Advanced Analytics, VECTOR and the Human in the Loop
Why faster decision-making makes human judgement more important, not less
Advanced analytics is changing the way organisations understand complex environments and make decisions within them. Increasingly supported by artificial intelligence, machine learning, predictive modelling and automation, these capabilities can process large volumes of information, identify patterns, detect anomalies, model potential outcomes and inform decisions at speeds that would be difficult to achieve through human analysis alone.
The opportunity is significant. So too is the potential for error.
An incorrect assumption, incomplete dataset, poorly framed objective or failure to recognise changing conditions can move through an analytics-enabled decision system significantly faster than through a traditional human process. As organisations become increasingly capable of converting data into analysis, recommendations and action, the challenge is therefore not simply whether advanced analytics should be used. It is determining where analytics strengthens decision-making, where human judgement remains necessary, and when responsibility needs to shift decisively back to the human decision-maker.
This is where Human-in-the-Loop (HITL) becomes important.
It is also where VECTOR provides a useful way of thinking about the relationship between advanced analytics and human judgement.
Advanced analytics as an enabler
Advanced analytics is best understood as an enabler within the decision environment rather than a substitute for decision-making itself.
Its capabilities range from business intelligence and statistical analysis through to predictive modelling, machine learning and increasingly sophisticated applications of artificial intelligence. At the more advanced end of this spectrum, AI can synthesise information, identify relationships, generate options, recommend priorities and increasingly support or automate elements of execution.
The technology may evolve considerably. The underlying leadership problem remains.
Advanced analytics necessarily operates within parameters established by people: the information available to it, the questions being asked, the objectives being pursued, and the assumptions, authorities and constraints within which it operates. A technically sophisticated system can therefore produce an operationally inappropriate answer when those underlying conditions no longer reflect reality.
This creates an important distinction:
Advanced analytics can optimise within a decision environment. Human judgement remains necessary to determine whether that environment still reflects reality.
VECTOR provides one way of making that distinction operational.
Regulate: establish the boundaries
The first requirement is not speed. It is control.
Within VECTOR, Regulate establishes the conditions within which effective decision-making occurs. In an analytics-enabled environment, this includes determining what systems are permitted to do, the information upon which they can rely, the level of automation appropriate to the task and the points at which human involvement or authority is required.
As artificial intelligence becomes embedded within analytical systems, this becomes increasingly important. The apparent precision of a model, dashboard or AI-generated recommendation can create unwarranted confidence, particularly when the reasoning behind it is difficult for the operator to interrogate.
Human decision-makers need sufficient understanding to recognise uncertainty, question assumptions and resist automation bias. They also need clarity regarding the boundaries within which an analytics-enabled system has been authorised to operate.
The faster the system can operate, the more important those boundaries become.
Regulation provides the discipline that allows speed without surrendering control.
Verify: establish reality
Advanced analytics can substantially strengthen verification.
It can analyse multiple information streams simultaneously, identify inconsistencies, detect emerging patterns and bring information to the attention of decisionmakers that might otherwise remain obscured. AI can extend this capability further by identifying relationships across datasets, assessing large volumes of unstructured information and helping decisionmakers explore competing explanations.
However, more information does not necessarily produce greater understanding.
Verification remains concerned with establishing the most reliable picture of reality available at the time. Decisionmakers still need to consider the provenance, currency, completeness and credibility of information, understand where uncertainty remains and distinguish evidence from assumption.
In an analytics-enabled environment, an additional question becomes important:
Are the assumptions underpinning the analysis still valid?
This matters because even an analytically accurate system can produce an operationally inappropriate answer if the circumstances upon which its assessment depends have changed.
Advanced analytics may improve our ability to see.
Human judgement remains necessary to determine what we are actually seeing.
Prioritise: determine what matters
Advanced analytics is particularly powerful when organisations face competing demands.
It can compare options, rank priorities, model alternative courses of action and optimise the allocation of finite resources against defined objectives. AI can increasingly assist by identifying relationships and potential consequences that may not be immediately apparent to the decision-maker.
The critical words are against defined objectives.
Someone still needs to determine what matters.
During a significant incident, different objectives may compete. Safety, operational continuity, reputation, financial impact, legal obligations, strategic objectives and the needs of affected people may not align perfectly. Optimising one may create unacceptable consequences elsewhere.
Prioritisation is consequently more than a mathematical exercise. It involves consequence, context, values, intent and judgement.
Advanced analytics can help determine the available choices and illuminate their implications.
Leadership determines which consequences matter most.
Execute: convert decisions into action
Advanced analytics can also improve execution.
Once objectives, authorities and boundaries have been established, analytics-enabled systems can support the allocation of resources, automate defined processes, coordinate activities and identify deviations from expected performance. Where AI and automation are integrated into these systems, actions may increasingly be initiated or undertaken without requiring continuous human involvement.
In some environments, requiring explicit human approval for every activity would remove much of the speed and efficiency that advanced analytics and automation were intended to provide.
Human-in-the-Loop should therefore not mean human approval of everything.
The appropriate degree of human involvement should instead be proportionate to uncertainty, reversibility and consequence.
