Decision Intelligence Frameworks — OODA Loop vs SEAL™ by diwo
“The straight line, a respectable optical illusion which ruins many a man.”
― Victor Hugo, Les Misérables
All of us, without exception, would love to tame the future and tilt odds in our favor all the time. This lofty aspiration and our incessant effort toward making right choices is what sets us apart from animal species. Decision making is a basic cognitive process of human behavior. Far from perfect, our prowess for decision making stands exposed more than ever in the hyper connected and rapidly changing world we live in.
Let us admit it. We need help. Human cognition alone cannot cope with the onslaught of daily information, sort through myriad of available choices, and make effective decisions day in day out. Can judicious use of AI and automation circumvent our cognitive limitations? The fast emerging field of decision intelligence is squarely focused on achieving just that.
Decision Intelligence: How can it help?
There is hardly any organization today that does not aspire to power their decisions and actions with intelligence. Enough evidence exists that those who are able to harness their organizational intelligence and align it to desired outcomes accrue a substantial advantage.
Over the decades of automation, significant strides have been made in codifying human experiential knowledge as well as extracting hidden knowledge from transactional data footprints. Focus of decision support automation has progressed from simple task efficiencies to enhancing decisions effectiveness and influencing outcomes.
The purview of decision intelligence is to explore outcome focused and human in the loop approaches to decision automation. A decision intelligence system is a man-machine collaborative system designed to enable and mature decision-action capability in an organization.
A decision intelligence system plays a dual role:
1. Operate as a non-intimidating intervention system to nudge and influence human decision-making behavior, and
2. Reduce cognitive burden of human decision-makers by assisting in quick exploration and evaluation of alternatives
So, how does one go about designing a decision-intelligence system?
Decision Intelligence Framework
The methodical design of a decision intelligence system must hinge around a framework — a mental model of decision making. The framework not only helps in defining scope and boundary of the system but determines the extent to which the system can be useful and in what situations.
Many frame decision intelligence agenda as “going from data to decisions” or “getting actionable insights from data”. Such articulations are too high level and vague to guide the design of a decision intelligence system. Moreover, by putting data at the center stage, these nudge us inadvertently toward collating all data as the necessary first step.
During the recent decade, we saw a proliferation of data lake or data hub technologies. In spite of substantial innovations in dealing with three V’s of big data (Volume, Variety, and Velocity), we have yet to see any noticeable impact on the decision-action capability of organizations. That does not imply that handling and managing data is unimportant for decision intelligence, however, we can safely conclude that it need not be the first step and some crucial piece is missing in crafting a decision intelligence system.
A well-designed decision intelligence system is less dependent on data as one might think, as it can help make effective decisions even with limited data and can tolerate errors and inconsistencies as well as deal with high degrees of uncertainty.
Decision making being a cognitive function, we need a deeper understanding of it, so that we can better augment and support it by intelligent automation.
Consider for example, the Covid-19 pandemic situation where a business is faced with the decision about when to revoke “work-from-home” directive to its employees so that business can be conducted as usual. Such a decision has a larger context but, for simplification, let us assume that state and federal guidelines have been relaxed and now it’s up to the individual businesses to make that call. Best action will no doubt depend on the type and size of the business but a host of other familiar and unfamiliar factors weigh in on the decision.
For a productivity focused business leader with a strong proclivity toward office attendance, the decision is trivial — “Open right away”. Such a leader is likely to make an impulsive decision and disparage a deliberate effort to weigh short and long-term consequences of the decision by the team. The impulsive decision may also be guided by a personal agenda such as appeasing select groups.
On the other hand, a different business leader who is more concerned about the well-being and psychological readiness of its employee base than short term business gains, may swing the pendulum to the other extreme which may inadvertently endanger the viability of the business.
