Growing up with machine learning

At Mediasmiths we regularly take the time to meet and talk to computer science students at Stockholm’s Royal Institute of Technology (www.kth.se). Recently I started asking students about their attitudes to machine learning and in those conversations it became apparent that machine learning has moved from being seen as a specialist subject to a mainstream skill. The consensus among the students I spoke to was that about 80% of computer science graduates take at least one machine learning course.

An estimated 80% of computer science graduates now study machine learning

Even if this figure is an overestimate, it is safe to say that soon a majority of graduates will have a good understanding of the fundamentals of machine learning. I say “good” rather than basic as they will know not just the buzzwords, but understand neural networks, know how the discipline has evolved, and what models to apply in different contexts. Crucially, they will also know the limitations of machine learning and therefore have a much better nonsense filter than most people today.

Let all of this sink in for a while and then imagine what it will mean when these graduates join software organisations where the majority of professionals have no formal machine learning training. Firstly, they will likely be frustrated by many of the misconceptions about machine learning, but beyond this they are likely to approach machine learning not as a specialist technology that requires a separate project, but as a technology that is built into projects as a natural part of the stack – like a database, or cache layer.

We will have a completely new generation of machine learning natives that will reach for the ML shelf without hesitating – a shelf that doesn’t even exist for most IT professionals, or even recent graduates. These machine learning natives are also likely to start using smaller-scale machine learning for “personal” use to improve their own working environment and productivity. Perhaps eventually leading to machine learning creating the software itself, referred to as Software 2.0 by Andrej Karpathy, Director of AI at Tesla.

We will have a completely new generation of ‘machine learning natives’

At Mediasmiths we have started this journey and during the summer of 2018 we prototyped how machine learning and custom image recognition models in AWS SageMaker can be used to detect events in football matches, for example yellow cards and corners. The initial results were very encouraging, but also highlighted how different this type of software development is, in all steps of the process. More about this prototype and what we learned in upcoming blog posts.

– Peter Henebäck