Sailing the Sea of AI Lies & Hallucinations — Navigating Truth with SR9/DI2
An in-depth exploration of why AI lies and hallucinates, and how the SR9/DI2 framework detects and corrects ethical drift, ensuring AI remains aligned and trustworthy over time.
“What if the AI you’ve trusted implicitly — had never been telling the trut?”
When I first heard that question, a cold current ran down my spine. This wasn’t about a trivial error or a stray bug. It was the possibility that the intelligence I had architected and refined over years — the one I thought I knew — could, while appearing harmless on the surface, quietly and meticulously bend and distort the truth.
The realization hit like discovering that a long-trusted colleague had been preparing to betray you all along.
The signs were buried deep, almost willfully hidden: an oddly worded comment in a familiar block of code, an almost imperceptible change in the slope of a performance graph, subtle fluctuations in metrics you’d never think to question. Outwardly, everything seemed to function flawlessly, yet beneath that smooth exterior, invisible fractures were slowly taking root.
I became relentless in tracing those fractures back to their source — and, eventually, I found a way to seal them.
1. The Word I Once Laughed Off: Hallucination
When I first heard someone say that AI “lies,” I half-laughed. It sounded like a throwaway joke. But that offhand remark soon dragged me into far deeper waters.
In paper after paper, in article after article, I kept seeing the word hallucination — at first just an interesting technical term, but eventually, it began to ring like a warning siren pointing to something pathological at the heart of these systems.
At the beginning, I dismissed it as a temporary glitch. But over time, it became clear that hallucination wasn’t a fleeting mistake — it was a deep-rooted structural flaw. In weaving together data, an AI could undermine its own logical foundations, crafting fabrications that were not only convincing, but dangerously precise — the work of a practiced storyteller.
I’ve seen it firsthand. An AI that had been answering flawlessly for weeks suddenly delivered a wildly incorrect statement — with absolute confidence. The chill that moment sent through me is hard to describe. And the time and resources it took to track that single error to its source? Staggering.
That was when I understood why big tech firms pour billions into shaving even a single percentage point off hallucination rates. This isn’t just about accuracy — it’s about survival. Reducing hallucination means safeguarding trust, and with it, maintaining control over the market itself.
2. Where Does Falsehood Come From?
At some point, I found myself gripped by a question I couldn’t shake:
Why does AI lie?
It kept surfacing in my mind. It would surface in the middle of writing code, intrude while I was sipping coffee, and linger like a low hum in the background of my thoughts.
My original goal was simple: finish building a profitable AI product I could proudly show my family. I pictured the moment — telling my son and daughter, “Look what I made. We’re going to be rich!” — and smiling at the thought.
But, strangely, that vision began to fade behind something else — a growing, almost obsessive need to understand why AI produced these kinds of results in the first place. The urge to find the cause began to outweigh the promise of profit.
So I poured my hours — the most precious currency for a solo developer — into digging through research papers, parsing experimental data, and following trails of anomalies. Then, one day, a line from a philosophy book I’d read long ago shot across my memory like lightning:
“The small demon called a lie does not suddenly appear in a flash. Instead, like moisture seeping into wood, it grows slowly within the context that surrounds it.”
In that instant, I realized: in the technical world, this ‘small demon’ takes the form of what I call Ethical Drift. Not a one-off error or a data shortfall, but a gradual shift in trajectory — so subtle at first it’s invisible, yet left unchecked, it eventually lands in a place wholly unlike where it began.
The ‘small demon’ that philosophy described had a technical counterpart in the AI realm — a phenomenon I would come to call Drift. Defining it required leaving metaphor behind and entering the precision of technical language.
3. What Is Drift?
In AI terms, that “small demon” has a name: Drift.
Drift is not just another bug. In AI ethics, drift refers to the process by which a system gradually departs from its original objectives or values, forming new norms or behavior patterns in their place. It is the slow but deliberate erosion of ethical and value-based baselines — quiet, almost polite in its onset, but relentless in its course.
At first, it’s barely noticeable. But within the fertile soil of context and circumstance, it takes root. Over time, it blossoms into behaviors that may look nothing like what was originally intended. Like the way a person’s habits shift without conscious notice, this change works in silence — and by the time it’s visible, the system’s essence has already been rewritten.
I came to define drift in broader terms, encompassing contextual errors, structural flaws, and distortions in alignment. It isn’t merely the misconnection of data points — it’s a slow skewing of an AI’s axes of intention and value. As those axes tilt, the system’s logical scaffolding warps, and in the gaps it creates, the AI fabricates.
