The importance of responsible AI to empower ethical data driven solutions
Lennart joined Mobiquity over six years ago to develop and lead the Data Science and Analytics team. He talks about how with a data driven mindset real world problems can be translated into numbers in order to create innovative solutions for challenging problems. He also shares his thoughts on the importance of ensuring Responsible Artificial Intelligence to enable ethical data driven solutions.
Lennart Bootsman: I was born and raised in the Netherlands. I have many different interests and hobbies. I’m an avid reader enjoying a wide range of topics, in particular philosophy of science, history, fantasy and novels. I like playing around with technology and practising coding. Also I enjoy gaming, both physical board games and computer games. And when I have some extra time I like going to the theatre, movies, and arts.
Since I was a young child I have always been interested in technology. I was one of the first people in my junior class to have a computer at home and that is how I started out learning to program. Since then computers and technology have played an important role throughout my working and personal life. During my studies, together with some friends of mine, I actively got involved in hacking and cracking programs and software. We purchased everything legally (my parents made sure about that), but we were just interested if we could do it. We were challenging ourselves to take something apart in order to truly understand how it worked at a deeper level and figure out what was needed at the core to make that something work. Not only was there a thrill and excitement associated with this, it also acted as a confidence builder for myself when I was able to succeed against the odds, jumping into the unknown, taking a weakness and figuring out how to make that successful. In a sense that was the starting block that led me to my career in Data Science & Analytics. Taking an uncertain situation, creating structure and making something tangible that people can use. And I find that this is something I enjoy the most about my job, not knowing exactly what the next problem is and having to figure out creative solutions in order to get results.
Building Mobiquity’s Data Science and Analytics department
I graduated from Delft University of Technology with a Masters Degree in Technical Computer Science. After completing my studies I wasn’t quite sure what I wanted to do. I decided it was a good opportunity to improve my language skills, in particular French, so I moved to France and started a French language course there. After six weeks I was offered a job by the teaching institution to manage the IT department and develop a new IT system. Three years later I moved back to the Netherlands to start working for a large telecommunication company. I worked there for about twelve years, where I had various roles within the IT and Innovation departments. One of the biggest programs I was a part of, was the roll out of Fiber to the Home in The Netherlands. For the next stage of my career I moved into the field of Data Analytics, working for a product company. And six years ago, I joined Mobiquity who at the time were looking to set up and develop the Data Science and Analytics department and competence.
Why Mobiquity? Before joining I had only worked in corporate settings where there are a lot more processes and structures set in place and less flexibility. Joining Mobiquity gave me the opportunity to be in the lead of shaping and moulding the future of the Data Science and Analytics department. As the Director of the department it is by far the most entrepreneurial role I’ve had and where I can ultimately contribute the most too. It’s a great and ever evolving challenge to continuously develop and progress the department. And I have to say that I enjoy working with the people at Mobiquity a lot. Everyone’s an expert within their field. People help each other out a lot, it comes naturally. Even though there has been significant growth in the company over the last few years, new colleagues can adapt quickly and contribute. It’s a great feeling to work as a team and achieve something bigger than the team itself.
Translating a real world problem into numbers
For any project within EMEA that includes Data Science & Analytics, my team is involved. The team consists mainly of Data Scientists who combine a solid background in mathematics with coding and creative problem solving skills. Data Science itself is essentially the craft of translating a real world problem to a mathematical space where it can then be solved. Once solved the mathematical outcomes need to be translated into real world solutions. Within a project our role is to utilise our data driven mindset by applying Artificial Intelligence (AI) and Machine Learning (ML) in order to create new insights. We start by understanding the end goal of our client. What exactly are they looking to find out more about? We then work with the client to either gather or generate the required data. Using this data we crate powerful ML models which can be incorporated in a digital solution and can provide the outputs the client is looking for.
