A Digestible Overview
Until recently, the term Artificial Intelligence or AI was considered a concept from science fiction rather than one used in the business world. Yet today, we all see and interact with AI daily – whether it's your smartphone assistant, driving your car, or embedded in the software used by your company. However, notwithstanding how often the term is used and the buzz it generates, it's rare you come across a straightforward definition of the concept in business terms. Whether you talk to business leaders or data scientists, the term is more often shrouded in ambiguity or mystery rather than in tactical terms. However, I strongly believe AI can be defined and discussed in simple business terms. And when this happens, it enables business and data leaders to collaborate more effectively to create powerful new solutions to real-world problems.
The following is a brief overview of AI and the related concept of Machine Learning in what I hope are easy to digest terms. My goal is to help business leaders better understand the concepts so they can identify opportunities to apply them. I also hope this helps data leaders find new terminology to distill complex topics in an easy to digest way to foster greater collaboration with business partners.
Artificial Intelligence
What is AI?
AI has quickly become one of the most discussed technological developments of the past decade. Yet for how often the term is used, it's hard to find a straightforward definition of the concept. Though as we learned in school, sometimes the best place to start is the dictionary.
Merriam Webster defines AI as the following:
“An area of computer science that deals with giving machines the ability to seem like they have human intelligence.” (Merriam Webster)
I recognize this definition is a bit vague, but that's why I like it. This is because the reality is that there is not a generally accepted set of criteria or parameters that define AI. Rather, it is the ability of a computer or machine to mimic human-like intelligence that defines AI. To get a more specific definition, we have to go a bit deeper. We now focus on the two primary types of AI discussed in the field:
Specialized AI: This type of AI solves a specific, narrow problem (or set of problems). All the AI applications we see and interact with today are examples of Specialized AI.
Artificial General Intelligence (AGI): This is the hypothetical ability of a computer to learn, understand, or perform any intellectual task like a human. There is not an agreed upon set of criteria for defining AGI, however there are some proposed tests (such as the Turing Test proposed by Alan Turing)
Most experts agree AGI does not currently exist. However, there is significant debate on how far we are from creating it. Some will say we are a decade or two away, while others say we are nowhere near close to creating. That's a debate I'll leave for another post.
From here on out we'll focus on Specialized AI, which is most relevant to today's businesses and the majority of the AI-related work we see.
Types of Specialized AI Systems & Examples
Within Specialized AI, there are three types of AI systems. Odds are you've interacted with each of these, though it's often hard to know which one of these it might be at surface level.
The three types of Specialized AI systems are the following:
Rules-Based Systems: These systems are primarily driven by rules programmed by a human. They demonstrate intelligence by acting on the rules within the system. For example, when you ask the customer service prompt on the phone to connect you with a representative.
Knowledge-Based Systems: These systems leverage a significant knowledge base (e.g., database, knowledge graph, etc.). They demonstrate intelligence by using the knowledge to act or provide information. For example, when you enter your symptoms into a website and it indicates possible illnesses you may be suffering from.
Machine Learning Systems: These systems use computers' ability to learn patterns in data to develop rules, as opposed to requiring a human to program the rules. These systems demonstrate intelligence by acting on or uncovering insights from data provided to the algorithms used by the system. For example, an algorithm finding credit card transaction behaviors indicating fraud (as opposed to a human identifying and programming those transaction behaviors).
All three types of systems are used extensively today. In practice, many systems combine one or more of the three systems.
Below are some common examples of each type of AI system:
Rules-Based Systems
Self-Service Kiosks
Interactive Voice Response (IVR) used for automated customer support
Most video game characters or opponents
Knowledge-Based Systems
Illness Symptom Checker
Exam Preparation Software
Medical Diagnosis Coding Software
Machine Learning Systems
Natural Language Processing
Autonomous Vehicles
Many Prediction Models
Machine Learning
While the concept of Machine Learning has been around since the 1950s, it grew significantly starting in the late 2000s. This was largely driven by a rapid increase in computing power (and reduction in costs), especially as cloud computing took off. It was also supported by the digital transformation of many industries, which created an ever-expanding amount of data that could be fed into Machine Learning algorithms.
