Podcast: Moving an Industry into the Age of AI with Martin Stamenov (Team Lead Machine Learning)

Martin Stamenov already starred in the 60th episode of the AI in Action Podcast back in December 2019, discussing the effects of Data Science, Machine Learning, and Artificial Intelligence on our daily lives. The first episode quickly turned into one of the five most popular episodes in the history of the podcast. Small wonder that he was invited to talk to Mark Kelly again, only around six months after Martin made his first appearance. This time, he addressed the topic of “Moving an Industry into the age of AI”. Read up on some of the most important things Martin had to say.

Automating complex tasks for greater efficiency

Machine Learning enables the automation of complex tasks and processes. Usually, these are processes that companies have to dedicate a lot of time and costs to. This is why they can save a great deal of important resources by implementing Machine Learning. However, the technology is still much too rarely used. One example Martin names is the translation sector as an old industry because historically speaking, these are the industries which are rarely at the forefront of technological advancement. This is a missed chance as Machine Learning could boost efficiency in two ways. Project managers have to carry out the difficult task of finding the most suitable candidates for certain projects from a pool of freelance translators. If there are thousands of translators who all work in different language pairs and bring different strengths to the table, this selection process can be arduous. This is where Machine Learning can be useful, making it easier for project management to search for the right translators.

However, the technology is having an even greater impact in another field. Translation processes are still predominantly manual, which is costly and lengthy. Machine Learning can create a pre-translation, to take over the most time-consuming components of the process, and produce texts that approximate the quality of human translations. This not only saves a massive amount of time and costs, but also helps the translator in the process.

Five steps forward, one step back

To successfully implement Machine Learning, companies need to take a step back and look at their processes. Martin recommends a multi-tier approach, that starts with an in-depth analysis and definition of processes that should be automated. Without this analysis and the detailed insights on processes this provides, it is simply not worth making a decision on whether or not to implement automation in the first place. During a second step, a closer look is taken at all technologies that can be optimized. Only by understanding which technologies are actually available on the market, can companies benefit from the optimal applications of Machine Learning. Technological advancements can open up new possibilities all the time — which is why this is should be an ongoing process.

Data as a reality check

Another aspect in the introduction of technology, which is often underestimated, is the importance of data. A considerable quantity is required to introduce Machine Learning to companies. If a company already has established a data routine, implementation becomes easier. Pre-trained models, in the form of Transfer Learning, are also immensely helpful during implementation.

"Data always forms some sort of a reality check."

Even if the perspective of automation and greater efficiency might seem alluring, it should definitely not lead to false assumptions regarding available data. Martin says that data always forms “some sort of reality check”. Without data, there can be no Machine Learning, regardless of how well processes are defined in advance.

Keep an eye on new developments

During the discussion, Martin places great emphasis on research, recommending to always keep an eye on the most recent developments. For example advancements in NLP only recently resulted in greater efficiency and user friendliness of the technology. To continue to make the most of Machine Learning now as well as in the future, new developments must always be monitored closely.

Listen to the entire podcast here: