Artificial Intelligence (AI) is the reason why there is a fair amount of enthusiasm in the world about this technology and that’s why I decided to write about it today.
I am from the opinion that I can better understand new things when I know what they are used for. Therefore, here are examples of how AI is being used in the market today:
- Predict which molecules will most efficiently bind with which proteins for the purpose of drug discovery.
- Predict which credit-card transactions are fraudulent versus legitimate.
- Predict, based on medical images, which tumors are benign versus malignant.
- Predict which automobiles have defects before they roll off the production line.
The best way I found to describe how to explain what is and how people can use AI is by adding a layer on top of this “black-box” technology and use something that most people are familiar with: economics.
Today in AI feels a lot like 1995 felt with the Internet. Most people will remember 95 as a transaction year in terms of the internet (Bill Gates wrote the Windows 95; Netscape went public, etc) and the reason for that is because the cost of communication, search and selling goods online was dramatically reduced. All the underlying economic models that have driven the Internet (supply <- demand; production <- consumption; prices <- cost). The only thing changed, is that the relative cost of a few key inputs has fallen dramatically; and it is the same with Artificial Intelligence. The reason why AI is important is that the thing for which the cost falls is a foundation input into such a wide range of activities that we conduct, and in the case of AI, that thing is the prediction. From now on, every time you read something about AI, replace it with the words: “cheap predictions”, and after that, the article will become less abstract and more practical.
How do you define prediction?
Prediction is defined as taking the information you have to generate information you don’t have. That includes what most people would traditionally call demand forecasting (take the last five years of sales to predict a forecast of the next quarter sales) and a less obvious is something like classification (a medical image that is the information we have and the information we don’t have is either a tumor is benign or malignant), the AI generating that classification, we call it prediction.
Prediction is taking information you have to generate information you don’t have.
The AI in the business side becomes an art more than a science because we will start converting non-prediction problems into prediction problems. Here is an example of how we are converting non-prediction problems into prediction problems: driving. We have had autonomous vehicles for a long time, but we always deployed our autonomous vehicles in a controlled setting so a factory or a warehouse, and way it was done is that they would program a robot to move around the factory floor and then they would give the robot a bit of intelligence. They put a camera on the front and would tell the robot, if somebody walks in front, then stop. If the shelf is empty, then move to another shelf. If, then.If, then.If, then. The problem was that we could never put that robot in a non-control environment because it would result in an infinite number of ifs. With machine learning, it was possible to reframe the problem: replace the infinite number of ifs with one prediction: what a person would do in this situation. The idea is to put an AI computer in the passenger seat and “observe” how a human would drive. In the beginning, the AI is not a very good predictor, however, with time and lots of data, they create their model: if the prediction of what the human was going to do is correct, they double down on their model; and if they are wrong, they take their model, making a different prediction potentially the next time the driver would do the same type of movement. As the AI “learns”, it will get to a point that the AI is such a good predictor of what the human driver would do, that the AI can just do it by itself. Converting problems that were traditionally prediction problems into prediction problems to take advantage of the new cheap prediction tool is what makes the market so excited about artificial intelligence and machine learning.
When the cost of something falls, it affects the value of other stuff, in economics that is called: “cross-price elasticity”, (The value of compliments go up and the value of substitutes go down). The capabilities of machine prediction increase, the cost of machine prediction falls and the value of human prediction will plunge, so human capabilities become less and less valuable because the machines can do so much better/faster/cheaper. As the price of machine prediction falls, human judgment, the input data, and the actions become more valuable.
Data is the new oil.
Ok, you might be thinking: if AI is just prediction and lowering the cost of prediction why is there so much fuss about it? The answer is because prediction is a key input to decision-making, and that is everywhere. A good CEO today is the one who makes good decisions. The goal is to have tools that help us make a decision in order to achieve better results, either into pharmaceutic companies, car manufactures or human resources, and that is why AI is so motivating today.
We are working on some great Artificial Intelligence models and if you would like to know more about what we are doing or if you would need any assistance when creating an AI model, don’t hesitate to contact us.
Thank you for the time and see you soon!
*Source: Prediction Machines: The Simple Economics of Artificial Intelligence, by Ajay Agrawal, Joshua Gans, and Avi Goldfarb.