The "Serenity Strategy" of Generative AI Adoption

Artificial Intelligence, in particular Generative AI in not magic.? It appears that way because we have a hard time explaining it.? As such, we struggle to? find firm footing as business leaders to approach adoption and manage how we build a sustainable capability in our organizations.? A simple definition of AI in three parts can offer a practical and useful understanding to frame how we manage it.?

Artificial Intelligence is a the systematic synthesis of vast amounts of information to derive knowledge and create a cognitive capability that can enhance human decision making and action.

Note: While aspects of this apply to machine learning and model building we do in the enterprise, I will focus on Generative AI which includes agents and services using ChatGPT, Claude2, Gemini, Llama…)

Synthesis of vast amounts of information

Garbage in, garbage out has never been so true.? The quality of the data used to build a “model” is essential.? Building them is like feeding a pig that will eat anything because of an extraordinary ability to digest matter and extract nourishment. This does not mean that we are fattening the swine to yield quality ham.? What if a model is built on information that only represents part of the reality we live in?? What if the data is from unreliable sources?? ..or if the data that is proprietary to someone else.? What if there is no data in the first place yet we receive responses (there are no shortage of anecdotes of GenAI service “hallucinating”)?? This will be the stuff of comedy in the coming years…and lawsuits and firings…but also some breakthroughs.

Unless you are already developing domain models in-house, the data used to build Generative AI models is the internet…the good, the bad, and the ugly. ? We have limited control over the quality but there are ways that we can begin to mitigate this risk.? The key action is to initiate a project to establish guardrails based on cross-functional risk analysis that defines policy of acceptable use, create scenario-based prompt standards and API templates, and define how you will monitor usage.? You’ll see this is a very similar exercise to defining mobility policy and strategy.? The process of defining boundaries invariably entails considering possibilities

Derive knowledge and create a cognitive ability

This is where technology raises the bar and will ultimately become truly powerful.? As the computer can process more information, it evaluates orders of magnitude more combinations of information. Detecting deeper correlations and relationships imperceptible to humans is only the first step.? It applies those relationships to learn how to perform progressively higher order reasoning and tasks steadily approximating human capability. ?

The primary issue is that Generative AI tools are a black box and we can not really explain where the answers came from or validate that they are accurate.? There are two ways to mitigate this risk.? First, ensure that your teams test use cases with various available LLMs to see how they perform with your different use cases where they have sufficient domain knowledge to catch inconsistencies with reality. This will not be “scientific,” but a cross functional team will begin to identify the nuances, strengths and weaknesses. ?

The second approach is to develop prompts with rich context and bounded by your own data that you trust. Your teams will understand this as Retrieval Augmented Generation (RAG).? A good prompt includes multiple parameters or variables that create greater specificity as to what you are asking. These parameters can be dynamic and based on your own data.? Service architectures like AWS Bedrock or those offered by Google Cloud and Azure will enable you to build Generative AI workflows that secure your data and manage access across various prompt scenarios.?

Don’t be fooled by anthropomorphic features of a genial chat bot that you are engaging directly.? It’s still a computer and your prompts have to give it instructions as if you are explaining a task to a primary school child who needs to be told everything that you expect of their answer and how they should complete the task. In fact, I saw a prompt during a presentation last week that told the AI agent to read the input text a second time…and it didn’t talk back!?

Enhance human decision making and action

ChatGPT seemingly came out of nowhere a year and a half ago.? Because it seems like magic we risk overestimating what Generative AI can do.? It does not replace well-formed human reasoning.? When you look critically at enough GenAI responses, you’ll realize that it is really just synthesizing and reporting on information it has found on the internet.? Anything resembling “perspective” is often a trite and “trained” platitude offered as a hedge. In fact, if you ask ChatGPT about its limitations to reason, the first thing it will tell you is that it “lacks real-world experience-” exactly what makes our people valuable.

With every new technology I am reminded of some wisdom shared early in my career.? “A fool with a tool is still a fool.”? This is not a cynical perspective, however.? I would also share something pertinent I heard recently that “it’s not the AI that will take our jobs, but the people who learn to use the AI.”? Our role and mandate as managers is maximize the capacity of people and find the practical possibilities to start to build a sustainable organizational competency. ?

Encourage each part of the organization to articulate what enhanced productivity means across various roles with the intent to ease low value tasks and support higher level thinking activities.? Generative AI has an incredible ability to perform mundane tasks that are the bane of an employee’s existence, especially when combined with a RAG approach.? Also, Generative AI will be most powerful when used to enhance a robust and inquisitive thought processes based on a structured analytical method. This means we need to ensure our people?are prepared to logically break down a problem into a series of questions and evaluate different outcomes or scenarios.? ?

For both levels, approach this as a sponsored program of initiatives that follow the guidelines established by your new policy and GenAI technology standards.? Also, consider introducing some education on higher order thinking or critical analysis and reasoning skills.? LinkedIn Learning, Khan Academy, and Coursera all have options.?

For some reason the “serenity prayer” kept haunting me as I wrote this!?

God, grant me the serenity to accept the things I cannot change, the courage to change the things I can, and the wisdom to know the difference.

Managing sometimes requires a leap of faith, but if we establish the right direction and parameters for our people, they will find ways to flourish with the enhanced capability that Generative AI offers them and achieve higher level outcomes for the organization. ?