Both extreme positions are likely to miss best option as they will not exert the effort to fully evaluate facts of the situation and explore all alternatives in between the two extremes. For example, they may overlook the caveats in the state and federal guidelines for opening up. They may also fail to fully comprehend the dependency of their business on other parties and thereby fail to coordinate the decision with others, for example, supply chain partners.
Pandemic in its early stages represents a highly fluid situation, where the facts are blurry to begin with and consequences are wide spread along social, political, economic, and psychological dimensions. The interplay across these dimensions have unintended consequences and are almost impossible to predict with any reasonable accuracy. One might feel in such situations that there is no “right” answer and any answer will do.
First and foremost, the responsibility of a decision intelligence system is to dissuade decision makers from making impulsive decisions. This may not be as simple as it appears, as humans have evolved with a strong tendency toward impulsive action. Furthermore, we suffer from the weight of our prior experiences and knowledge making it difficult for others to persuade us.
Often, while making an impulsive decision, we are aware of its disadvantage but still go forward. We may we feel ill-equipped to handle the cognitive burden or stipulate lack of time. This is where the second role of a decision intelligence system comes into play — mitigate perceived or real time constraints and cognitive burden (thinking aversion) constraints of the decision maker.
Without having a deeper understanding of how people make decisions, it will be impossible to build an effective decision intelligence system. A decision intelligence system must be built around a sound decision making framework. Human and artificial agents can then collaborate following the structure and discipline of the framework.
The purpose and scope of decision intelligence automation is to implement artificial intelligence agents operating by the directives of a decision making framework.
A decision making framework may appear to be deceptively simple. Isn’t that obvious, one might say? How will somebody make a decision otherwise? Decision making frameworks when described can trigger such reactions yet we have been relying on such frameworks across various disciplines and different eras in human history for problem solving and for making sense of the universes within which we have to function. The least a framework does is to provide a structure and discipline without which an organization is bound to stay at the lowest level of decision intelligence maturity.
Early frameworks assumed that decision-making occurs at conscious level of processing guided by rational behavior. Today’s understanding of decision-making theories is much more nuanced. Complex cognitive processing is understood to be at play both at conscious and subconscious levels. Rationality and logical reasoning often appear after the fact in explanatory stories aimed to justify a decision. Explanation generation is not necessarily a factual recount of why and how a particular choice was made. More often than not, producing an explanation is an independent decision-action activity with its own context and objectives in cohort with the primary decision-action activity.
There are quite a few decision-making frameworks in circulation which can serve as the basis of a decision intelligence system. In this article, we discuss two such frameworks — OODA Loop and SEAL™. First appeared in 1995, OODA loop has its origins in military combative operations whereas SEAL™ is a relatively new framework by Diwo, LLC, a Michigan based technology startup providing decision intelligence products and services.
OODA Loop
“What Is the Aim or Purpose of Strategy? To improve our ability to shape and adapt to unfolding circumstances, so that we (as individuals or as groups or as a culture or as a nation-state) can survive on our own terms.”
— John Boyd
Developed by military strategist and United States Air Force Colonel John Boyd, not so much as a decision-making tool but more as a mental model for individual and organization learning and adaptation. OODA stands for observe, orient, decide, and act. The notion of the loop signifies the repetition of the OODA as situation evolves. It was applied by Boyd to train combat operations personnel during military campaigns. Since than many have applied OODA to non-military and non-combative settings in commercial and educational enterprises.
Observe — The “Observe” element in brief is about sensing the current situation which includes your individual situation and the environment around you.
Applied to IT systems, some equate this step to simply establishing connectors to data sources and ingesting data but the intent of the step is to also reflect upon what should be observed, why, and how frequently.
In Covid-19 context, it would mean collecting data about virus infections via test results, hospitalized patients, and casualties. This would include determining the extent of data to be captured.
