Don’t be scared!
The tech sector has a tendency to over-hype things at times. A few years ago, I started noticing that every job listing or start-up would make sure to add the machine learning buzz words every chance they would get. This led to a bit of insecurity and FOMO.
I counter-acted for this by moving my focus away from the marketing speak and started talking to some real tech folks working in the domain. These helpful folks set my mind at ease by telling me that it was just a bit of maths and statistics in action. They advised me to just start dabbling with the technology and learn a bit of the maths and statistics that goes behind it.
Some examples
Once I started understanding how ML worked, it became clear to me that I could do some pretty cool things with it in many different areas. For instance, not so long ago, online stores like Amazon would recommend other products you might be interested in based on some basic algorithms. This type of recommendation engine would be a perfect candidate where an ML-based system would do a much better job. Better recommendations would result in better sales and everyone wins.
Generally speaking, any situation where you have lots of historical data and you want to predict how things will go in the future, is a good ML candidate. However, it is important to understand that the old adage .. “garbage in, garbage out” holds true for ML systems as well, maybe even more so.
I remember when Gmail came around, I was extremely happy with the spam filtering provided by the service. It’s unclear to me if it was always an ML-based system but surely nowadays they use ML to improve the filtering to counter ever evolving spam mechanisms.
Surely, you have heard of some of the famous ML applications like Google’s AlphaGo or Telsa’s autopilot but it’s important to realise that gradually ML is becoming pretty ubiquitous. A quick example of this that comes to my mind is the use of ML in the financial sector where they use intelligent systems to detect fraudulent transactions.
Expect no miracles
Once I started understanding more about ML, I frankly was a bit underwhelmed by the whole thing. ML is great when it comes to solving very specific problems but we are still quite far from general AI systems that are capable of learning and performing everyday tasks. So don’t worry, computers are still pretty dumb!
Hardware needs
I want to point out here that any reasonably sized machine learning system requires a lot of “learning”, which basically translates into need a lot of hardware resources to make it happen. I am now a bit used to waiting for my example ML code to finish “learning” after every change I make. This can be generally countered by working on multiple things at a time and moving to the other task while the ML system is busy learning.
Thanks to the cryptocurrency/block-chain craze, most of us have access GPUs at a reasonable cost. As GPUs are extremely fast with mathematical operations and ML relies on lots of maths, it’s a match made in heaven! It’s also important to mention here that with scalable cloud infrastructure at everyone’s fingertips these days, things have become much easier than before.