leonard Aukea, Head Of Machine Learning Engineering And Operations, Volvo Cars

BEFORE WE TALK MORE ABOUT THE MACHINE LEARNING SPACE TODAY, CAN WE HAVE A BRIEF ABOUT YOURSELF?

My background is in data science and machine learning (ML) engineering and I’m currently handling the machine learning engineering and operations at Volvo Cars. Before joining Volvo Cars, I worked across multiple industries. This experience exposed me to multiple different problems within the ML space, and I was able to find synergies and similarities in terms of what problems different businesses have, and what pain points were there. Once I joined Volvo Cars, I moved toward operationalizing ML technology in the current context. My main task here is to set up a dedicated infrastructure for machine learning and find common features of ML systems that can be reused across different projects. And in this regard, one of the core areas of focus is increasing the safety of the vehicles produced by Volvo
Cars.

SO IN LIGHT OF YOUR EXPERIENCE, WHAT ARE SOME OF THE TRENDS OR CHALLENGES THAT YOU HAVE WITNESSED HAPPENING IN THE ML SPACE TODAY?

I think the main challenges right now are not actually technical; rather they are mainly cultural. Also, I feel there’s a trend when AI and ML became this huge buzzword, and everybody just wanted to jump onboard and magically get a lot of value out of ML and AI.

They sprinkled the data scientists across different units and domains of their respective organizations and they eventually became siloed. But now we are realizing data scientists and ML engineers alone don’t really have the capability to operationalize machine learning systems and maintain them over time. Because most ML engineers have their background in academics and theoretical side of the technology and they lack the real-life business context to run the ML algorithms most efficiently. In this regard, building cross-functional teams that can collaborate with each other is one of the major organizational challenges apart from the cultural challenge.

Besides, machine learning is 100 percent self-aware software engineering. Therefore, getting a machine learning system to production means businesses need to treat it like software. But there are additional challenges that are explicitly related to machine learning due to the nature of the algorithms being stochastic in their nature, so you have to have to accept some margin of error in your results. This is also something that one needs to clarify when they communicate with stakeholders and actual users of ML technology.

I think the main challenges right now are not actually technical; rather they are mainly cultural

SO WHEN IT COMES TO YOUR ORGANIZATION ARE THERE ANY TRENDS THAT YOU ARE LEVERAGING IN YOUR ORGANIZATION TO SEAMLESSLY PROVIDE MACHINE LEARNING TO YOUR CLIENTS THERE?

We are working heavily on adopting ML ops, philosophies, and principles to streamline ML development and empower different demo teams across different domains. First of all, we are conducting educational sessions and building a foundation of organizational best practices. We are also developing a central team for maintaining and operating the ML infrastructure. For them, we have abstracted away certain tools into common APIs that can be easily used and accessed by these particular teams. We’re also pushing them to maintain and care about system design so that they don’t acquire too much technical debt over time. This allows us to have a central cross-functional team comprising ML ops, operations engineers, data scientists, and AI product managers, which enables us to streamline and deliver end-to-end ML implementation projects.

HOW DO YOU ENVISION THE ML SPACE THE NEXT 12 TO 18 MONTHS DOWN THE LINE? IS THERE ANY PIECE OF ADVICE THAT YOU WANT TO GIVE TO THE UPCOMING PROFESSIONALS IN THE FIELD?

I envision it being even closer to software engineering and development and I feel that this transformation is currently ongoing. However, there are currently two groups in this space, wherein one views ML as software engineering practices and the other group thinks of ML as a service offering. So we will be seeing contention between these two ideologies and eventually, it will evolve ML at a much faster rate as both groups have been covering some crucial aspects of ML. So I think there will be parallel streams. But in essence, the way we use ML will resemble more and more the way we do software.

In the ML field in general, it’s necessary to build a good foundation in terms of software. And it will definitely make you way more effective and increase the probability of you actually getting to the production stream in the first place; otherwise, the castle will crumble.