An Introduction on Artificial Intelligence and Machine Learning

This article is written for the reader who would like a basic understanding of Artificial Intelligence (AI) and its subset – Machine Learning. The author also introduces Deep Learning, which is an evolutionary step of machine learning. Keeping the article at a high-level, the author provides the novice examples of AI applications and references for detail reading. The author’s goal is to give the reader an awareness of the subject and spur interest without getting knee deep into the math and intricacies of programming code.

Back in the mid to late 50’s, Jon McCarthy and Arthur Samuel coined the terms Artificial Intelligence and Machine Learning respectively. Artificial Intelligence refers to the field of computer science that develops software that takes data and produces solutions to stated problems. Machine learning is a subset of AI. It is the use of a data trained model to produce decisions, like moves in a chess game, or predictions, like who will win the Kentucky Derby.

Applications of AI is nothing new. I remember in the 60’s watching Tom Landry, the coach of the Dallas Cowboy football team, use statistics from past games in calling plays off-the-field to the dismay of the quarterbacks. However, these initial applications were limited by data storage and computer speeds. But with advancements in big data[¹]and computer power, artificial intelligence has now cracked open the crystal ball of fortune telling into the “wow” category (data analytics replaces luck as the enabler).

Sticking to sports for the moment, watch the movie Money Ball on how the Oakland Athletics used statistics to recruit overlooked players for a winning team. Better yet, follow the Houston Astros on how they play baseball by using “the shift” to place players on the field to catch predicted hit locations; they do this more than any major league team in baseball. Note, these are no longer off-line recruiting predictions, but on-line game decisions; imagine, if a batter at home plate could predict the next pitch with high probability. This is happening in the field of medicine, retail sales, self-driving vehicles, global positioning systems, investing, weather, facial recognition, even personal relationships.

Artificial intelligence is an umbrella term covering the areas of Machine Learning (ML) and its subset Deep Learning (DL). Let’s review ML and its approaches:

• Supervised Learning is where an analyst teaches a model (algorithm) on how to respond to a set of data in terms of a decision or a predication. The analyst typically begins with a predefined historical database that its contents are understood and the problem solution is known. Essentially, the data is structured under the supervision of an analyst. The algorithm is taught through an iteration of tests to identify known patterns which leads to a solution.

An example is weather forecasting. By using regression analysis, the model compares the known historical weather patterns to current weather conditions. If conditions match the known data patterns, the model provides a weather prediction. The results of several hurricane models[2]look like a bowl of spaghetti paths on television. Note the historical database is constantly updated with new data – match or no match. Thus, a new model is born.

• Unsupervised Learning is when the analyst does not structure the data due to the amount of data and variables involved. In these cases, an algorithm groups features in clusters to manage the data (i.e., essentially making the data supervised) through an iteration of tests without human intervention. One of the benefits of the unsupervised approach is lead time. It would take too much time to structure all data sources.

For example, in healthcare, large amounts of symptom data are analyzed by an algorithm into patterns to help identify diseases; thus possible treatments. As stated above, time to structure all data sources associated with various diseases by an analyst would be extensive.

• Reinforcement Learning is when the algorithm learns via feedback (i.e., trial and error) rather than training as in supervised learning. Its learning is based on successful results that reinforces previous action taken.

An example is robotics. Perhaps you’re thinking of “Data”, the humanoid in Star Trek, who advises us of the percentages of successful actions. No we are not there, yet. But, we do now have self-driving vehicles. These vehicles determine its next action continuously based on the previous results. It’s like training a pet; based on a reward (treat), the dog repeats an action, like fetching a ball. Unless you own a cat – you can’t train a cat, at least not ours – well, maybe it takes a deeper training.

Deep Learningis a bit more complicated in that it incorporates a hierarchical layer of algorithms or filters and learns from each layer in an iterative manner, like a neural network of a human brain. It is useful in recognizing patterns from unstructured data and often used for image and speech recognition applications.

An example is facial recognition. A database incorporates image data of known criminals and matches them with images collected by surveillance cameras. When a crime happens, the images from the cameras provide pictures of the faces of those involved. These images can then be matched against the database. With a match, law enforcement can then pursue the appropriate suspects.

Table 1. Machine Learning Approach Characteristics

Supervision Learning Process

Let’s review Machine Learning and document a high-level process for implementing it. Note that it is an iterative learning process that only ends when the analyst retires the model:

Figure 1. Supervision Learning Process

After some research, I was taken back with so many machine learning applications already in use: weather forecasting, medical diagnoses, law enforcement, and self-driving vehicles. Also, I did not realized that it was the advancements of big data and faster computing that allowed the break-thru of AI in our daily lives. Most of us, I believe, think that artificial intelligence is still science fiction. Not so! We as business analysts need to pursue AI education and recognize the many business opportunities opening up to all of us.

Author:Mr. Monteleone holds a B.S. in physics and an M.S. in computing science from Texas A&M University. He is certified as a Project Management Professional (PMP®) by the Project Management Institute (PMI®), a Certified Business Analysis Professional (CBAP®) by the International Institute of Business Analysis (IIBA®), a Certified ScrumMaster (CSM) and Certified Scrum Product Owner (CSPO) by the Scrum Alliance. He holds an Advanced Master's Certificate in Project Management and a Business Analyst Certification (CBA®) from George Washington University School of Business. Mark is also a member of the Association for the Advancement of Cost Engineering (AACE) and the International Association of Facilitators (IAF).

References/footnotes:

1. Big Data: a data source that provides large amounts of trusted data (text, video, audio) at a high speed.
2. The four best hurricane forecast models — ECMWF, GFDL, GFS, and UKMET (see Internet for expansion of acronyms). They take several hours to run on the world's most advanced supercomputers.

• Overfitting sometimes happens when the model does not recognize matches due to an enlarged database. Analysts can prevent this by testing the model against unknown data.

•  YouTube AI and Machine Learning videos and articles on the Internet.

• For example Edureka and Intellipaat training videos
• https://machinelearningmastery.com/a-tour-of-machine-learning-algorithms/

• Many AL and Machine Learning books available on Amazon.
  • Python Machine Learning: Programming and deep learning for beginners the crash course for python programming, neural networks, artificial intelligence and data science by Leonard Smith | Jul 31, 2019

• Vendor books and white papers
  • IBM has issued Machine Learning for Dummies