Orient — The “Orient” element is about making sense of the observed situation from multiple perspectives pertinent to the goals one wants to achieve. In practice, this translates to applying filters to the observed data which we often do unconsciously based on genetic heritage, cultural dispositions, personal experiences, and domain knowledge. The purpose of making this process explicit is to ensure that we are aware of the unconscious filters and even conscious filters to offset undesirable cognitive biases. This allows one to accurately assess the current situation and the gap between the current and the desired situation which is necessary to carve out the future course of action.
Applied to IT systems, some equate the “Orient” process to data munging or converting data to information including data aggregations and knowledge mining.
Orientation in Covid-19 context would understanding spread patterns and projections.
Also included will be the characteristics of the current treatment cycle.
Decide — The “Decide” element is where a preferred course of action is selected after reviewing alternatives courses of action. The selection action becomes the hypothesis to be subsequently tested.
Applied to IT systems for decision support, this element would map to decision modeling and decision analysis.
For Covid-19, Decide would include forming hypotheses to be tested to minimize the multi-faceted impact of the pandemic and for discovery and production of vaccine.
Act — The final element is “Act” which amounts to testing the decision by actual implementation. This includes developing an action plan and oversee its execution.
Applied to IT systems for task automation, this element would map to process flow design, and process orchestration.
There is a comprehensive version of the OODA Loop, shown in figure above, but most consider it as a high level sequence of 4 steps and miss finer nuances underlying John Boyd’s intent.
OODA Loop Limitations as a Decision Intelligence Framework
The OODA Loop was developed by Boyd to create among his audience a way of thinking and to imbue in them the mental agility which is the key to survival in military combative operations. As it was not intended to be necessarily a guide for decision automation, it has limitations to be considered a generic decision intelligence framework. This is not to say that OODA loop is necessarily wrong, its limitations are more due to its incompleteness and its special focus on decision making in combative situations.
Speed of decision making — Speed of decision making is the prime focus of OODA Loop theory, as arriving at decisions quickly is often the decisive element in war. The penchant for speed can easily obscure other themes integral to decision making.
While timeliness of decisions and actions is important in all situations, timeliness concept is not equivalent to speed.
For example, in the Covid-19, it is highly desirable to find a vaccine at the earliest possible but rushing into one may pose greater risks. Same is the case with relaxing social distancing constraints. In the short term, it may seem to improve economic indicators, but in the long term early relaxation decision may prove to be disastrous if it triggers another wave of infections before a cure is found.
Decision making as a war game — OODA Loop is influenced by modeling decision making as a war game. There is an enemy and the overarching goal of decision making is to defeat the enemy. In the business enterprise setting, one consider the competition or even the market as an enemy, and may always assess situations as win-lose. Not all decision making situations in organizations are combative situations to begin with. Even in combative situations, there is often a room for win-win strategies. As information is freely shared and complex dependencies manifest, more and more decision making situations require mutual understanding and collaboration.
Covid-19 situation is a grim reminder of such complexity. How do you identify the enemy in this case? Is it the virus we are fighting or the callousness of politicians, or media biases, or ignorance of masses, or anxieties and fears of people, or economic devastation, or the enormous strain on the health care system?
Cognitive Burden as a Weapon– Contrary to decision intelligence goal of reducing cognitive burden, OODA Loop deploys cognitive burden as a weapon. It does so by overwhelming the enemy by putting cognitive burden on them by rapid OODA looping. It does not so much care about reducing the cognitive burden of the decision maker. Underlying rapid looping is the inherent assumption that enemy is responding to each and every one of your actions and at the same speed. This however is not always true. For example, in guerrilla warfare, enemy may take a quite rational approach to prolong a conflict and stretch out time.
Sequential View of OODA elements — Many view OODA elements as steps in a four step sequence which gets repeated over and over, even though that was not necessarily the intent of Boyd. These four elements must be viewed as continuous processes where decision emerge as a consequence of interactions among them. Sequential perspective does not allow for feedback loops between the steps which is common in complex decision making. The OODA framework also does not allow pursuing multiple choices of actions, for example doing A/B testing in marketing or simultaneous trials of COVID-19 vaccine prospects.
