April 25, 2023
Bill Stoughton, VP, Data Science Lead
Business trends tend to elicit either a sense of, “I’m feeling left behind,” or, “There’s too much hype, I’m going to focus elsewhere.” I think these opinions capture a lot of the reaction to AI.
But why is it mystifying in the first place? Lots of technical terms mixed with jargon. And like any hot trend, the expansion of the term into areas it doesn’t belong. We’ll start by describing what AI is, and the three high-level categories that it falls into. The reality is, your company is almost certainly doing some of this. We will then define some of the common terms that are sometimes used interchangeably but shouldn’t be.
This article will demystify some of the rhetoric, bring clarity to what Artificial Intelligence is (and isn’t), and prepare a business leader to venture down the path “eyes open,” able to separate what’s real from the hype, and armed with a conceptual understanding of what AI could do for their business.
The term “artificial intelligence” was attached to the idea of computers mimicking human intelligence in 1956 at a conference at Dartmouth. The promise was always way ahead of practical applications until about 20 years ago, when separate paths that enabled significant AI traction started to converge.
· Massive computing power enabled through parallel processing
· Growth in the amount of data available and the ability to make its availability ubiquitous (the cloud)
· The open-source movement
Since terminology complicates the landscape of these discussions, it helps to differentiate what AI is, versus the techniques and platforms used to do it. From a practical sense, AI can be thought about in three broad categories:
· Narrow or “functional” AI – purpose specific, designed to solve a defined problem by predicting something, optimizing a scenario, recognizing specific patterns, etc. Things like a spam filter, fraud detection, predictive maintenance, and many others are examples of narrow AI. It is highly likely that your organization is already doing some of these things.
· Generative AI – a lot of attention to this lately with things like ChatGPT’s newest release, and the integration of generative capabilities into search engines. It uses a deep learning framework to understand the content of a user’s input and can generate relevant responses. There are lots of practical applications like interactive chat, content generation, search engine interfaces, or writing your kid’s term paper.
· General AI – the ability to perform any task that a human can. A robot at home that can make you a drink, debate the difference between the US and French revolution, and tell you how to fix the dishwasher. This does not exist … yet.
How do you ‘do’ AI? The landscape of terminology here can also be confusing. We are starting to hear terms like Machine Learning (ML) used interchangeably with AI, which are related things but not the same.
Generating business value from AI requires an ecosystem that defines objectives and includes an environment that enables the implementation of these solutions. In thinking through how to make all of this operational we need to parse out the terminology that defines it.
There are a set of terms that describe outcomes – like AI itself. This list is by no means exhaustive but those relevant in most business contexts are:
· AI - computer programs that mimic human intelligence
· Natural Language Processing (NLP) – a subfield of AI dealing with the understanding and generation of natural language
· Computer vision – another subfield of AI that uses algorithms to interpret and understand visual data
The terms used to describe outcomes need to be distinct from the description of approaches to get these things done. The primary collection of approaches to accomplish AI-focused outcomes is Machine Learning, but there are others like rules-based systems, expert systems, decision trees, and evolutionary (genetic) algorithms.
Deep Learning is an important subset of Machine Learning that involves multiple layers of neural networks to represent very complex data relationships.
Lastly, there are specific techniques that are deployed within, for example, the Machine Learning approach. Techniques such as regression, K-means clustering, Principle Components Analysis (PCA) can be used for basic analytics, or as part of a Machine Learning framework. More often, other techniques such as neural networks are utilized in Machine Learning (Neural networks are computer algorithms modeled on the human thought process).
The diagram below represents these three types of terminology and how they interact and overlap. Consider the space just outside of the AI circle as analytics basically defined.
Because machine learning plays such a dominant role in the AI discussion, especially for business applications, it’s worth spending a few moments on the 3 basic approaches. The term itself describes computer algorithms that can improve through experience. Machine Learning is a subset of AI (but not all AI is machine learning). The specific types of algorithms that can be involved with a machine learning solution vary and can include basic approaches like regression (linear, logistic, etc.) but more often involve more complex approaches like neural networks.
Regardless of the specific type of algorithm there are three broad approaches within machine learning, each with their own set of optimal use cases.
The basic context for this approach is usually the prediction of some outcome or classification of something with an objective truth. It is used when you know the answer and you provide the algorithm with lots of examples(training data). The algorithm then maps inputs to outputs based on all the data you provide it.
Does this image include a cat? Will this customer respond to my offer? Did our logo appear in any social media yesterday?
When faced with complexity, humans like to categorize things. It helps us to simplify and organize our thinking and is a natural cognitive process to deal with complexity. Unsupervised learning is an algorithm and approach to identifying undetected patterns and relationships in the data.
The output can be things like clusters or groupings. Practical applications include customer segmentation, market basket analyses, or grouping customer reviews into topics or themes.
Reinforcement learning algorithms are goal-oriented approaches that learn how to maximize some form of reward signal, learns from their own actions(and can “fail fast”).
Practical applications include optimizing email subject lines, product recommendations, or optimizing search strategies through targeting and bidding.
Not all analytics is AI, nor should it be. Analytics involves using statistical or mathematical methods to analyze data and gain insights. Regression analysis, predictive modeling, hypothesis testing, and data visualization are some of the traditional tools that can be used to generate analytic insights. They have been for years and will continue to be.
An analytic approach might become AI when it evolves to include the use of matching learning algorithms to analyze data and make predictions or decisions based on that data. Using predictive modeling and regression analysis to target likely responders would be considered an analytic approach. Using machine learning algorithms to make those same predictions based on customer behavior, demographics, and other real-time factors in an automatically self-learning environment would be considered an AI approach. The difference between the two can be subjective.
To solve a complex business problem, multiple AI-driven approaches can be combined to create a solution. For example, if the goal is to create an effective prospect segmentation strategy you might:
· Use supervised learning to find best potential customers based on likelihood to respond, potential tenure, and revenue potential. All in an environment that continually learns
· Use unsupervised learning to identify trends and natural groups. Use these groups to customize targeting and treatment
· Use reinforcement learning to test your messaging and media investment strategy for each of these groups
Discussions about complex topics can go in any direction, but they can sometimes start in the same place. Bringing clarity to the terminology used to describe a complex topic enables these discussions to be more productive. This is what we’ve tried to do with this article – establish a common language using real examples and as little jargon as possible. If this helps anyone move past, “I’m feeling left behind,” or, “There’s too much hype!", then it’s done its job.
To continue the conversation, reach out to us.
Blend360 co-creates data science solutions with clients to achieve their business goals. Clients are the heroes in our story, and we provide the tools and expertise needed to succeed. We use advanced analytics, machine learning, and AI to enable data-driven decisions that drive growth and innovation. As a leader in the data science industry, we collaborate with businesses of all sizes to create innovative solutions that make a difference.