By: James Raboin

Editor: Reid Varner

### History (AKA “the boring stuff you glossed over in school”):

As most tech things do, AI/ML (Artificial Intelligence and Machine Learning) started out with a bunch of nerds who had too much time on their hands. Between the 19th and mid-20th century, there were dozens of critical mathematical discoveries that led to the rise of Artificial Intelligence and Machine Learning. Most notably, Markov Chains, Bayes Theorem, and Turing’s Learning Machine were the forerunners of the AI revolution. Going forward, we will use Artifical Intelligence (AI) and Machine Learning (ML) interchangably, but there are subtle differences between how they should be used in a technical sense.

Between the 1950’s and early 2000’s, Machine Learning saw a dramatic increase in popularity. This was due to a multitude of things, but the most critical changes were the increased computing power of computers and the dramatic rise of data collection and Big Data. The former makes sense, right? A machine is only as fast as its slowest component, so obviously more complex calculations can be computed as computing capabilities increase. The latter, however, may not be so obvious. Let’s dive into how AI/ML works, shall we?

Real Image of the IBM Watson Supercomputer

### How it works: Skynet Edition

We can split AI/ML up into 3 categories: Supervised Learning, Unsupervised Learning, and Reinforcement Learning.

Supervised Learning: Think of this like a teacher/student relationship. The teacher (a human) gives input data and the correct “answer” to a student (the inputs could be “4 legs, Furry, Barks” and the answer would be “dog”). This ML student processes the data using statistics and high-level math and generates a mathematical equation. This equation then predicts outcomes of future events based on all the data it processed thus far. For example, let’s say we want to train our ML student to predict if a driver has received a speeding ticket in the last year. What we would do is compile a list of drivers in an Excel file and input data into it (the input data would look like “Type of Car”, “Age”, “Location”, and the output data would be “Has Received Ticket”). We would hand this over to Mr. Robot-o and he’d spit out a math equation based on the Excel file we gave him. He now tests out his equation on more data and sees that his equation is 93% accurate! Wow! Gold-star for you, binary buckaroo. If we wanted to test him in a live environment, we would just throw him onto a Highway median and see if he predicts the “Has Received Ticket” column correctly. Pretty neat, right?

A common real-world application you may be familiar with is the “Spam” filtering in emails. You, the “teacher”, tell the email server when you think an email is Spam or not (by clicking “Mark as Spam”). Over time, the server will have a huge list of spam emails that it will use to create an equation that will filter future spam emails!

I’m impressed you’ve read this far, here’s a kitten as a reward

Unsupervised Learning: The difference between unsupervised learning and supervised learning is one sole distinction: are we given the correct answers in advance or not? In supervised learning, the machine is given input data along with the corresponding correct answer. In unsupervised learning, the machine is never given an answer, only input data. It uses the input data to come up with its own “gut instinct” based on common patterns in the data. Think of this like giving a child 100 Matchbox cars and telling him to organize them. The child then takes the common features among the cars (color, number of wheels, body type, etc.) and forms 5 groups of similar cars. This is the “equation”; clusters of similar things. So now, after all is said and done, you hand him a red Corvette toy. He ponders for a bit and then assigns it to a group. You now know the “label” of the red Corvette AND have a group of similar cars to the Corvette.

So why is this useful? Think about Amazon’s recommendations. It operates in a very similar way. It “clusters” groups of items together based on how often they are bought together. So if someone buys the complete Star Trek collection on DVD, Amazon knows that the DVD’s are grouped with other similar items like toy Lightsabers, video games, and a one-person dining set.

Another common application is grouping businesses together to sharpen a company’s target market and finding common customers between them. Someone who owns a Ford Pickup, recreationally fishes, listens to country, and watches the NFL will probably be interested in buying cowboy hats.

UNRELATED PSA: HAN SHOT FIRST, NOT GREEDO

Reinforcement Learning: This is like teaching your dog commands. You give them treats when they do the right command and say “no” when they don’t. Let say we are teaching an ML model to play Chess using Reinforcement Learning. He would start out making many inefficient moves and would be “punished” any time it loses a piece. Conversely, it would also be “rewarded” any time it takes an enemy piece. Over thousands and thousands of games, it would learn to play Chess by avoiding bad things that happen and learning to proactively make good things happen. This field of study is incredibly complex, so I won’t go into more detail than that.

### Roll Credits:

We can wrap up the intro to AI/ML here. Consider yourself an expert in the burgeoning field of Machine Learning. I’ll phone Google and let them know that you all are adequately equipped to be working on bleeding edge technology.

Links for further reading:

Stanford Professor Gives a Brief AI Intro to Business Students:

Mathworks Intro to ML:

https://www.mathworks.com/discovery/machine-learning.html