4 Key AI Concepts You Need to Understand

Here is a brief excerpt from an article by Bob Friday for InfoWorld magazine. To read the complete article, check out others, and obtain subscription information, please click here.

* * *

Artificial intelligence (AI) is taking the world by storm, with innovative use cases being applied across all industry segments. We are decades away from replacing a doctor with an AI robot, as seen in the movies, but AI is helping experts across all industries diagnose and solve problems faster, enabling consumers like myself to do amazing things, like find songs with a voice command.

Most people focus on the results of AI. For those of us who like to look under the hood, there are four foundational elements to understand: categorization, classification, machine learning, and collaborative filtering. These four pillars also represent steps in an analytical process.

Categorization involves creating metrics that are specific to the problem domain (e.g. finance, networking). Classification involves determining which data is most relevant to solving the problem. Machine learning involves anomaly detection, clustering, deep learning, and linear regression. Collaborative filtering involves looking for patterns across large data sets.

[Every artificial intelligence solution is built on these four foundations; here’s your quick guide to two.]


AI requires a lot of data that is relevant to the problem being solved. The first step to building an AI solution is creating what I call “design intent metrics,” which are used to categorize the problem. Whether users are trying to build a system that can play Jeopardy, help a doctor diagnose cancer, or help an IT administrator diagnose wireless problems, users need to define metrics that allow the problem to be broken into smaller pieces. In wireless networking, for example, key metrics are user connection time, throughput, coverage, and roaming. In cancer diagnosis, key metrics are white cell count, ethnic background, and X-ray scans.


Once users have the problem categorized into different areas, the next step is to have classifiers for each category that will point users in the direction of a meaningful conclusion. For example, when training an AI system to play Jeopardy, users must first classify a question as being literal in nature or a play on words, and then classify by time, person, thing, or place. In wireless networking, once users know the category of a problem (e.g. a pre- or post-connection problem), users need to start classifying what is causing the problem: association, authentication, dynamic host configuration protocol (DHCP), or other wireless, wired, and device factors.

* * *

Here is the direct link to the complete article.

Bob Friday is co-founder and CTO of Mist Systems.

Posted in

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.