Why Context is Easy to Overlook And How AI Can Help
In an earlier article, I discussed how an AI agent can reduce the cognitive burden for a human when decision-making. This essay delves deeper into the how, discussing the role of context in an AI-powered decision intelligence system.
How often are our words taken out of context, and how often do we take others’ words out of context? And more importantly, how does it feel? Perhaps, as if the world barely understands us. On top of that, we may carry guilt or shame from the human imperfection of misunderstanding others.
Even when we overlook context, it seems context doesn’t overlook us — the lack of it affects our choices, conclusions, and therefore, our relationships. This is even more complex professionally, where we not only interact with layered humans but also machines. The best outcomes arise when the interplay of human communication and digital interaction exists in harmonious synchronicity.
A decision intelligence system is the equivalent of a personal AI assistant, but for an enterprise — converting noise into an orchestrated electronic symphony. But how? In order to understand, we must first get intimate with context itself.
The Power of Context
“Though context may not be everything, nothing exists without context.”
Context is a force, and therefore differs from information, which is only inert matter.
But wait, “Isn’t context just another piece of information — information about information?”
It is information, and it is also more than that. It can be merely characterized as an influential factor in decision-making, but that definition would be lacking.
As linguists suggest, context imparts meaning onto information. And even more significantly, it has the power to invoke thoughts and actions. Context is many factors that we don’t understand and have no control over.
However, to be understood and manipulated by AI, it must have concrete representation. When we have a lack of context, we may find ways to cover up for that lack, and other times context manipulates us. So we can observe some common behavioral patterns:
- We presume context when it is missing
- We fill-in context out of convenience
- We make context the scapegoat of our errors and omissions
- We allow context to distort our perceptions
- We allow context to bring us together or drive us apart
To have power over context, we must, alongside AI, grasp these intrinsic behavioral patterns fully.
1. We Presume Context When It Is Missing
Let us look at an example of how context can invoke certain thoughts and behaviors based on the meaning we impart.
A human sees or hears the word, “apple” and then imparts meaning, and conjures up the ideas or action which range from wanting to upgrade their iPhone to craving hot apple pie with cold vanilla ice cream.
2. We Fill-in Context Out Of Convenience
More than we would like to take credit for, we fill-in context both unconsciously and consciously. And some people, politicians, for instance, are especially adept at making conscious substitutes out of convenience and always like to leave room for plausible deniability.
We are all influenced by our present beliefs, values, memories, motives, and goals.
After hearing the sentence with a mythological meaning, the listener adds additional context, that diverts the focus from mythology to ‘fruitology’ based on their personal agnostic spiritual beliefs.
3. We Make Context The Scapegoat Of Our Errors And Omissions
Humans often cover up our own errors and omissions in one of two ways, related to context:
- “Since I had no context, I had no choice except to provide my own.”
- “I said that, but it’s not what I meant. You quoted me out of context.”
Current pandemic media coverage and the ensuing discussions with friends, family, and co-workers provide plenty of real-life examples of both. How many variations have we heard up until now on what is safe and what is not safe in this current climate?
4. We Allow Context To Distort Our Perceptions
There are many examples within the fields of psychology and cognitive science speaking to the distortion of our perception through context. Figure 4, The Ebbinghaus Illusion showcases how two separate visual contexts can trick our brains into misperceiving the size of the red dot. While it seems larger in the image on the right, it is very much the same size as the one on the left.
5. We Allow Context to Bring Us Together or Drive Us Apart
When we communicate with each other, we each speak and listen from within our own context, therefore creating an interaction ripe for misunderstanding.
While providing a shared context seems like the obvious solution in theory — in reality, we know it’s not that simple. Even if we begin a conversation agreeing on a shared foundational context, it tends to get distorted along the way.
At a minimum, it would take a continual effort to maintain a shared context.