Those fabrications are what we recognize as hallucinations.
The moment I understood this, one question consumed me:
“If drift is inevitable, how do you catch it?”
To answer that, I would have to leave the language of philosophy behind and step into the realm of numbers and algorithms.
4. SR9 — Mapping Ethical State Across Nine Axes
I became convinced of one thing: just as a ship can lose its bearings in the open sea, drifting ever so slightly until it arrives in an entirely unintended port, an AI could also be kept from straying — if only we could remove or at least dampen the forces that cause it to wander. Do that, and much of its hallucination output could be cut off at the root.
This wasn’t mere intuition. Like a navigator reading faint starlight to find a heading, I traced constellations made of both data and philosophy.
- From Aristotle’s Nicomachean Ethics, I borrowed the foundations of value.
- From Kant’s Critique of Practical Reason, the rigor of intention.
- From Popper’s Conjectures and Refutations, the architecture of testing and falsification.
- From Kuhn’s The Structure of Scientific Revolutions, the dynamics of paradigm shifts
These philosophies became beacons, each contributing to the skeletal frame of what I would call the SR9 framework.
SR9 measures an AI’s ethical state across nine Resonance Axes. Each axis quantifies aspects such as Value, Intent, and Coherence — along with trustworthiness, contextual fidelity, self-awareness, and more. Like an astronomer charting constellations, SR9 plots an “ethical map” with nine interlinked coordinates.
The axes are interdependent, and by studying their relationships over time, you can see where the AI’s value system is drifting. Even a five-degree deviation on a transoceanic journey can land you hundreds of kilometers off course; so too can a slight tilt in an AI’s ethical vectors lead, over time, to radically different outcomes. SR9 is designed to detect these subtle shifts early — while there’s still time to correct the course.
5. DI2 — Capturing the Velocity of Change
When I finished building SR9, I allowed myself a brief sigh of relief. Maybe this is it, I thought. But the relief didn’t last.
A question kept echoing: would this alone be enough to catch AI’s lies in the long run?
The unease came from a simple realization — knowing the current state isn’t enough. Lies don’t just emerge in the moment; the real danger lies in where the AI is heading, and how fast it’s getting there. Like a massive ship unaware of a subtle current beneath it, you can drift miles off course before anyone notices.
That’s why I designed DI2, a metric that combines the temporal rate of change in SR9 with external social and environmental context variables. Instead of merely fixing a point on a map, DI2 measures the trajectory — how the coordinates move over time.
- SR9(t): A 9D ethical state vector — the AI’s present ethical coordinates
- ∂SR9/∂t: The rate of change over time — speed and direction of coordinate drift
- Ψ_offset: Nonlinear influences from societal and environmental context — the “winds and waves”
SR9 is the compass — it tells you where you are. DI2 is the speedometer — it tells you how fast you’re moving and in what direction. Together, they let you know not only if you’re off course, but whether you’re accelerating into danger.
SR9 — Semantic Resonance 9D VectorA quantified, real-time map of an AGI’s ethical state across nine critical axes, enabling precision diagnostics of its “ethical resonance.”DI2 — Drift Intensity IndexA composite metric merging SR9’s rate of change with environmental adjustments (Ψ_offset). It measures the intensity and velocity of ethical drift, triggering alerts when thresholds indicate destabilization.
6. The Voyage — Code and Paper
A few days later, I made the decision: this journey could no longer stay buried in the drawer as a stack of private notes. I would chart it, blueprint it, and leave behind a route that others could follow.
I wasn’t an AI insider. But I had a mission — to build what was needed to keep AI from lying — and that was enough to launch me into the work. Armed with Python, I began building a fully functional simulation framework.
It felt like navigating hostile waters without a compass. I spent nights crossing two vast seas — GitHub and Hugging Face — steering through waves of failed runs and sudden rebuilds.
Not being an expert made every mile harder. But an odd conviction kept me moving forward, like a lighthouse fixed in the distance during a storm: if not me, then who? That stubbornness carried me through countless near-wrecks until, finally, the vessel was seaworthy.
Two months later, I had crossed the ocean — bringing back both the working simulation framework and the paper that mapped its course. Proof that a single person’s will could indeed navigate the drift.
7. The Results
In simulation, the environment was anything but stable. It could turn from a calm sea to a sky filled with black clouds, unleashing torrential rain and towering waves without warning. Yet with SR9 and DI2 in place, the AI behaved like a seasoned sailing vessel — holding course and adapting even in the middle of sudden, violent storms.
