Martin Rosvall, Professor of Physics with a focus on Computational Science, Umeå University. Lead Scientist and Co-founder of Sift Lab AB.
In mid-November, I was invited to give a keynote speech at the Nordic Retail and Wholesale Conference 2021, organized at Umeå University. It was rewarding in several ways. Speaking at a live event feels exotic in times of pandemic. The contact with the audience on site is something completely different than staring into a Zoom grid.I talked about the limitations and opportunities of machine learning. My goal was to help the audience look beyond the AI hype. It's the best way to convert collected data into increased understanding and productivity. With an overbelief in too-good-to-be-true solutions, we instead risk heading towards a new AI winter when inflated expectations are not met.
Inflated Expectations
After a couple of examples of overhyped applications, including Google Flu Trends and machine learning for crime based on portrait photos, which have proven to be useless in practice, I went through the basics of how machine learning actually works. Many throw around AI terms but few understand the basics. This risks further inflating expectations without grounding in reality.Then, I drew parallels with our research at the intersection of machine learning and network theory. To understand many complex systems, from how people spread infectious diseases to how customers shop in retail and e-commerce, it is crucial to understand how the pattern of interactions - between people in disease spread and between customers and products in trade - affects how the systems work.
Beyond the AI Hype
At Sift Lab, however, we do not just help our clients understand how their complex systems of customers and products work. We also help them influence their systems through various activities, such as targeted campaigns. Therefore, I concluded by talking about an important difference between academic and commercial research: the analysis tools must be able to compile, model, analyze, and act on constantly growing data in a transparent process that the user understands and can add their specialized knowledge to without being absorbed by manual labor.Not without pride, I showed broadly how Sift Lab's platform differs from other solutions and what our customers think about using it.This week I am organizing a research hackathon for deeper collaborations between medical professionals and AI researchers. I hope it will be as rewarding as meeting the curious audience at the Nordic retail conference.
Three Tips for Utilizing the Opportunities of Machine Learning
1. Garbage in = garbage out. It doesn't matter how good the machine learning algorithms are if the available data is limited. Collect as rich data as possible and ensure compliance with GDPR. For example, there is much value to be extracted from the transaction data of retail and e-commerce merchants.
2. Invest in a transparent and flexible solution. Then you can combine your own expertise with the power of machine learning algorithms that grow with the amount of data. Uniting human insights with the algorithms' ability to identify patterns in large data sets is unbeatable.
3. The future is data-centric. With an application for every problem, it quickly becomes overwhelming, and valuable insights or activities are lost. A platform that integrates data management, analysis, and activities to leverage insights transforms data into growth without extra work.
