Can Machines Really Learn?

03/15/2015 09:08 am ET | Updated May 15, 2015

In the age of data science, machine learning and pattern matching are the building blocks of competitive advantage. In a perfect world, you just hire a bunch of data scientists, have them deploy clever algorithms and the machine will output a clear path to higher sales, better ROI and world peace. Sadly, that's not how it works.

A Checkered Past

Arthur Samuel was not a particularly good checkers player, but he was a groundbreaking computer scientist. In 1959, he began to teach a computer to play checkers, thinking that it was a good model for rudimentary problem solving. He defined machine learning as "a field of study that gives computers the ability to learn without being explicitly programmed."

Arthur Samuel playing his computer checkers game

We can benefit from this definition, but first we must define the verb "to learn." For our purpose, "to learn" is not cognitive; it is operational. Paraphrasing from Alan Turing's famous paper, "Computing Machinery and Intelligence," let's not ask the question, "Can machines think?" Let's ask, "Can machines perform the way we (who can think) do?"

Machine Learning & Pattern Matching

In practice, machines can really learn. Two of the most popular methods are machine learning and pattern matching. They are computer science techniques used to predict, categorize, analyze and even make guesses.

Main categories of machine learning tasks include the following:

    Supervised learning -- where you teach it

    Unsupervised learning -- where you let it learn by itself

    Reinforcement learning -- where it learns by trial and error

    Deep learning -- where it uses hierarchical or contextual techniques to learn

The goal of each of these tasks is to "teach" a computer program to apply generalized rules to data sets and yield useful results. How could these techniques be beneficial?

Practical Examples

    Recognition -- Speech to text, voice, face, fingerprint, parts, etc.

    Natural Language Processing -- Translation, sentiment analysis, etc.

    Recommendation -- If you like this, you may love this too.

    Diagnostics -- For medical or mechanical systems.

    Categorization -- For categorizing text, images, audio, spam filtering, etc.<

    Prediction -- What stock will go up or down? Who may be a terrorist?

    Analysis -- Share of basket, viewer profiling, fraud detection, etc.

    Yield Management -- Programmatic advertising auctions, airline seat pricing, etc.

    Autonomous Vehicles -- Cars, robots, drones, etc.

    And much, much more.

There are literally thousands of tasks you can accomplish using machine learning and pattern matching algorithms. You are limited only by your creativity, your quest for knowledge and the quality of the data you are working with.

Quality Data

All of data science is subject to GIGO (pronounced guy-go), which stands for Garbage In, Garbage Out. You can employ the finest data scientists in the world, but if they are working with bad data, you are going to get bad results.

1st Party vs. 3rd Party

Your data (first party), the data you generate or have collected, may be perfectly suited for analysis and offer rich opportunities. Other people's data (third party), the data you purchase or receive from other organizations, not so much. If you've ever received a postcard offering to extend the warranty of the leased car you returned two years ago, you know the value of most third-party data sets. A quick audit of the databases throughout your organization combined with a look at data you can trade for, or partner to obtain, will set the stage for meaningful outcomes.

Do-It-Yourself Machine learning Fun

If all of this inspires you to explore pattern matching and machine learning capabilities, Google has a Prediction API that is super easy to work with and is awesome fun.

"Given a set of data examples to train against, you can create applications that can perform the following tasks:

  • Given a user's past viewing habits, predict what other movies or products a user might like.
  • Categorize emails as spam or non-spam.
  • Analyze posted comments about your product to determine whether they have a positive or negative tone.
  • Guess how much a user might spend on a given day, given his spending history."

Want Help?

We have a team ready to help you get ready to work with your data, understand the opportunities afforded by machine learning and pattern matching and even do a data science readiness assessment. Just shoot me an email, and I'll be happy to work with you to help you achieve your business goals.

I'm the Managing Director of the Digital Media Group at Landmark|ShellyPalmer, a tech-focused investment banking and advisory firm. You may also know me as Fox 5 New York's on-air tech expert. Follow me @shellypalmer or visit for more info.