COMPETITIVE ADVANTAGE FROM THE FOUR WAVES OF DIGITAL (PART 2)
Ten Human Abilities and Four types of Intelligence to Exploit Human-Centered AI
Getting started on your Automation, Analytics, and AI strategy
In the first part of the Digital Revolution series, I looked at how data, automation, analytics, and AI are four inseparable parts of the revolution — the unbeatable quartet. While all of us can enjoy the music of the unbeatable quartet, most of us would be challenged to create one. Many of us would not know where to start.
The situation is not too dissimilar when it comes to building your competitive advantage using data, automation, analytics and AI. We have seen organizations that have developed their big data strategy, created a data lake and are now searching for some good business use cases. There are others who have focused on RPA, found some good early successes, but are now struggling with a ‘bot’ management problem. Some others have siloed analytics and automation Centers of Excellence that are competing to expand into AI. We have also seen AI strategies at large enterprises that speculate with no grounding on what data is available or what has been done to date in more traditional analytics.
Some companies don’t care much for such strategies and launch straight into action. Business executives want to start from the business use cases within a specific business unit, functional area, or a shared service area. IT organizations view the quartet as just extensions of software technology. Analytics and AI are just additional functionality to be implemented in software (see my series on Data Scientists are from Mars and Software Developers are from Venus for the dangers of this view). Data Science and analytics organizations tend to focus on specific techniques or algorithms — machine learning, deep learning, computer vision, NLP etc.
Business use cases, technological functionalities, and analytics/AI techniques are not the unifying paradigms for building lasting strategies. We believe a fundamental understanding of human abilities and how we plan to automate or augment human abilities is the appropriate way to approach the quartet.
Use Cases
Business Lens
The most popular mechanism to develop an automation, analytics or AI strategy are business use cases. These use cases are specific to an industry sector and also to functional areas. The use cases span all industry sectors from primary industries like agriculture, forestry, fishing, oil and gas, mining etc. to secondary industries like manufacturing, energy, chemicals, construction, food etc., to tertiary industries like financial services, healthcare, travel and transport, logistics, media, telecommunications etc. In addition, the use cases also span the different elements of the value chain from research and development, product development, marketing, sales, customer service, operations, finance, HR, etc., There are literally hundreds, if not thousands of these business use cases that have been mapped, delivered, and their ROIs realized in the past 3–4 years.
Technology Lens
While business executives naturally start from the business use cases — industry sector and functional area, technology executives start from the IT stack. They view the data, analytics, automation, AI strategy primarily in terms of building the technology capability around data warehouses, data lakes, streaming data for the data side; cloud storage and compute, microservices architecture, technology vendors for robotic and intelligent process automation, and cloud-based AI/ML solutions for the analytics and AI areas. The technology group should help ensure that while the business use cases are delivered, they are also enabling consistency with an existing technology architecture or enhancing the architecture as new capabilities need to be developed and deployed. For example, a technology architecture that enables not only continuous integration and continuous delivery (CI/CD) from a software perspective, but also continuous learning (CL) from a model perspective. Similarly, facilitating the convergence of Internet of Things (IoT), 5G, and AI (e.g., smart sensors or federated learning) can lead to enhancements of the existing IT stack.
Technique Lens
Business and IT executives have learnt to co-exist and build bridges between two entirely different mindsets, capabilities, qualifications, and issues. Now, we have a third group of executives — data scientists, analytics, and AI professionals who bring a different lens to the use cases. The primary concern for this group is what techniques can be used to extract insights from the data to make better decisions or what techniques can be used to automate tasks. This group views the use cases in terms of the type of technique(s) that can be used e.g., machine learning, deep learning, NLP, computer vision, process mining, etc. For example, different analytics and AI techniques can be mapped to industry sectors or to functional areas.
Although all these lenses are useful to develop unique techniques (i.e., technique lens) that are scalable (i.e., technology lens) and deliver business results or ROI (i.e., business lens) they fall short in one critical dimension. They fall short in not addressing the fundamental questions — what tasks and decisions humans are performing and how can we make them more efficient and effective. This requires us to turn the clock back by almost 35+ years to examine the taxonomy of human abilities.
