What’s the first thing that comes to mind when you hear the following phrases?
- Artificial grass
- Artificial sweeteners
- Artificial flavors
- Artificial plants
- Artificial flowers
- Artificial diamonds and jewelry
- Artificial (fake) news
These phrases probably evoke thoughts such as “fake,” “not real,” or even “shabby.” Artificial is such a harsh adjective. The word “artificial” is defined as “imitation; simulated; sham” with synonyms such as fake, false, mock, counterfeit, bogus, phony and factitious.
The word “artificial” may not be the right term to use to describe “Artificial Intelligence,” because “artificial intelligence” is anything but fake, false, phony, or a sham. Maybe a better term is “Augmented Human Intelligence,” or a phrase that highlights both the importance of augmenting the human’s intelligence as well as to alleviate the fears that AI means humans become “meat popsicles” (quick, name that Bruce Willis movie reference!). And while I don’t expect this name change to stick (if it does, please give me some credit), I’m using this blog as an excuse to introduce some marvelous new training materials on artificial intelligence and machine learning.
But before I dive into details, let’s first frame the artificial intelligence conversation.
Focusing on the “How” Won’t Lead You to the “What” and “Why”
Organizations have access to a growing variety of internal and external data sources that might yield better predictors of business performance. And while having a process to ideate, validate and prioritize the different data sources that one might want to explore for its predictive capabilities, in the end the data by itself is of little value – organizations need to become more effective at leveraging data and analytics to power their business models (see Figure 1).
But in order to “monetize” that growing bounty of data, you’re going to need to become an expert at advanced analytics to tease out the customer, product, service, and operational insights that are the real sources of economic value (see University of San Francisco “Determining The Economic Value of Data” research paper). Business leaders need to become knowledgeable about advanced analytics capabilities so that they can envision “What” business use cases to target and “Why,” before they get pulled into the “How” discussion.
Preparing for the “How” Discussion
To help business leaders understand where and how to apply the different classes of advanced analytics (i.e., machine learning, neural networks, reinforcement learning, artificial intelligence), I’ve created an advanced analytics roadmap. I then mapped the advanced analytics roadmap against the Big Data Business Model Maturity Index (see Figure 2).
While certainly not perfect (and certainly not definitive given continued advanced analytics advancements), Figure 2 attempts to classify the different advanced analytics capabilities into a roadmap that organizations can use to help them understand when and where to apply the different advanced analytics capabilities. This is my attempt to try to summarize the advanced analytics confusion, hype and excitement into something actionable.
With that as my goal, here are the different levels of advanced analytics:
- Level 1: Insights and Foresight. This is the foundational level that includes statistical analytics as well as the broad categories of predictive analytics (e.g., clustering, classification, regression) and data mining. The goal of the level 1 is to quantify cause-and-effect, establish confidence levels and measure goodness of fit.
- Level 2: Optimized Human-decision Making. This level includes machine learning, deep learning and neural networks. The goal of these advanced analytic algorithms is to enable computers to learn on their own; to identify patterns in data, build models that explain the data, and predict outcomes without having pre-programmed rules and analytic models.
- Level 3: The Learning and Intelligent Enterprise. This level includes artificial intelligence, reinforcement learning and cognitive computing. These advanced analytic algorithms self-monitor, self-diagnose, self-adjust and self-learn. These analytics perceive the world around them, create goals, make decisions towards those goals, measure decision effectiveness, and learn in order to refine the decisions that advance towards the goals (maximize rewards while minimizing costs).
It is important to be able to summarize and present the wide realm of advanced analytics within a frame that we can explain to business leadership (because eventually we’re going to come to them for money). So using Figure 2 as our business framework, let’s deep dive into each of the advanced analytics levels.
Level 1: Insights and Foresights
The goal of Level 1 is to quantify “cause-and-effect” (i.e., quantify relationships in the data) and predict what is likely to happen at some measureable level of confidence. Level 1 sets the foundation for determining “goodness of fit,” or the extent to which observed data matches the values predicted by analytic models. Level 1 includes the following advanced analytic capabilities:
- Statistics is a branch of mathematics dealing with the collection, analysis, interpretation, presentation and organization of data. Statistical analytics and methods are used to support hypotheses (decisions) and provide credibility to modeling results and outcomes (via confidence levels and “goodness of fit” measures). Check out “Statistics for Dummies Cheat Sheet” for more information about different statistical techniques.
- Predictive Analytics and Data Mining include anomaly detection, clustering, classification, regression and association rule learning. Predictive analytics and data mining uncover statistically significant patterns in large data sets; they uncover relationships buried in the data in order to quantify risks and opportunities. Check out “23 Types of Regression” to see the different types of regression techniques available to the data scientist.
Level 2: Augmented Human Decision-making
Level 2 builds upon the predictions created in Level 1 in order to prescribe actions and recommendations. Level 2 is the domain of analytic capabilities focused on natural language processing (NLP), text translation, voice recognition, and photo/image/facial recognition. Advanced analytic capabilities in level 2 focus on learning and then making inferences from that learning. Level 2 includes the following analytic capabilities:
- Neural Networks and Deep Learning leverage a system of highly interconnected analytic layers to decompose complex data formats (e.g., images, audio, video) in order to learn about the data and create inferences about the data. For example, Figure 3 shows how a series of interconnected neural network layers work to identify a written number.
But beware, as there is not just one neural network technique, as can be seen in Figure 4.
Machine Learning empowers systems and applications with the ability to gain knowledge without being explicitly programmed. Machine learning focuses on the development of computer programs that can change when exposed to new data. Machine Learning algorithms identify patterns in observed data, build models that explain the world, and predict things without having to explicitly pre-program rules and analytic models (see Figure 5).