Low-consequence and readily reversible activities undertaken within well-understood parameters may be suitable for substantial automation. Increasing uncertainty or consequence may require greater human supervision. Decisions carrying significant strategic, ethical, legal or potentially irreversible consequences are more likely to require explicit human authority.
The principle is not human involvement everywhere.
It is human involvement where judgement matters.
This distinction becomes increasingly important as execution accelerates. An incorrect decision moving slowly through an organisation creates one form of risk. An incorrect assumption replicated across an automated decision system at machine speed creates another.
Accordingly, faster execution increases rather than diminishes the importance of knowing when human intervention is required.
Continuous Reassessment: what has changed?
Continuous Reassessment is where advanced analytics and the Adaptation Point become particularly closely connected.
Advanced analytics is exceptionally well suited to persistent monitoring. It can detect deviations, identify emerging patterns, monitor thresholds and recognise changes in the operating environment far more rapidly than a human team attempting to monitor every variable simultaneously. Predictive techniques can also help identify where conditions may be moving rather than simply describing where they are now.
AI can extend that capability further by drawing relationships between disparate information, highlighting anomalies and supporting increasingly sophisticated assessments of emerging conditions.
However, detecting change and interpreting its significance are different activities.
Conditions always change.
New information emerges. Competitors and adversaries respond. Resources are consumed. Priorities shift. Assumptions weaken. Decisions create both intended and unintended consequences.
The relevant question is therefore not simply:
Has something changed?
It is:
Has reality changed enough to require a change in approach?
That is the Adaptation Point.
HITL at the Adaptation Point
This provides a more useful way to think about Human-in-the-Loop.
Rather than positioning the human as a permanent approval mechanism that slows every analytics-enabled activity, human judgement can be deliberately concentrated at the points of consequence, uncertainty and adaptation.
Where activities remain within established assumptions, authorities and tolerances, analytics-enabled VECTOR can continue to operate.
As uncertainty increases, information conflicts or operating conditions begin to diverge from expectations, human oversight should correspondingly increase.
Where there is significant change to the assumptions, consequences, authorities or intent underpinning the existing approach, an Adaptation Point is reached.
At that point, human judgement becomes decisive.
The response may be to ADAPT. The existing approach no longer sufficiently reflects reality and deliberate change is required.
Equally, the response may be to HOLD. Circumstances have changed, although the underlying approach remains appropriate.
This distinction matters.
Adaptive leadership is not constant movement. Responding to every new signal can be as damaging as failing to recognise genuine change.
The value of continuous reassessment lies not simply in identifying change, but in distinguishing between change that is merely occurring and change that is consequential.
Advanced analytics can detect change. AI can help interpret and anticipate it. Human judgement determines whether the change matters.
Human-in, human-on and human-out-of-the-loop
There is consequently value in thinking beyond a binary choice between human decision-making and automation.
A proportionate decision model can recognise different levels of human involvement.
Human-out-of-the-loop may be appropriate where automated systems operate within tightly defined parameters, the consequences of error are limited, actions are readily reversible and intervention provides little additional value.
Human-on-the-loop allows an analytics-enabled or automated system to operate within established authorities while a person supervises its performance and retains the ability to intervene.
Human-in-the-loop requires human judgement or authority before consequential action occurs. This becomes increasingly important where uncertainty is significant, actions are difficult to reverse, consequences are substantial, competing values are involved or circumstances have moved beyond the assumptions on which the system's existing authority was based.
These are not necessarily fixed categories.
Human involvement should be capable of increasing or decreasing as circumstances change.
VECTOR can operate across that spectrum. What changes is where decision authority sits and the conditions under which it shifts.
This creates an important role for Continuous Reassessment.
Monitoring should not be concerned only with whether an action is producing the expected result. It should also consider whether the circumstances still justify the degree of autonomy originally granted.
The Adaptation Point can therefore serve as both a leadership concept and a useful principle for designing human-machine decision systems.
Knowing when to intervene
Advanced analytics is likely to continue increasing the speed, volume and sophistication of decision support available to organisations. Artificial intelligence will form an increasingly important element of that capability.
The appropriate response is neither to resist this development nor to assume that faster automated decisions are inherently better decisions.
The requirement is to combine the relative strengths of machine capability and human judgement.
Advanced analytics can detect, compare, model and anticipate.
AI can increasingly interpret, recommend and execute.
Human beings remain responsible for purpose, consequence and judgement.
VECTOR provides a disciplined way of connecting these capabilities. Human-in-the-Loop provides a mechanism for retaining human authority where it matters. Continuous Reassessment provides the means of determining whether circumstances remain within the assumptions and tolerances supporting the existing approach.
The Adaptation Point then presents the critical question:
Has reality changed enough to require a change in approach?
Sometimes the answer is ADAPT.
Sometimes the answer is HOLD.
The advantage may therefore not belong to the organisation that removes people from the decision cycle.
It may belong to the organisation that knows precisely when human judgement matters most.
Stable Intent. Adaptive Action.
Montane Protective Security provides intelligence-informed protective security, risk assessment and security advisory services to organisations operating in complex environments.
Comments