Ten Human Abilities
In 1984, Fleishman and Quaintance in their book on Taxonomies of Human Performance: The Description of Human Tasks documented 52 distinct human abilities. These human abilities covered the broad spectrum of perceptual, cognitive, and motor abilities of humans. For example, under perceptual abilities they consider near vision, far vision, night vision, perceptual speed, visual color discrimination etc. Under cognitive abilities they consider deductive reasoning, inductive reasoning, mathematical reasoning, oral/written comprehension, oral/written expression etc. Under motor abilities they consider manual dexterity, reaction time, control precision, and multiple coordination.
Why are we interested in these human abilities when we are talking about data, automation, analytics, and AI? As we saw in our earlier post, we defined AI in terms of sense, think, and act that correspond to the perceptual, cognitive, and motor abilities of humans. We believe that starting from human abilities that we want to automate, assist, and augment we can provide a coherent framework for integrating the three different lenses we saw earlier. Rather than start from the fifty-two human abilities we examine a small grouping of ten human abilities.
- Collecting, Organizing, and Summarizing: Ability to collect, organize and summarize data that is structured or unstructured — text, audio, image, and video.
- Sensing, Querying, and Conversing: Ability to sense, search, query structured databases, text, audio, image, and video data. Ability to communicate and converse with the machine in natural language.
- Recognizing, Identifying, and Classifying: Ability to recognize, identify, classify, understand, and visualize structured data, natural language, audio, image and video data.
- Trending, Forecasting, and Predicting: Ability to identify trends from past data, forecast the future from the past data, predict and project across multiple scenarios.
- Diagnosing, Deciding, and Recommending: Ability to diagnose root causes, make decisions supported by reasoning, and making recommendations based on historical patterns.
- Reasoning — Deductive, Inductive, and Abductive: Ability to follow a logical process to arrive at a conclusion based on deduction, induction or abduction.
- Learning — Supervised, Unsupervised, and Reinforcement: Ability to improve performance over time or with additional data or environmental input.
- Optimizing — Continuous vs Discrete, Unconstrained vs Constrained, Deterministic vs Stochastic, and None, One, or Many Objectives: Ability to find the ‘best’ solution given a solution space and a number of constraints.
- Simulating, imagining, and adapting: Ability to simulate scenarios, hypothesize new ones, act autonomously, learn from data and over time adapt to changing environmental conditions.
- Discovering, creating, and inventing: Ability to discover new hypothesis or facts, generate or create new artifacts, such as answers to questions, poetry, original art or music.
This list is by no means a comprehensive list of human abilities like what Fleishman and Quaintance have provided. We have chosen not to include some of the motor abilities that will be more relevant for robotics; similarly, we have not included capabilities that current generation computers are already good at, such as, number crunching, documenting, or remembering. This list reflects some of the key human abilities we have seen the use cases address. It captures the majority of the business use cases and also many of the techniques used today in analytics and AI.
Four Intelligence Types
The rationale for looking at the human abilities is to develop a human-centered approach on how we can replace, supplement, or enhance these human abilities with advanced analytics and AI. When we look at a human-centered approach to AI, we have found it useful to look at two distinct dimensions — (a) how humans interact with AI and (b) how the AI is interacting with the environment.
Along the x-axis we have two extremes, where at one end the AI is independently taking actions and making decisions; and at the other end we have a human-in-the-loop system where the human is ultimately responsible for the decisions and actions, but is using the AI to inform his or her decisions and actions.
Along the y-axis, we have one extreme where the AI is fixed and is not adapting to the environmental changes; and at the other end we have an adaptive system that changes with the environment.
When we combine these two axes, we have four distinct ways by which AI is being used today and has been used in the past:
- Automated Intelligence: Manual or cognitive tasks that are repetitive or non-repetitive when sufficiently simplified and standardized can be automated. These are instances where the tasks are sufficiently straightforward that we do not want any human involvement e.g., copying text from one screen to another, and also tasks that are relatively static and will not change for months or even years and there is a large amount of time that is currently spent on such tasks within the firm.
- Assisted Intelligence: There are a number of actions and/or decisions that require human judgement. While the AI algorithm can do the number crunching, finding patterns, predictions, etc., we still require humans to take the final decision or act based on the recommendations. This is especially true for decisions or actions that involve our health, finances, and other critical emergencies. This type of assisted intelligence is not new — since the dawn of the digital revolution and the advent of computers. We have used computers in this manner. Of course, some would challenge the notion that number crunching and doing calculations is not AI. But in the use of technology we are in the same continuum with AI being the most advanced type of intelligence. Another key aspect of assisted intelligence is that humans learn from the machines and get better at using their judgement. However, any change in the recommendations of the AI requires re-evaluation and re-programming. Hence, we say that they are fixed in their interaction with the environment and not adaptive.