Fundamentally, Machine Learning does two things: 1) quantifies relationships in the data (quantify relationships from historical data and apply those relationships to new data sets), and 2) quantifies latent relationships (draw inferences) buried in the data.
There are two types of machine learning:
- Supervised machine learning is a type of machine learning algorithm used to draw inferences from data sets with label responses such as fraud, customer attrition, purchase transaction, part failure, social media engagement, or web click.
- Unsupervised machine learning is a type of machine learning algorithm used to draw inferences from data sets without labeled responses such as finding hidden (unknown) patterns, groupings or relationships in data.
See the blog “Top 10 Machine Learning Algorithms” for detailed list of machine learning algorithms.
- Adversarial Machine Learning is a fairly new area of machine learning. Adversarial Machine Learning sits at the intersection of machine learning and computer security. It seeks to enable the safe adoption of machine learning techniques in adversarial settings like spam filtering, malware detection and biometric recognition. Machine learning techniques were originally designed for stationary environments in which the training and test data are assumed to be generated from the same distribution. However in the presence of intelligent and adaptive adversaries, this working hypothesis is likely to be violated. For example, a malicious adversary can carefully manipulate the input data exploiting specific vulnerabilities of learning algorithms to compromise the whole system security.
- Finally, Ensemble machine learning combines several machine learning techniques into one predictive model in order to decrease variance, bias, or improve predict effectiveness. Ensemble methods can be divided into two groups:
- Sequential ensemble methods where the base learners are generated sequentially. The basic motivation of sequential methods is to exploit the dependence between the base learners. Weighing previously mislabeled examples with higher weight can boost the overall performance.
- Parallel ensemble methods where the base learners are generated in parallel (e.g. Random Forest). The basic motivation of parallel methods is to exploit independence between the base learners since the error can be reduced dramatically by averaging.
See the article “Ensemble Learning to Improve Machine Learning Results” for more details on ensemble machine learning.
Level 3: The Learning and Intelligent Enterprise
Level 3 focuses on creating an intelligent enterprise that can self-monitor, self-diagnose, self-correct and self-learn. Level 3 is the domain of continuous “learning and adjusting” advanced analytic techniques such as reinforcement learning, artificial intelligence and cognitive computing. Level 3 includes the following analytic capabilities:
- Reinforcement Learning focuses on how software agents take actions in an environment so as to maximize cumulative rewards while minimizing costs. Reinforcement learning uses trial-and-error to map situations to actions so as to maximize rewards. Actions may affect immediate rewards but actions may also affect subsequent or longer-term rewards, so the full extent of rewards must be considered when evaluating the reinforcement learning effectiveness. Reinforcement learning is used to address two general problems:
- Prediction: How much reward can be expected for every combination of possible future states
- Control: By moving through all possible combinations of the environment, find a combination of actions that maximizes reward and allows for optimal control
See “Transforming from Autonomous to Smart: Reinforcement Learning Basics” for more details on reinforcement learning.
- Artificial Intelligence is the ability for a computer system to acquire knowledge within a particular environment, apply the knowledge to successfully interact within that environment, and learn from the resulting interaction so that subsequent interactions get more effective, even to the point where an artificial intelligent application could re-program itself to more successfully perform (survive?) within a complex environment or situation (now that should scare the singularity folks).
Artificial intelligence involves the study of agents that perceive the world around them, form plans, and make decisions to achieve their goals. An intelligent agent is an autonomous entity that observes through sensors and acts upon an environment using actuators (i.e. it is an agent) and directs its activity towards achieving goals (i.e. it is “rational,” as defined in economics). There are 4 general types of intelligent agents:
- Simple reflex agents
- Model-based reflect agents
- Goal-based reflect agents
- Utility-based reflect agents
Cognitive Computing is a relatively new concept that is being championed by IBM Watson. Cognitive computing involves self-learning systems that simulate human thought processes and decision-making in complex situations. From Wikipedia, we get cognitive systems features including:
- Adaptive: may learn as information changes, and as goals and requirements evolve
- Interactive: may interact easily with users so that those users can define their needs comfortably
- Iterative: may aid in defining a problem by asking questions or finding additional source input if a problem statement is ambiguous or incomplete
- Contextual: may understand, identify, and extract contextual elements such as meaning, syntax, time, location, appropriate domain, regulations, user’s profile, process, task and goals
You can’t get to the “What” and the “Why” by focusing on the “How”
It is also important to understand the “How” in order to envision the “What” and “Why.” Sometimes the wide variety of advanced analytic techniques and algorithms cause confusion, and cause business leaders to slow down or even stop until they understand these advanced analytic capabilities better. The goal of this blog was to provide enough of an explanation of advanced analytics to business leaders so that when we get engaged in an envisioning exercise, we get turn off the governors that limit creative thinking.
Appendix: Marvelous Sources of Advanced Analytics Knowledge
There are many sources of excellent education available on advanced analytics, such as Andrew Ng’s deep learning classes on Coursera. One of my favorites is the content provided by the “Machine Learning for Humans” site. It has excellent material and includes a free downloadable e-book.
I’ll continue to share new sources of great educational material on advanced analytics as they get released into the wilds. Understand the “how” will help organizations to envision the realm of what’s possible. Many times, that envisioning is only limited by the organizations creativity and management commitment.
Figure 3: Why Convolutional Neural Networks
Figure 4: The Asimov Institute – The Neural Network Zoo
Figure 5: Nvidia – What’s the Difference Between Deep Learning Training and Inference?
 The technological singularity is the hypothesis that the invention of artificial super intelligence will abruptly trigger runaway technological growth, resulting in unfathomable changes to human civilization (a.k.a. Skynet).
Figure 6: Philosophy of Artificial Intelligence: Simple Reflex Agent