SUBSCRIBE TO OUR BLOG
Before we get into it, let me clean up some of the preconceived notions that pop culture has foisted upon our industry. When we put AI to work, we’re using it to be complimentary and supplementary to human decision-making. We’re not going to surpass human decisioning because nothing – nothing – can replace the creativity of the human mind. That’s why John Connor always wins.
What we’re doing, more precisely, is applying machine-decisioning to make human-decisioning more effective. We’re taking out the underlying work that computers can easily do, and setting the human mind free to work on the bigger and more complex problems.
Let’s start with the context of Machine Learning and this broad term called Artificial Intelligence. Machine Learning is synonymous with Neural Networking. What’s that? It’s a networked system that mimics the human brain in how it makes decisions. We can teach it, and it makes decisions on its own.
In the world of Neural Networking, there are two broad categories. These are: Supervised Learning, and Unsupervised Learning.
Supervised learning is the model where humans train the machine to perform a task. Imagine, as a light example, teaching an algorithm to sing the ABCs. We define the letters, we define the order and the notes. Eventually, the algorithm can sing its ABCs quickly, clearly, and accurately. Then, we can add that algorithm to a network, and it can sing the ABCs any time the network needs.
Unsupervised learning is a bit different, in that is doesn’t need any human interaction. You can feed an algorithm a piece of information, and it can train itself based on the information it receives.
For a better case for Supervised Learning, let’s look at the posted speed limit on the highway. When you see a speed limit sign, you know all your numbers and letters and you can easily read it. The camera on your car can see that too – and like the above, we’ve already taught the algorithm its letters and numbers.
But in this case, it needs more supervision to learn specifically how to identify a speed limit sign. What is the shape of the sign? Does it have the right colors? In addition to the words Speed Limit, does it also have Trucks or School? Once the algorithm is taught all these factors, it can identify a speed limit sign quickly, and accurately, and we can let it do that job within the network.
For Unsupervised Learning in the real world, think of a car travelling down the street. That car has sensors that report its location with latitude and longitude, and that gives us an idea of its location. As the car travels down the street in a certain direction, the coordinates are reported, and there’s always a chance that those coordinates could be off.
In this case, the car may report that it’s in the street, then suddenly report that it’s inside a building next to the street. But in reality, the car is still moving, and the vehicle hasn’t reported any other unusual data. We can tell an algorithm to learn that this is errant data – that it can be corrected and dismissed – and it can do so all on its own.
We have to be able to ingest and process an extreme amount of data. The way to do that at scale for a global market is applying these algorithms to do the heavy lifting for us. It’s a fundamental part of our Open Location Platform, and it’s how we’re powering the tech that will bring us into the future.