Summa Intelligentiae: What is AI? A Clear Definition
Quaestio Prima
01
What is Artificial Intelligence?
A Business Leader's Clear Definition
“Recrease?” A made‑up word, yes, but a very real rhetorical correction in the U.S. Navy Submarine Force. “Increase” and “decrease” are forbidden and crews must say “raise” or “lower.” Decades of successful Nuclear Navy operations has codified the lesson: doctrine makes language precise so meaning is understood and ambiguity eliminated. Clear, concise, direct communication is foundational to success.
“AI” in business is even muddier than “recrease.” It’s marketing’s hottest buzzword, the highlight of every earnings call, and the proper antecedent to “bubble.” Used as a linguistic Swiss Army knife, the label now says everything and therefore nothing. So, how do we restore clarity and, importantly, make sure our definition is shared?
Insights to Expect
How AI has evolved from inception to present
The gaps between scholarly and practicaldefinitions of AI
How to bound and codify whatever “AI” is being discussed
The organizational unlock of establishing precise definitions
AI did not arrive as a single breakthrough. The field co‑evolved with computing and repeatedly reinvented itself. Early decades favored symbolic systems (i.e. logic, rules, and knowledge bases) followed by cycles of hypergrowth, “AI summers”, and minimal advancement, “AI winters.” Then, statistical learning and data availability accelerated progress, culminating in today’s deep learning and transformer models. The graphic below illustrates a simplified development of the main branches of traditional AI. The path from the inception of the Turing Test to AI influencers has been anything but linear.
Simplification of the main branches of AI: Symbolic, Connectionist, and Hybrid.
A Quick Tour (without the gift shop)
Symbolic AI (a.k.a. GOFAI...Good Old-Fashioned AI): Systems manipulate symbols and rules to derive conclusions. Includes expert systems, automated reasoning.
Statistical/Machine Learning (ML) Era: Algorithms learn patterns from data. Includes supervised, unsupervised, reinforcement learning. The modern workhorse behind commercial AI and parent to deep learning, reinforcement learning, etc.
Deep Learning & Transformer Architecture: Engines of recent breakthroughs in perception and language. The upstream branch that encompassing generative models and large language models.
Everyday AI you don’t call “AI”: Your phone’s Face ID; your bank’s card‑fraud anomaly flags; your email’s spam filter. Mature, narrow systems doing specific jobs with high reliability.
Defining AI
A concise, academically grounded definition
Systems that perform tasks by learning patterns from data or by applying symbolic reasoning, rather than following explicitly programmed rules for each scenario. [Synthesized, mainly Russell, S., & Norvig, P.]
In short, an AI system must either learn from data to improve or apply symbolic manipulation and logical inference to produce results without an engineer hard-coding (explicitly defining) every branch.
Where policymakers have landed (useful guardrails)
U.S. Code (National AI Initiative Act): "A machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments" (15 U.S. Code § 9401)
EU AI Act (Article 3): "A machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments" (EU Artificial Intelligence Act, Article 3)
What, exactly, is "AI" today?
A family of techniques, some learned from data and others rule‑driven, that produce useful outputs without scripting every scenario. Generative models are a branch, but they are not the whole tree.
Deciphering Market(ing)-Driven Definitions
Let’s address the elephant on the earnings call. When everything is labeled "AI", the label stops helping you decide. Generative features get slapped on products like a wedding surcharge on a basic venue. Same building, bigger bill. So, even when vocabulary and definitions are misaligned, how can we examine the product while ignoring the packaging? To borrow from Bill (Shakespeare, not Gates), we should smell the rose.
Three-part test for a precise definition:
Inputs: What data sensors, or documents feed the system? Who owns and controls the data, and how often is it refreshed?
Outputs: Exactly what is produced (scores, classes, text, tool actions, etc.) and how is it consumed in our (i.e. your) workflow?
Process: Is the substrate learned (statistical/ML) or prescribed (symbolic/rules)? What evaluation evidence exists (offline tests, human review, real-time monitoring)?
Note: How outputs are validated may be a part of the process through iterative improvement or a part of the output. If validation is weak or absent, downstream controls and feedback loops must be added. Evaluations (i.e. evals) will be covered in greater depth in the future.
This isn’t pedantry; it’s necessary conversational due diligence. Beyond conversational alignment, this approach is a minimalist risk management approach. The NIST AI Risk Management Frameworkbegins with context and mapping, then measuring and managing risks. These steps are not possible if you can't articulate what the system is and how it behaves.
Precision over accuracy (why this saves time and money)
In this context, precision means everyone shares the same specific definition for the tool in front of you (i.e. inputs, outputs, and process) even if accuracy (alignment with academic definitions or consistency with other tools) varies. Precision lets you right‑size controls. A fine‑tuned model predicting energy derivatives needs different guardrails than a scripted FAQ chatbot with branched responses. Without precision, you’ll overspend on controls where they aren’t needed and under‑govern where they are.
A pragmatic organizational definition
A recommendation to be internally precise and accurate
We define AI as any machine‑based system we deploy (employ) that either learns from data to improve predictions/decisions or applies symbolic logic to derive conclusions, implemented for a clearly defined business objective. Everything else is automation.
Remember, and so that we give William his full credit:
“What’s in a name? That which we call a rose / By any other name would smell as sweet.”
Names are fine; substance is better. A precise definition yields clarity of communication, unlocking speed and eliminating marketing noise.
Apply and act
After the three‑part test, classify risk (impact, reversibility, autonomy), set guardrails (data governance, human oversight, monitoring), then decide: adopt, sandbox, or abandon. Not all “AI” is the same, and not all “AI” is AI.
Final Thoughts
Scholarly definitions converge on two engines: learning from data (connectionist) and logical reasoning (symbolic). These approaches are now blended in modern solutions (hybrid). Regulators have broad, capability‑based definitions that anchor governance.
Common usage over‑labels; cut through it with a substrate‑first lens and the three‑part test.
The Three-Part Test breaks an ambiguous term into components: inputs, outputs, process. Demand evaluation evidence and consider monitoring requirements.
This investigation of meaning is the first building block of effective implementation. It lets you abandon misfit tools early, pick simpler alternatives when warranted, and invest where the solution serves the problem, not vice versa.
With a precise, shared definition, you can return to the right order of operations: define the problem, then pick the solution—sometimes AI, sometimes not.
Did you enjoy this email?
Share it with a friend and don't forget to subscribe