Power Over Context
In order to tame the beast that is Context, we must take the following four actions:
- Chase the beast out of hiding
- Size up the beast
- Look after the beast
- Train the beast to act on demand
1. Chase the Beast Out of Hiding
Context is the masterful at burying itself within the decision-making and decision-action processes. It is difficult to distinguish it from information or process, but it must be extricated to do anything more with it.
So, how?
Isolating Context from its surrounding environment is an intricate modeling exercise, and is predominantly manual. It is like developing business process and decision models, but requires extra effort on top of these models. A decision model may interact with multiple context models. Inversely, a context model may too interact with multiple decision models. Both decision and context models eventually get embedded in business process models.
Context modeling is impossible to automate fully. Still, AI agents can help simplify the modeling effort in certain ways, for example:
- AI agents can extract concepts from business documents and supporting models for inclusion in the context model
- AI agents can evaluate user-defined concepts for their suitability for the context
- AI agents can help define and refine the context relationship with its surroundings
2. Size Up the Beast
When we chase the beast out, we give it an independent existence and identity. We must then size it up by developing a fuller and concrete structural and behavioral representation of the context — manipulable by computers.
We can compare the untamed Context to Lernean Hydra — the multi-headed beast in Greek and Roman mythology (see Figure 5). We have to pay attention to each head individually — the shape, the contour, where it begins and ends, because we don't know when and where each head might raise itself and cause havoc.
An AI agent can help in sizing up the beast by:
- Automatic discovery and flagging of errors and omissions in contextual knowledge representation
- Automatic correction of common errors
3. Look After The Beast
Giving the context an independent identity and form is like getting the beastly creature out of the jungle and putting it into a cage. In the wilderness, the creature knew how to survive, but in the cage, we have to look after it to keep it alive and useful.
We can do this by keeping current. Individual components of context may require different approaches for keeping them current.
AI agents can help look after the context in the following ways:
- AI agents can continuously monitor the environment and apply attention filters to refresh the context
- AI agents can predict uncertain aspects of context, using pre-trained AI models
- AI agents can converse with the human context manager for updating context in a timely and consistent manner
4. Train The Beast To Act On Demand
Training the beast means making it ready to work for you on demand.
When tamed by training, the hostile Lernean Hydra transmutes to benevolent Sheshnaga of Hindu mythology who serves not only as a seat in the ocean for Lord Vishnu but also protects him from the elements.
You may have the current Context information always at your disposal—it’s of no use though, if you do not have a clear understanding of how and where to engage Context in the decision-action process.
From a cognitive standpoint, Context is primarily of use in two ways— for controlling complexity, and for reducing uncertainty.
An AI agent can be trained as an attention filter to retain only those parts of input signals from the environment which are relevant for the Context at play. On the other end of the decision-action process, another AI agent can use Context to help narrow down action choices. At other places in the decision-action process, different parts of Context guide the exploration and evaluation of strategies.
Context Awareness In A Decision Intelligence System
Context-awareness is the distinguishing characteristic of a decision intelligence system. It puts it in an altogether different league than other data-driven intelligence options, such as BI, Intelligent Search, and Machine Learning powered prediction engines.
Many researchers and implementers of decision intelligence systems fall short of engaging the context fully in their design and architecture. Being context-aware does not end with using contextual information for reference purposes only.
Is there a deeper reason for the limited conscious utilization of context? Or is it merely the lack of available data or lack of attention?
Could it be that our years of training in the reductionist approach is somehow hindering us from being fully context-aware? In our zeal for separation of concerns, we may be inadvertently seeking to design components that work across all contexts.
Application independence has been a critical design principle in database systems design. The database schema design is for a domain—it is generally oblivious to actual query patterns—the real clients of the database.
In training machine learning models too, we seek for robustness across all contexts.
A decision intelligence system has to transcend beyond drawing rigid boundaries influenced by heuristics, such as separation of concerns. What transpires at the edges is often much more important than what happens within.
Understanding context, measuring and estimating the effects of context, and effectively deploying context is a fundamental aspect of designing a decision intelligence system.
