Foundations of Machine Learning
Why Machine Learning Is Taking Off Now
3 min read
A 65-year-old idea, a recent boom
If Arthur Samuel described machine learning back in 1959, why does it feel like a technology that only became unavoidable in the last decade or two? The idea was sound the whole time — what was missing were the practical conditions needed to make it useful at scale.
Three shifts changed that.
1. Computing power reached everyday devices
Training a model — and even just running one — takes real computational muscle: lots of repeated calculations over lots of data. For most of computing history, that kind of power lived only in research labs and specialized data centers.
That's no longer true. Phones, laptops, and even small embedded devices ("edge devices") now carry processors capable of running machine learning models directly, without needing to phone home to a data center for every calculation. That shift — powerful computation becoming cheap and ordinary — put machine learning within reach of far more products and companies.
2. Connectivity caught up
Machine learning models are trained centrally but often need to be used everywhere — on a phone in someone's pocket, in a car's onboard computer, inside a retailer's point-of-sale system. That requires moving data and predictions back and forth reliably.
Widespread, faster internet connectivity and bandwidth made that practical. A recommendation engine that needs to update in real time, or a fraud-detection system that needs to score a transaction in milliseconds, depends on a network that can move data fast enough to make the prediction useful before the moment has passed.
3. There's simply far more data to learn from
Machine learning is only as good as the data it learns from, and data has become vastly more abundant. Smartphones, wearables, connected appliances, and the broader Internet of Things constantly generate streams of data — location, behavior, sensor readings, purchase history — that machine learning systems can use as training material.
More devices capturing more data, continuously, is what gives modern machine learning systems something 1959's version never had: a large enough, rich enough body of experience to learn genuinely useful patterns from.
Key takeaway
Machine learning's core idea hasn't changed since 1959, but three practical shifts made it viable at scale: cheap and widespread computing power (including on edge devices), fast and reliable internet connectivity, and an explosion in the amount of data being captured every day. Together, they turned a research concept into infrastructure that's now built into everyday products.
What's next?
Machine learning is one member of a family of related terms — AI, ML, and deep learning — that get used interchangeably but aren't the same thing. Next, we'll draw clear lines between them.