Human-Machine Collaboration — The focus of OODA loop was on description of a simplified mental model for decision making. Even though, one can possibly extend it to involve man-machine collaborative decision making, there are no easy hooks for such extensions.
SEAL™ Framework by Diwo
Evolved specifically for designing human-machine collaborative decision intelligence systems, SEAL™ is a proprietary framework underlying diwo’ s decision intelligence suite. SEAL stands for sense, explore, act, and learn. Though still transformational in nature, SEAL is designed to support and augment fundamental cognitive processes of human decision making rather than imposing on people to learn unfamiliar paradigms.
On first glance, SEAL may appear similar to OODA, since four elements of OODA can be loosely mapped to elements of SEAL as follows:
1. Sense element of SEAL encompasses both Sense and Orient elements of OODA
2. Explore element of SEAL subsumes Decide element of OODA
3. Act element of SEAL subsumes Act element of OODA
4. There is no explicit equivalent of Learn element of SEAL in OODA
The details under these elements are quite different however. This is partly because of the intent of SEAL to achieve a man machine symbiosis by reducing cognitive burden of decision makers and partly due to its continuous business optimization focus by explicit incorporation of feedback loops and learning. SEAL elements are realized by concurrently executing and collaborating intelligent agents.
Interwoven in SEAL is a formalized concept of “Opportunity” and the belief that business enterprises are primarily opportunity driven. In brief, an opportunity represents a time sensitive business situation which if not addressed proactively is highly likely to have an adverse impact on business. The impact may be due to increased risk and financial loss, missing of a growth prospect, or both.
Unlike Sense in OODA which is focused on sensing the current situation with the reactive intent, Sense in SEAL is proactive by design and predicts future situations in that qualify as opportunities early so that business have ample time to become ready to react. Transactional, contextual, and internal environment is continuously observed in Sense. Observations are run through prediction engines and their potential impact analyzed. Sense agents may reveal multiple opportunities from the same snapshot of observation, thereby, requiring branching of subsequent activity to address them concurrently.
For example, in the Covid-19 case, Sense of SEAL will be lot more concerned about what factors causes the virus to spread and when and where it can spread in the future than just knowing where it has already spread.
Sense in SEAL requires active collaboration among human and machine agents as neither of them on their own can handle massive amount of data and make sense out of it. Human agent determines what is of interest, priority, and urgency at any time. Machine agents adapts to this feedback and does heavy lifting of sorting through data, running predictions, and performing impact analysis in the background.
To illustrate human machine collaboration in the Covid-19 case, it is the human agent who sets the policy that in the short term controlling spread is lot more important than directing all resources for finding a cure. Accordingly, the machine agent will focus on understanding the spread. If human agent set a different policy, the machine agent would adapt to support that policy.
An opportunity after raised by Sense enters the Explore phase where human and machine agents collaborate to explore strategies for addressing the opportunity and evaluate outcomes of alternatives to narrow down the choice to actions to be rolled out. The collaboration not only allows for incorporation of experiential knowledge and experimentation with decision levers but most importantly allows for building trust and confidence in selected alternatives. Selected alternatives (similar to OODA hypotheses) move to the Action phase where actions are actually executed either manually or via a process automation substrate.
In Sense as well as in Explore, the machine agents actuate cognitive interventions to mitigate thinking aversion and decision fatigue of decision-makers.
The Learn element of SEAL is not so much as a distinct phase but a facet that permeates throughout all other elements. Learning happens at multiple levels - for adaption and fine tuning of predictive models and man-machine interaction. At a macro level, it is about understanding and improving the efficacy of action alternatives executed during ACT phase.
In summary, while OODA and SEAL are similar in emphasizing that a disciplined and facts based approach to decision making is essential for sustainable success in any endeavor, SEAL provides a comprehensive framework for implementing a human-machine decision intelligence system.