Shedding Some Light on Machine Learning

Stephen DeAngelis

February 12, 2020

“The hype machine is cranked up to an 11 on the topic of machine learning,” writes John McDonald (@jpmcdon), CEO of ClearObject.com.[1] McDonald goes on to note the term “machine learning” (ML) is often used synonymously with artificial intelligence (AI), but he objects to such usage because, he asserts, machine learning is neither intelligent nor artificial. He cautions, “Before you get swept away by the gust of hot air coming from the technology industry, it’s important to pause in order to put things into perspective. Maybe just explaining [machine learning] in reasonable terms will help.” It’s important to understand what machine learning is and is not because, as Kevin Casey (@kevinrcasey) writes, “Machine learning is already pervasive: Most people probably don’t realize it.”[2] Bill Brock, Vice President of engineering at Very, explained to Casey, “Whether or not you know it, odds are that machine learning powers applications that you use every day. Machine learning has revolutionized countless industries; it’s the underlying technology for many apps in your smartphone, from virtual assistants like Siri to predicting traffic patterns with Google Maps.” The bottom line is: Machine learning may be over-hyped, but it’s also proven itself extremely useful. Casey insists, “This is not pie-in-the-sky futurism but the stuff of tangible impact.”

A simple machine learning primer

In a nutshell, machine learning leverages algorithms to pore over large datasets looking for patterns. Mike Colagrossi observes, “The idea is to get the algorithm to learn or be trained to do something without being specifically hardcoded with a set of particular directions.”[3] He goes on to note, “Over the years, machine learning developed into a number of different methods.” Those methods are:

1. Supervised learning. “In a supervised setting, a computer program would be given labeled data and then be asked to assign a sorting parameter to them. This could be pictures of different animals and then it would guess and learn accordingly while it trained.”

2. Semi-supervised. “Semi-supervised would only label a few of the images. After that, the computer program would have to use its algorithm to figure out the unlabeled images by using its past data.”

3. Unsupervised. “Unsupervised machine learning doesn’t involve any preliminary labeled data. It would be thrown into the database and have to sort for itself different classes of animals. It could do this based on grouping similar objects together due to how they look and then creating rules on the similarities it finds along the way.”

4. Reinforcement. “Reinforcement learning is a little bit different than all of these subsets of machine learning. A great example would be the game of Chess. It knows a set amount of rules and bases its progress on the end result of either winning or losing.”

Alan J. Porter (@alanjporter), Head of Strategic Services for [A], observes some scientists aren’t enthralled with machine learning because they believe “ML is great at recognizing patterns but not much else” and “ML assumes tomorrow is going to be the same as today.”[4] Understand those basic premises and you can appreciate what machine learning can do without falling for the hyperbole.

Artificial intelligence, deep learning, and machine learning

Artificial intelligence is an umbrella term under which you can find machine learning and its companion deep learning. Deep learning is a specific type of machine learning. These relationships can create a bit of confusion. Ronald Schmelzer (@rschmelzer), Managing Partner and Principal Analyst at Cognilytica, explains why people object to using ML and AI synonymously. He writes, “It may take time and effort to train a computer to understand the difference between an image of a cat and an image of a horse or even between different species of dogs, but that doesn’t mean that the system can understand what it is looking at, learn from its own experiences, and make decisions based on that understanding.”[5] He goes on to note, “For the layperson, we want to stress that AI is not interchangeable for ML and certainly ML is not interchangeable with Deep Learning. But ML supports the goals of AI, and Deep Learning is one way to do certain aspects of ML. Or to put it another way, doing machine learning is necessary, but not sufficient, to achieve the goals of AI, and Deep Learning is an approach to doing ML that may not be sufficient for all ML needs.” He explains:

Over the past 60+ years there have been many approaches and attempts to get systems to learn to understand its surroundings and learn from its experiences. These approaches have included decision trees, association rules, artificial neural networks of which Deep Learning is one such approach, inductive logic, support vector machines, clustering, similarity and metric learning including nearest-neighbor approaches, Bayesian networks, reinforcement learning, genetic algorithms and related evolutionary computing approaches, rules-based machine learning, learning classifier systems, sparse dictionary approaches, and more.

Mark Tim writes, “Apart from machine learning, artificial intelligence is … completely wide and different in scope.”[6] He goes on to explain, “[AI] is not a system; rather AI is implemented in the system.” For years, artificial intelligence was divided into two categories: weak and strong. Strong AI was defined as artificial general intelligence and weak AI was everything else. As the artificial intelligence field has matured, those definitions have become outdated. As I see it, there are currently three levels of AI being developed. They are:

  • Weak AI: Wikipedia states: “Weak artificial intelligence (weak AI), also known as narrow AI, is artificial intelligence that is focused on one narrow task.” In other words, weak AI was developed to handle/manage a small and specific data set to answer a single question. Its perspective is singular, resulting in tunnel vision.
  • Strong AI: As noted above, strong AI originally referred to Artificial General Intelligence (i.e., a machine with consciousness, sentience and mind), “with the ability to apply intelligence to any problem, rather than just one specific problem.” Today, however, there are cognitive systems that fall short of AGI but far surpass weak AI. These systems were developed to handle/manage large and varied data sets to answer a multitude of questions in a variety of categories. This is the category into which cognitive computing falls. Cognitive AI can deal with ambiguous situations whereas weak AI cannot.
  • General AIThe AGI Society notes the ultimate goal of AGI researchers is to develop “thinking machines” (i.e., “general-purpose systems with intelligence comparable to that of the human mind”). The development of these potential thinking machines keeps some scientists and many science fiction writers up at night.

In the near-term, weak and strong AI systems will dominate conversations and provide the most benefit to businesses. Cognitive systems should prove particularly useful since they can function in ambiguous situations — the kind of situations in which individuals and companies often find themselves.

Concluding thoughts

McDonald observes, “If you plot the rise of computing cognitive power on a timeline scale starting with the beginning of civilization, it’s so fast in human history that it appears to be a vertical wall. We are only beginning to figure out the impact of this on humanity, and that’s what scares so many about the topic.” Terence Mills (@terence_mills), CEO of AI.io and Moonshot, concludes, “Both AI and ML can have valuable business applications. Determining which one is best for your company depends on what your needs are. These systems have many great applications to offer, but ML has gotten much more publicity lately, so many companies have focused on that source of solutions. However, AI can also be useful for many simpler applications that don’t require ongoing learning.”[7] Business executives need to know, however, that AI and ML capabilities are not an either/or situation. By leveraging cognitive computing platforms, like the Enterra Cognitive Core™ — a system that can Sense, Think, Act and Learn®, you can have both capabilities to use as needed.

Footnotes
[1] John McDonald, “It’s Time To Demystify Machine Learning,” Forbes, 14 June 2019.
[2] Kevin Casey, “How to explain machine learning in plain English,” The Enterprisers Project, 29 July 2019.
[3] Mike Colagrossi, “What’s the difference between A.I., machine learning, and robotics?” Big Think, 28 May 2018.
[4] Alan J. Porter, “Machine Learning Isn’t Rocket Science,” CMS Wire, 9 September 2019.
[5] Ronald Schmelzer, “Is Machine Learning Really AI?Forbes, 21 November 2019.
[6] Mark Tim, “Differences between AI and Machine Learning, and why it matters,” e27, 11 August 2019.
[7] Terence Mills, “Machine Learning Vs. Artificial Intelligence: How Are They Different?” Forbes, 11 July 2018.