As I sit down to write this article, I can’t help but hum the tune to “If I only had a heart” from The Wizard of Oz. Who didn’t love the “heartless” Tin Woodman from L. Frank Baum’s children’s novel and then its film adaptation starring Jack Haley?
Like many, my introduction to the world of Artificial Intelligence (AI) began in science fiction stories like the above mentioned. Also, like many, I have been guilty of using the terms Artificial Intelligence and Machine Learning synonymously. I have since learned that they are not the same thing. Machine Learning, I now know, is a branch of AI and a growing trend with a significant number of organizations striving to adopt Machine Learning technology and potential applications.
At Explorance, we initiated the Machine Learning Center of Excellence over a year ago. So, I sat down with the man at the helm of this team – Alexis Tremblay – to get insight into the mind of someone who lives it each day. If I approached this expecting to grasp this mysterious and somewhat mythical world of Machine Learning straight off the bat, my ego would soon find itself hugely disappointed. The first tidbit of knowledge Alexis bestowed upon me was this:
“A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks T, as measured by P, improves with experience E.” – Tom Mitchell
For me, it might as well have been Parseltongue. Since the Explorance library held no spellbook that could help me decipher Tom Mitchell’s definition, I started our chat by asking Alexis to explain it to me. So began our interesting – though at times bizarre – conversation.
Q: For a mere muggle like myself, can you explain what the definition above is saying in everyday English?
AT: Yes, it is a very obscure definition. Essentially what it is saying is similar to when a human is learning a new skill by exposing themselves to new experiences; a machine needs tons of data to learn a particular task. The process of learning is periodically evaluated, and it improves over time.
Q: How does Machine Learning compare to human intelligence?
AT: It doesn’t. The first thing you should know is that Machine Learning is not human-level Artificial Intelligence. It doesn’t have the capacity to reason; it doesn’t have common sense, nor does it have abstract thinking. In other words, avoid anthropomorphism when speaking about machine learning. We all rooted for WALL-E, who was smitten with EVE, but it was just a robot in the end.
Q: WALL-E wasn’t just a robot. It was an adorable robot. Change my mind?
AT: Yes. WALL-E was cute.
Q: What would you say are some of the challenges and limitations of Machine Learning today?
AT: I think the challenges are in the human perception of what Machine Learning can do because they’ve been overhyped. A lack of proper understanding tends to inflate expectations. In my experience, I’ve been asked to do some weird things that approximated magic. Machine Learning is not magic!
Machine learning is not magic!
Q: So, you can’t use Machine Learning to help a muggle become a wizard?
AT: (Silent stare)
Q: How should non-magical people – like me – approach Machine Learning?
AT: Machine Learning must be seen more like a single function that does a single thing. Often, it’s more about automating one feature that can be used inside a product. This is the correct way to see and use Machine Learning. If I can give a specific example, we have a model that’s making three predictions, that’s the raw output from the model. Then there’s the step of how to use those predictions – the business logic. For me, that will be done outside the Machine Learning team. How you use the predictions is up to someone else. We’re not building a product; we’re building a feature. Eventually, everyone will understand that it’s one specific function that does one thing and does it well – single responsibility principle.
Q: What are you most excited about for the future of Machine Learning?
AT: For me, it all comes back to the Foundation series by Isaac Asimov. In it, a mathematician predicts the future based on what’s happening in society. I read it as a teenager multiple times, and then I discovered Actuarial Science, which is described as statistically predicting the future. Machine Learning is something like that – using historical data to predict what’s going to happen. It allows you to structure data in vast amounts and see patterns in the data. Humans are really good at pattern recognition, but we’re not aware of it. So, trying to reproduce that in a machine is fascinating.
Q: So, you want to predict the future? Are you also planning to take over the world?
AT: I plan to be Supreme Master of the Universe. Earth is too small for my ambition.
Q: What are you and the Machine Learning team working on currently? Let me guess, an algorithm to find the Tesseract.
AT: Nothing like that. We’re doing natural language processing and text analytics. We even tried to use Machine Learning to enhance our capacity internally for the annotators. We’re building tools that will make us more productive.
Q: Is there anything else you’d like the world to know about Machine Learning?
AT: Machine Learning was not well known; it was purely academic at first. Up until a year or two ago, it was in its infancy. Now it’s in its teenage years and democratizing. Enterprises are opening to the idea of stochastic output. I’m excited to see what’s next – there’s always something new occurring, and that keeps the field interesting.
Alexis Tremblay is a Machine Learning Engineer and team lead at Explorance, a Montreal-based software company and the world’s largest provider of Journey Analytics Solutions.
Blog•Blue Text Analytics•Machine Learning•