George Polzer: What is AI?

This is the first article in a mini-series exploring the interconnection of Artificial Intelligence (AI) and Blockchain by our Strategic Alliances Manager, George Polzer. George is an AI Business Strategist and Chairman of the AI & Blockchain SIG of the Enterprise Ethereum Alliance.

What is AI?

The best starting point to learn about AI is to understand what an algorithm is. Given a list of positive numbers, identify the largest number in the list. The algorithm or procedure will determine if the 1st number in the list is larger than the second, if yes, number 2 is discarded and repeat. This is an algorithm! It is a mathematical recipe; a step-by-step procedure for solving a problem or accomplishing some end and typically by a computer.

The study or science of algorithms dates back to 2000 BC, when the Egyptians developed the earliest known algorithm for multiplication without using tables. Data structures and algorithms are one of the major subjects studied by every computer science student. In 1998, Larry Page, the co-founder of Google, published his Google PageRank algorithm to estimate the importance of a Web page with his PageRank method with no machine learning. Although all AI and machine learning are built with algorithms, not all algorithms are AI. What differentiates them as AI algorithms is the capability to self-learn just from the patterns in the data sets. The algorithm self-trains without requiring software engineers to write explicit programming logic to do so.

The AI Algorithm

For example, given the data points in the below chart, find a line through the points so that the line minimizes the distance all the points are from that line. The AI algorithm that solves this is called “linear regression” and the more technical term that qualifies this as machine learning is “gradient descent” or “optimizing a cost function”. Linear regression is the simplest form of AI yet offers the most fundamental value as it automates the prediction process. This means that once the computer has learned the best fitting line based on a data set, the model trained by the linear regression algorithm can then predict, for example, the X value (lets say the price of a home) based on Y (the home’s square footage). This is an extremely oversimplified example but image having millions of data points, and multiple dimensions or axis through which you have to fit not just a line but an n-dimensional hyperplane.

Why are Machine Algorithms so Important?

Consider if company A has algorithms (software) that are manually improved by software engineers, and company B has AI algorithms (software that can self-improve, that is, learns from data) at an increasing rate, company B wins.

In the next post, George will explain how Blockchain is as transformational as AI and how they differ and complement each other.