Machine Learning still requires Common Sense

Stephen DeAngelis

July 10, 2018

Artificial Intelligence (AI), cognitive computing, and machine learning are receiving a lot of press coverage and have been subject to a lot of hype. Although there are differences between the terms, they are often used synonymously. Of the three terms, machine learning is the most widely discussed.[1] Nick Heath (@NickJHeath) explains, “From driving cars to translating speech, machine learning is driving an explosion in the capabilities of artificial intelligence — helping software make sense of the messy and unpredictable real world.”[2]

What is machine learning?

As noted above, some people equate machine learning with artificial intelligence. AI is actually an umbrella term under which machine learning falls. Cynthia Harvey explains, “All machine learning systems are AI systems, but not all AI systems have machine learning capabilities.”[3] With that bit of trivia out of the way, Harvey goes on to note there are four types of machine learning. They are:

  • Supervised Learning: “Supervised learning requires a programmer or teacher who offers examples of which inputs line up with which outputs.”
  • Unsupervised Learning: “Unsupervised learning requires the system to develop its own conclusions from a given data set.”
  • Semi-supervised Learning: “Semi-supervised learning, as you probably guessed, is a combination of supervised and unsupervised learning.”
  • Reinforcement Learning: “Reinforcement learning involves a system receiving feedback analogous to punishments and rewards.”

Dr. Marc Deisenroth (@mpd37), a lecturer at Imperial College London, adds, “Machine learning can be considered the engine of modern AI. It provides the underlying technology that drives AI. AI is about complex systems that behave intelligently. In order to reach this goal, AI poses many questions, and machine learning provides the technologies toward answering these questions. In other words, AI is about systems and questions whereas machine learning is about practical solutions to these challenges. Another difference is that AI strives for intelligence, whereas machine learning does not necessarily do this.”[4] So far in the discussion, one important word has been missing: algorithm. Richard Nass, Executive Vice-President of OpenSystems Media, explains, “It’s fair to say that the real key to accurate and useful machine learning consists of assembling the right combination of algorithms, compilers, and hardware architecture. If you don’t have the right components in any of those three areas, machine learning won’t work as it should.”[5]

Machine learning and common sense

Machine learning has no innate intelligence. Algorithms are simply instructions that tell a machine what to do with the data on which they train. Pete Warden (@petewarden), CTO of Jetpac, laments the fact that more attention is given to algorithms than to data. “Academic papers are almost entirely focused on new and improved models,” he writes, “with datasets usually chosen from a small set of public archives. Everyone I know who uses deep learning as part of an actual application spends most of their time worrying about the training data instead.”[6] Cory Doctorow (@doctorow) adds, “The problem is as old as data-processing itself: garbage in, garbage out. Assembling the large, well-labeled datasets needed to train machine learning systems is a tedious job (indeed, the whole point and promise of machine learning is to teach computers to do this work, which humans are generally not good at and do not enjoy). The shortcuts we take to produce datasets come with steep costs that are not well-understood by the industry.”[7]

In other words, data scientists need to use some common sense in selecting and preparing data on which machines train if they want common sense insights to emerge on the other end. Warden points out, for example, that one researcher searching for images of sunglasses on which a machine could train discovered the dataset included images of an archaic device for magnifying sunlight. When searching for images of a garbage truck, glamour shots were found in the dataset. And, when searching for images of cloaks, “a bias towards undead women” turned up. A little common sense goes a long way in preventing poor machine learning results. Doctorow adds, “It’s an important lesson for product design, but even more important when considering machine learning’s increasing role in adversarial uses like predictive policing, sentencing recommendations, parole decisions, lending decisions, hiring decisions, etc. These datasets are just as noisy and faulty and unfit for purpose as the datasets Warden cites, but their garbage out problem ruins peoples’ lives or gets them killed.”

Machine learning in the real world

People have their lives touched by machine learning without giving it much thought. Personal assistants like Siri and Alexa answer questions, set alarms, and much more. Movie recommendations on Netflix, music recommendations on Spotify, and friend recommendations on Facebook are all driven by machine learning. Nanette George (@NanetteGeorge), Senior Marketing Manager at CloudFactory, notes five areas our lives are going to be changed thanks to machine learning.[8] They are:

1. Autonomous Vehicles. “Self-driving vehicles could lead to a safer, cleaner, more efficient future for transportation. Software developers use ML and deep learning (DL) algorithms to power computer vision that allows the vehicle to make decisions in ways that are similar to human decision making.”

2. Education. “Teaching people how to write can be difficult to scale. Even for experienced high school teachers and college professors, it can be a challenge to review written work and provide meaningful feedback to every student in each of their classes.” That’s just one example of how machine learning can be used in education. It can be used in dozens of other ways in a modern classroom as edtech continues to improve across subject areas.

3. IoT and IIoT Predictive Maintenance. “Equipment maintenance is one of the many costly challenges facing companies that deploy fleets of machinery into the field. The Internet of Things (IoT) and Industrial Internet of Things (IIoT) use built-in sensors on everyday objects, from fuel gauges to tires, to gather data and share it across a network. The system uses ML to analyze data such as temperature and humidity to predict performance and future outcomes.”

4. Inbound Logistics Planning. “Logistics planning ensures the right person receives the right number of supplies at the right place at the right time. Inbound logistics focuses on the management of suppliers and the goods they send into a business. It’s a complex process of managing orders, shipping, warehousing, inventory control, and utilization. By gathering and feeding data on existing planning into an ML model, businesses can predict and recommend future processes.”

5. Retail Commerce. “Companies that sell products in a store or online have been collecting big data for some time now. They gather demographic data on consumers, their spending habits, and preferences. The challenge has been to aggregate online and offline data and to recognize patterns in the data that could positively influence pricing, inventory, customer experience, and profitability. Machine learning makes it possible for retailers to discover patterns in the data that they can act on to influence the customer’s experience with their brand.”

Those are only a few of the areas in which machine learning will change our lives in the years ahead. We need to make sure common sense is also a part of that future.

Footnotes
[1] A quick Google search brings up 348 million hits for machine learning, 304 million hits for artificial intelligence, and 20 million hits for cognitive computing.
[2] Nick Heath, “What is machine learning? Everything you need to know,” ZDNet, 14 May 2018.
[3] Cynthia Harvey, “What is Machine Learning?” Datamation, 3 January 2018.
[4] Colin Smith, “Machine learning research: the driving force of Artificial Intelligence,” Imperial College London, 4 October 2017.
[5] Richard Nass, “Machine learning starts with the algorithms,” Embedded Computing Design, 5 December 2017.
[6] Pete Warden, “Why you need to improve your training data, and how to do it,” Pete Warden’s Blog, 28 May 2018.
[7] Cory Doctorow, “Garbage In, Garbage Out: machine learning has not repealed the iron law of computer science,” Boing Boing, 29 May 2018.
[8] Nanette George, “5 Exciting Machine Learning Use Cases in Business,” IoT for All, 14 November 2017.