1. The Original Meaning of GIGO
For nearly seventy years, “garbage in, garbage out” (GIGO) has been a foundational law of computing. Coined in 1957 by Army mathematician William D. Mellin, the principle was a warning: because computers cannot “think,” a careless programmer who provides incorrect figures will inevitably receive an incorrect answer.
In traditional building automation workflows, this truth is easy to see. Inaccurate sensor data leads to poor control decisions. Incorrect point mapping creates unreliable graphics. Poorly written sequences of operation result in inefficient or unstable system performance. The machine, in its deterministic rigidity, acts as a mirror: it can only reflect the quality of its input.
This skepticism predates even the silicon chip. In 1864, Charles Babbage, father of the Difference Engine, was asked: “If you put into the machine wrong figures, will the right answers come out?” He was paralyzed by the confusion behind such a question. To the deterministic mind, the answer was obvious: no.
2. Why Generative AI Changes Conversation
The rise of generative AI introduces a more nuanced reality. Modern AI tools, especially Large Language Models (LLMs), are no longer passive mirrors. They have developed internal “sanity checks” that allow them to detect, and often correct, errors in their own processing before a single word is externalized.
Consider Neural Machine Translation (NMT). Researchers have discovered that these models don’t just translate text; they actively perform an internal Grammatical Error Correction before they even begin translating. Using Representational Similarity Analysis, scientists can watch the model’s neural layers at work. Feed it a sentence with a noun-number disagreement. In early layers, the model sees the error. But as the signal moves deeper, the internal representation of that “garbage” word physically shifts closer to its correct grammatical form.
This is driven by Robustness Heads, special attention mechanisms that focus on specific linguistic units. In English, they fix article errors (“a apple” → “an apple”) based on phonetics. In French, they focus on noun genders. The AI isn’t just repeating patterns; it is making a judgment call about what you meant to say.
Similarly, LLMs possess a “second opinion” mechanism. There is the first-order model that picks the most statistically likely next word (standard auto‑complete). But hidden inside is a second‑order evaluative signal, often called X_eval, which performs an independent check on “question‑answer fit.” It allows the model to look back at a completed sentence and realize, “That answer doesn’t actually fit the facts.”
Most fascinating is the Post-Answer Newline Token (PANL), the very first blank line or “enter” character after a response. The PANL acts as a summary of the AI’s self-awareness. It predicts whether the model has made an error far better than standard confidence scores, and, most importantly, it predicts correctability: whether the model actually knows the right answer deep down, even when its surface output is wrong.
This is the AI equivalent of pausing mid‑sentence, realizing you made a slip of the tongue, and correcting yourself before you finish speaking.
3. AI as a Communication Amplifier
For building automation professionals, this technical evolution leads to a practical shift. AI can now function as a communication amplifier, not a replacement for expertise, but a tool that reduces friction between what you know and what you can express.
Have you ever typed a rushed, grammatically flawed prompt into a chatbot and received a surprisingly useful answer? That is AI acting as an amplifier. It interprets imperfect language, organizes rough thoughts, and helps convert early-stage ideas into clearer, more actionable communication.
In daily BAS work, this means:
- Summarizing trend logs into plain‑English observations.
- Drafting troubleshooting reports from field notes.
- Clarifying sequences of operation when your initial description is messy.
- Improving technical communication with facility managers or clients.
A prompt does not need perfect grammar or polished structure to produce valuable results. What matters more is whether the user provides clear intent, relevant context, and sound assumptions.
4. Where GIGO Still Applies
Despite these advances, the core warning behind GIGO has not disappeared. It has simply shifted. AI may be forgiving of form, but it remains highly sensitive to intent. If your underlying thinking is misleading, your assumptions are wrong, or your problem is poorly scoped, the output can still become misleading—even if it sounds confident and well‑written.
GIGO still applies in several critical ways:
- Flawed assumptions → The AI will reflect them, potentially amplifying errors.
- Missing context → The model may fill gaps incorrectly, producing plausible but wrong answers.
- Vague or contradictory goals → Outputs will mirror that confusion, leading to unreliable recommendations.
In short, GIGO remains true for intent, if not for form. AI can polish silver, but it cannot turn a fundamentally broken question into gold.
5. A Building Automation Example
Let’s make this concrete. Suppose a controls technician is troubleshooting an air-handling unit (AHU) overheating. The technician types into an AI assistant:
“Why is my AHU supply temp high? The setpoint is 55 but it’s 68. Maybe the valve is stuck.”
The prompt is far from perfect, grammar is casual, the problem description is incomplete, and the technician has already jumped to a hypothesis. In a deterministic system, such a vague input might produce useless output. But a modern LLM can interpret the intent:
- It recognizes the equipment type (AHU).
- It understands the discrepancy (setpoint vs. actual).
- It notes the proposed hypothesis (stuck valve).
- It can then respond with a structured list of possible causes: valve failure, sensor drift, incorrect scheduling, heating coil issues, or control loop tuning.
The AI might also ask clarifying questions: “Has the valve been manually exercised recently? What is the valve position command vs. feedback?” This turns rough, shorthand field notes into a systematic troubleshooting guide.
But where does GIGO still apply? If the technician had typed: “AHU hot. fix it.” That lacks intent. The AI might guess, but the result could be generic or wrong. Or worse, if the technician assumes the wrong sensor type (e.g., thinking a supply air sensor is a return sensor), the AI will likely follow that flawed assumption unless corrected. The machine can polish your language, but it cannot verify your field measurements or override your false premise.
6. From Perfect Prompting to Clear Thinking
This leads to a crucial realization: the goal is not perfect prompting, it is clear thinking.
In the old deterministic world, you needed a perfectly structured command. In the new probabilistic world, you need a well‑reasoned intent. The most effective AI users are not those who memorize arcane prompt templates (like COSTAR, CRISPE, or RACE). They are professionals who:
- Think critically about the problem before typing.
- Provide relevant context and constraints.
- Formulate a clear goal, even if the wording is rough.
- Treat AI as a thinking partner, not an oracle.
- Verify outputs against field knowledge and system behavior.
For BAS professionals, this means integrating AI into your existing discipline. Use it to draft sequences, interpret trend data, or explain a control logic puzzle. But always cross‑check the results against your own understanding of the mechanical system, the sensor layout, and the sequence of operations.
7. The Takeaway: Intent In, Value Out
We’re stepping into an exciting new era where AI isn’t just a tool, it’s like a collaborative partner equipped with a built-in “sanity check.” Remember, the old machines were mere mirrors of our actions, but now, we’re working alongside thinking partners. While they might not truly “understand” you in the human way, they offer something just as meaningful: a safe space for expressing your ideas more easily. They help transform our often messy thoughts, frustrations, tentative plans, and rough ideas into clearer, more practical language, making collaboration smoother and more intuitive.
The GIGO era is not dead. It has evolved. A better principle for the AI age is:
“Intent in, value out.”
When technicians, engineers, facility managers, and leaders bring clear thinking to the process, AI can amplify their expertise. It can turn rough notes into stronger documentation, help organize complex observations, and reduce the friction between what a professional knows and what they are trying to communicate.
But a final caution remains: Only honest thinking can tell the difference between an output that is genuinely useful and one that is merely polished. AI can improve the form of our communication, but it cannot guarantee the truth of our assumptions. In BAS work, the strongest results come when AI is paired with field verification, sequence knowledge, and disciplined troubleshooting.
So, the next time you type a messy prompt and get back gold, remember: the magic isn’t in the grammar. It’s in the clarity of your intent. And that clarity is still a human responsibility.
Intent in, value out — but only honest thinking can tell the difference.
