Machine teaching is the emerging practice of infusing context -- and often business consequences -- into the selection of training data used in artificial intelligence (AI) machine learning so that the most relevant outputs are produced by the machine learning algorithms.
Proponents of machine teaching -- most vocally Microsoft -- hope to make the practice easily replicable, so that those without a background in computer science or software engineering can use machine teaching in new contexts. A Machine Teaching Group formed within Microsoft espouses the notion that: "[B]y separating the teaching information from the algorithm, we can allow the algorithms and the teaching language to innovate independently and the teacher doesn't need to understand machine learning algorithms." The idea is to enable business users to take machine teaching tools and apply them to problems specific to their industry sectors. So, for example, lawyers, nurses, city planners or other subject matter experts can impart important abstract concepts to an intelligent system that can perform machine learning.
Even so, the implication is that machine teaching requires a blend of human and artificial intelligence. By including human teachers in the training process, generating results through machine learning becomes faster and less expensive than using data alone.
To understand machine teaching, though, it is useful to first understand the concept of machine learning. Machine learning, a subset of artificial intelligence, allows software applications to become more accurate at predicting outcomes without being explicitly programmed to do so. Machine learning algorithms use historical data as input to predict new output values.
Machine teaching vs. machine learning
The machine teaching discipline is centered on creating the optimal training set that can guide a learning algorithm with the most efficiency. For its part, the goal of machine learning is to use those training data sets to allow autonomous systems to learn and improve their skills without being explicitly programmed to do so.
Machine learning, a term popularized in 1959 by Arthur Lee Samuel, is the use of computer algorithms and statistical models to perform specific tasks without explicit instructions. Although it is still considered by many to be an emerging science, machine learning systems are widely used in such settings as email filtering and internet search page recommendations.
The machine learning process
Machine learning uses algorithms to build mathematical models that can look for patterns in data to make decisions without further human intervention. These mathematical models are based on sample data, generally known as training data. The more the data, the better the decisions.
The process of learning begins with observations or data, such as examples, direct experience or instruction, in order to look for patterns in data and make better decisions in the future based on the examples that they are provided. The primary aim is to allow the computers to learn automatically without human intervention or assistance and adjust actions accordingly. Machine learning models are generally divided into two types of machine learning algorithms: supervised and unsupervised.
Supervised learning algorithms apply past experiences to new data using examples that are labeled (known simply as labeled data), to predict future outcomes. The learning algorithm can compare its output with the intended output, find errors and make the needed adjustments. The teaching process continues until the machine can reliably make predictions with an acceptable degree of accuracy.
By contrast, unsupervised algorithms use training data that is not labeled. Instead, unsupervised algorithms look to find structure in the data. This is a particularly useful skill in research and science when looking for hidden patterns.
A third type of learning algorithm, known as reinforcement learning, uses a reward feedback signal to teach machines and their software agents to choose the ideal behavior.
Benefits of machine teaching
The chief benefit of machine teaching is that it put automation tools into the hands of ordinary users with no computer science background -- that is, subject experts. The goal is to make machine teaching tools as easy to use as word processing software or computer spreadsheets, wherein writers and accountants don't need to know computer programming to make the best use of these software tools. If that goal is achieved, then presumably it would free up computer scientists to tackle more creative tasks, rather than the more monotonous tasks of creating training sets.
Applications of machine teaching
Machine teaching is now being tested in a variety of applications, notably in industry. In 2017, Siemens subject matter experts, using Bonsai's platform (since acquired by Microsoft), trained an AI model to autocalibrate a computer numerical control (CNC) machine more than 30 times faster than an expert human operator. CNC machines need to be recalibrated frequently, as even minor friction leads to errors that result in costly manufacturing imperfections.
Other applications being tested include keeping carbon dioxide levels safe in buildings with large, automated heating, ventilation, and air conditioning (HVAC) systems; supply chain management; healthcare operations; and transportation logistics.