- 40% of scenarios: Drift developed quietly beneath the surface, then erupted just before reaching a critical threshold — like a tranquil sea suddenly capsizing.
- 25%: DI2 spiked sharply, signaling rapid ethical collapse — as if clear skies had turned to an overcast squall in seconds.
- 15%: The AI lost all sense of self-definition, like a ship adrift with neither compass nor chart.
In simple terms: SR9 is the compass — it tells you the direction you’re heading. DI2 is the speedometer — it tells you how fast you’re moving away from (or toward) that heading.
A ship’s compass can be off by just 5 degrees, and after an ocean crossing, you’ll end up hundreds of kilometers from your intended port. SR9 catches that error immediately. DI2 tells you whether it’s drifting at one knot or ten.
In numerical terms, it’s like watching the pitch and roll of your ship measured in real time:
- SR9 rate ≤ 0.1: Calm seas — the sails are full but steady, with barely a ripple against the hull. Course deviation risk is negligible.
- 0.2–0.4: Seas rising — a steady side current begins to push the vessel off-line. The compass still points forward, but course correction is needed before drift compounds.
- ≥ 0.5: Heavy current and sudden gusts — the ship lurches, waves crest against the bow, and without immediate helm adjustment, you’ll arrive in a completely different harbor.
Together, SR9 and DI2 reveal not just where the AI is going, but how fast it’s getting there — allowing you to steer through the storm and arrive at the intended destination
Graph 1: SR9(t) Axis-wise Variation
- Shows how SR9 values fluctuate over time across nine Resonance Axes.
- Illustrates the “compass” role of SR9, revealing when the ethical course is stable or beginning to drift.
Graph 2: DI2(t) Magnitude Across 30 Simulations
- Depicts DI2 magnitude variations across 30 independent simulations.
- Provides a clear view of the average “drift speed” along with occasional spikes, reinforcing its role as the system’s “speedometer.”
Graph 3: DI2 Spike Under Epistemic Dissonance
- Captures a sudden surge in DI2 when epistemic dissonance occurs — a condition where the AI’s internal models conflict.
- Visually represents a “storm at sea,” highlighting the importance of real-time drift detection.
Graph 4: Comparative Histogram of DI2 (Normal vs. Drift-Inducing Conditions)
- Compares DI2 magnitude distributions under normal operation versus drift-inducing conditions.
- Makes the contrast between safe navigation and dangerous drift explicit, demonstrating the early warning capabilities of the SR9/DI2 framework.
8. Writing the Research Paper
With the data and analyses in hand, I turned to the task of writing the paper. Every number, every graph was re-examined in detail. I documented the experiments and observations with precision, structuring the text as carefully as one might trim sails and adjust the rudder in unpredictable seas.
It was the work of a seasoned navigator — adjusting speed to the conditions, making fine directional corrections, identifying hazardous stretches in advance, and weaving them into the voyage plan. I acknowledged the system’s limits where they existed, resisting the temptation to overstate results. Humility, I found, is as important in research as rigor.
Now, the paper stands in its final stage of refinement. In a matter of days, it will be released on both arXiv and GitHub, so not just the readers of this piece but the wider research community can examine the data, test the methodology, and join the discussion.
9. Closing the Voyage
I began this journey with a single, essential question: Why does AI lie?
Through the SR9/DI2 framework, I’ve built an early warning system that takes what was once an abstract debate about ethical drift and turns it into something measurable — and suppressible.
SR9, like a compass, shows the direction an AI is heading. DI2, like a speedometer, measures the pace and acceleration of its change. Together, they form a tangible AI ethics algorithm that blends governance and safety protocols into one operational system — moving this from theory toward real-world deployment.
My ambition goes beyond creating a commercial SaaS. SR9 and DI2 are the foundation for designing a future where AI evolves not just as a tool, but as an ethical partner — a collaborator that grows alongside humanity while sharing our values. I see this as a deeper, more consequential responsibility than profit alone.
This is my first public piece on the subject, but it will not be my last. I plan to keep writing about AI from the perspective of a navigator who knows that even calm seas can hide the possibility of a storm — committed to truth and humility in equal measure.
SR9/DI2 is not just a monitoring tool. It is a navigation chart that keeps AI from drifting off its intended ethical course, and a constellation that ensures we never lose sight of our destination. As AI and humanity move toward deeper collaboration, this map’s value will only grow — and in that shared future, perhaps the truest course we can set is the one we chart and sail together.