- Augmented Intelligence: The more recent wave of advanced analytics and AI have focused on machine learning. As humans make judgements, often inconsistently and frequently irrationally, algorithms can learn about human judgements and factor them into its own recommendations. For example, a loan officer with a decade long experience may use his judgement (based on his past experience) very differently to a loan officer who has less than a year experience. Machine learning systems can learn from the experienced loan officers to adapt their recommendations. As these systems adapt to their interaction with the humans and also to changing circumstances (e.g., during periods of recession there is likely to be an increase in defaults) they are adaptive. We say that they are augmenting the decision making of humans.
- Autonomous Intelligence: Finally, there are some situations where we want the AI to be adaptive to the environment and also operate without any involvement of humans. We categorize these types of uses as autonomous intelligence. A great example of autonomous intelligence is the quest of autonomous cars. We want the cars to drive themselves under any condition — heavy snow or rain or traffic conditions — with no intervention from humans. This falls in the quadrant of adaptive systems with no human-in-the-loop.
As we go through these four types of intelligences — from automated to assisted to augmented to autonomous, we require progressively more scrutiny, governance, and oversight. The risks increase significantly as we move across these four types. The jump from augmented intelligence to autonomous intelligence requires both a technical leap and also social acceptance.
The ten human skills we saw earlier can be replaced (in automated and autonomous intelligence) or enhanced (in assisted and augmented intelligence) by advanced analytics and AI. Let’s take a very simple skill like organizing. AI can help you today in organizing your email into standard folders (e.g., primary, social, updates etc.) automatically. It can assist you by performing complex searches, extracting the thread of conversations, etc. to aid your actions. More sophisticated email responders can draft replies to emails based on your past conversations and present it to you before sending. This would be augmented intelligence. Personally, I wouldn’t trust these email responders yet to do it autonomously without my intervention. So, we really haven’t reached the stage of autonomous intelligence in this case. The standard ‘out-of-office’ email responders are static and can fall more under automated intelligence.
We can take each of the human skills that we saw in the previous section and cycle through all four types of intelligence. It is this fact that makes them a powerful tool for us to use in developing our strategy.
Bringing it all together: Starting your journey
Most organizations start their data, automation, analytics, or AI journey by developing use cases with a business, technology, or technique lens. Given that what we are solving for is the ability to replace, assist, or augment human skills, we believe that starting from the ten human skills and four types of intelligences might be better. Let’s illustrate this with an analogy.
Let’s say you have an eight-year old daughter and you want her to be good in multiple areas — be good academically, be good in music, be a ballet dancer, and do sports. The way you would help her develop in these areas is not by making her focus just on one, e.g., ballet or sports or music. You would have her practice multiple skills over a period of time and have her progressively gain the expertise. It is highly unlikely that you would want her to practice 24x7 on ballet and when she has mastered it have her move on to music and then sports and so on.
When we select business use cases purely based on the functional area — we are often trying to just excel in one area and are not fully leveraging the investment across multiple functional areas and business use cases. It is like having your daughter be the best ballerina before you have her start playing tennis!
Similarly, if we approach it primarily from the technology lens, we build the capabilities without seeing the benefits manifest themselves in the business. It is similar to having your daughter master her motor skills in terms of stretching before you put her through her first ballet lesson. While stretching is important for ballet it needs to be done together. We often see organizations build their data warehouses or data lakes and then start worrying about finding the right business use cases.
If we approach the use cases from a technique lens — let’s say deep learning- we often end up developing the ‘best hammer in the world’ constantly treating everything as a nail to hit on. The organization develops interesting techniques and becomes well known academically for their contributions to the academic discipline, but that does not translate into revenues for the firm. Of course, if the mandate for the group is to excel academically it would be the right strategy. This would be akin to your daughter building the strength and stamina for a rigorous exercise. While this would be extremely useful for a long tennis match — just having the strength and stamina alone will not guarantee that you will win.
In summary, one of the best ways to start your journey is by looking at the ten skills and the four types of intelligences within the business, technology, and technique lenses. Starting with a human-skill-intelligence centered approach can help us get clarity on when and how to bring data, automation, analytics and AI across the different business/functional areas, use the appropriate technologies and techniques and deliver on the outcomes. The outcomes invariably are increasing efficiency/productivity improvements from automation and AI; and better effectiveness of decisions and actions from advanced analytics and AI.