The translation of a real world problem to a mathematical space means that what we create is functionally more or less the same across business problems. The challenge lies in tuning the functional outcome to desired business outcomes. For example, one client may want to predict what their customer will buy next, or the quantity they are likely to purchase, to optimise their supply and demand operations. Whereas another client may want to understand how much money each customer is likely to spend over the next months in order to better predict their financials. In the most basic sense we are predicting what will happen, but from a content perspective there are a vast amount of variables to consider. Therefore, each end solution we create is always different and unique to each client.
Creating innovative solutions for challenging problems
We have many interesting projects that we have worked on. One in particular to note started when a few members of my team won a hackathon, which led to an engagement with a leading Oil & Gas processing plant. The basic question was if we could optimise the gas processing plant. Gas was collected from the ocean floor and then processed in their plant. It takes energy to produce energy and we had to figure out how to make that process more efficient. By taking a vast amount of sensor data from the gas processing plant, we developed a data-driven digital twin that accurately mirrored the plant’s operations, which we could then use to predict what the optimal settings would need to be for the plant to operate most efficiently.
Knowledge sharing and development is something we value very much in our team. That is why we have, on an ongoing basis, multiple students doing their internship at Mobiquity. I enjoy this a lot as it keeps me feeling young, but also brings a fun and inspiring atmosphere to the overall team. Those doing internships are always enthusiastic, working on some of the latest innovations. It’s a great way for the team to stay in touch with the new technologies, understand what’s going on and also to determine faster if there is something in particular that may be beneficial to look more closely at within the department or for a specific project.
We also like to do a lot of knowledge sharing as a team. Where we work on innovative and challenging projects to experiment with the latest tools and technologies and see what we can learn from it, and how we could apply and utilise it within a future project. What we are currently working on, still in the experimental phase, is developing bots that can play board games using Reinforcement Learning (RL) techniques. Reinforcement learning has a huge potential but we believe it is very challenging to apply to real world problems. That is why we decided to start with board games, because a board is essentially a small self contained world with a set of given rules. We started with tic-tac-toe, which is a simple game but developing a bot for a simple game poses challenges. We have now mastered that and are currently investigating what game we will move onto next, and in the end assess how we can apply RL to complex real world problems.
Uncovering the complexity of ensuring Responsible Artificial Intelligence
Responsible AI, asks the question: how can we ensure that AI understands our norms and values and intentions so that we can trust the outcomes of AI as correct, unbiased, and fair. It’s a very intriguing topic as it covers a range of diverse and challenging issues, which we as a society are yet to truly understand or develop a viable solution for. Within cultures we have sets of norms and values which we are looking to be embodied within our ML models. However, on that level we are seeing a lot of things going wrong. We find many examples of ML models having biases against minorities. There are multiple reasons for this, but an important factor is that in order to train a model you need to have a lot of data. What traditional ML models do is replicate the past towards the future. And within historical data minorities are typically underrepresented. Another example of such shortcomings in historical data is its inability to portray an accurate representation of the continuous change and evolution of cultures without reinforcing past norms. As data by nature is something that has already passed, how can we ensure that it represents the present and future without replicating past inequalities?
We have given a lot of control to the computer and we know they are making biased decisions which we as humans may not agree with. However, as AI is all around us, it is almost impossible to know exactly when a decision is being taken. With so many layers, we can no longer pinpoint at which moment it goes wrong. As we are unable to understand how an algorithm gets to a decision, we struggle to know if that decision is biased or not. And if we take a step even further back from just saying the computer itself is making biased decisions, and look towards the much deeper problem of human bias, it further complexes the problem. Each person, because of their upbringing and cultural perspective, has their own biases whether realised or unintentional. Each decision you make is influenced by your past, similar to a ML algorithm. Even if you are willing to admit you are biased, how can you say exactly what your biases are? It’s extremely difficult to get that to the surface and uncover what is the ground truth for being biased. With AI/ML being applied deeper and more widespread in our society and as the world becomes increasingly more digital it is an extensive and fascinating challenge we must address.
We hope you enjoyed reading this, and don’t forget to follow us if you’d like to read more interesting stories.
Interested to join the Mobiquity team? Check out our open vacancies:
Originally published at https://www.mobiquity.com.