Today most organizations have individuals or whole departments focused on Machine Learning. Similar to AI, the term is used a lot but it's hard to find straightforward definitions that are helpful in business terms. This creates an aura of ambiguity – even an intimidation factor – around the concept. Yet, I believe there is ample opportunity to make Machine Learning approachable. Furthermore, there is significant opportunity and a need to further involve business stakeholders in creating Machine Learning solutions. I once heard a statistic that less than a third of Machine Learning solutions ultimately get used (I've also seen this in my experience). I believe the three major drivers of this are that:
The solution doesn't solve the problem in a sufficient manner
The inability to get adequate buy-in from business stakeholders
Challenges in effectively deploying the solution within the technology infrastructure
All three of these challenges can be mitigated if there is better understanding, alignment, and communication amongst business, data, and technology stakeholders.
Defining Machine Learning
The main difference between Machine Learning and other types of AI systems is that there are no rules programmed by a human within the system. Rather, Machine Learning uses the ability of computers to detect patterns in data to generate the rules or insights.
I find the following two definitions helpful to further explain this:
“[Machine Learning] is the field of study that gives computers the ability to learn without being explicitly programmed.” – Arthur Samuel, 1959
“…a set of methods that can automatically detect patterns in data, and then use the uncovered patterns to predict future data, or to perform other kinds of decision making under uncertainty…”
What is often confusing is that there is still a significant amount of computer programming required to build and maintain Machine Learning systems. However, the key distinction is that the programming is done to prepare the data, identify and establish the Machine Learning algorithm that will be used, and/or to implement the solution. In a pure Machine Learning solution there is, however, no programming of any specific rules to drive action or generate insights – i.e., demonstrate intelligence – as is the case with Rules-Based AI Systems.
Why and When is Machine Learning Used?
The main benefit of Machine Learning is that it does not require explicit programming by a human. It is often used when programming by a human is costly or not feasible. This could be due to the complexity of the problem, a lack of detailed domain expertise, or a volume of data that cannot feasibly be analyzed by other means. This makes Machine Learning a highly scalable solution.
For example, if the volume of data increases or the amount of rules required increase due to added business complexity, you don't necessarily need to scale up your engineering team. Similarly, if you aim to tackle problems in a new domain, you may not necessarily need to hire an additional domain expert to build the model (that's not to say domain expertise isn't required).
Below are some examples of why and when Machine Learning might be used:
Common Machine Learning Problem Types & Examples
How does Machine Learning Work?
Machine Learning is a data-driven process. It relies on a set of defined input variables and, in some cases, a set of defined output variables. This is usually represented in matrix form (i.e., The Machine Learning Matrix).
The Machine Learning Matrix
The data is provided to a Machine Learning algorithm, which typically seeks to satisfy a given objective (e.g., predict the output variable with sufficient accuracy, group similar data points together, etc.). This is how the Machine Learning System learns. It can then apply what it learned to new data to make predictions, generate insights, or take actions.
Types of Machine Learning Methods
There are various methods available based on the problem and data. In practice multiple methods are often applied together to create a solution. The two most common methods are Supervised Learning, which is characterized by a Response Variable or Label, and Unsupervised Learning, which does not include a Response Variable.
The table below summarizes the various Machine Learning methods:
Machine Learning Methods
Why is this important?
AI offers nearly unlimited potential and transformative powers for any business or product. This is why so many organizations have invested in the discipline – both in terms of technology and people. Yet the reality is many organizations question their investments and whether they're realizing the returns the sought. What this leads to is a small number of haves and many have-nots when it comes to organizations investing in AI.
I believe those leading the way are the organizations and individuals who can successfully bring together the business, data, and technology domains to build needed solutions and create a differentiated product or competitive advantage. Once they do so it often becomes hard for others to catch up (because the data and learnings generated from the Machine Learning solution(s) becomes an asset itself). This is why I think it's so important to distill these topics into easy to digest terms that cut through the ambiguity, and allow individuals from all three domains to work together more effectively. I hope this post helped you in better understanding these concepts and/or giving you new terminology to help others to do so!
If you you liked this article and want to learn more check us out at www.thedatalinguist.com. We provides courses, coaching, and content to help you become fluent in Business, Data, and Tech to help you realize your potential and maximize your impact in any data related role!
References
Géron, A. (2019). Hands-on Machine Learning with Scikit-Learn, Keras & TensorFlow (2nd Edition).
Murphy, K. P. (2012). Machine Learning a probabilistic perspective.
Comments