What is AI, really?
Put simply, artificial intelligence is a computer making decisions for itself to accomplish an objective.
This can be as simple as if/then or the now-famous neural net. Instead of a human telling the computer what to do at each step of running a program, artificial intelligence allows the computer to make its own choices.
AI is already in your life. Assistants like Siri and Alexa might be in their primitive stages, and you may not trust them (with good reason not to), but Google’s search results are based on AI, Facebook shows you content in your feed using AI, Netflix tells you what they think you’ll be most interested in, YouTube blocks videos that break copyright law with AI, and financial institutions determine the likelihood of loan default with it.
There’s a lot of buzz around “machine learning” and “deep learning.” In both of these subsets, the developer acts as a teacher, explaining thousands of times what is right and what is wrong. Instead of the code being the logic and making the decisions inside the software program, the data is the logic.
Imagine teaching a 2 year old what a chair is. You can sit on a chair, you can place things on a chair, you can have a back rest attached the chair or it can not have a back rest. If a chair is over a certain height, it can also be called a stool. All this, and its primary purpose is for a person to sit on it.
Now teach a child what a table is. You can sit on a table, place things on a table, but it doesn’t have a back. It can be a high-top or it can be low. Its primary purpose is for someone to place things on it.
So you may point out 10 different chairs and 10 different tables and the child’s brain starts to recognize the patterns. Soon enough, the child won’t confuse the two, although it will try to stand on the table every once in a while.
This is how AI, both machine learning and deep learning, work. You show a model (a general code structure) lots of examples of one thing vs another vs another, and that model recognizes the patterns.
Then, when you show a model an example it hasn’t seen, it makes its best guess of what that object is. Just like if you showed a child a chair and asked it “Is this a chair or a table?” The child or model may get it wrong, but with each new example, the odds at getting the question right grow larger.
What is AI? It’s code that learns from the data you use to teach and extrapolates when you give it a scenario it hasn’t seen.
Sometimes, it’s telling the difference between a picture of a chair and of a table. If you gave the code all 7 Harry Potter books and then asking it to write the 8th. Sometimes, it’s predicting your likelihood of defaulting on a loan and sometimes it’s making a guess as to your sexual orientation from your Facebook profile picture.
So, what is AI, really? It requires quite a few examples to teach a specific task or concept, and then it extrapolates the examples to anything else you give it.
Like a child.