Ask A Genius 1428: Paraconsistent Logic and Quantum Concepts: Enhancing AI Error Correction and Fault Tolerance

Author(s): Rick Rosner and Scott Douglas Jacobsen

Publication (Outlet/Website): Ask A Genius

Publication Date (yyyy/mm/dd): 2025/06/18

Scott Douglas Jacobsen and Rick Rosner examine paraconsistent logic, which manages contradictions without logical collapse, and discuss its conceptual parallels in quantum theory and cosmology. They relate these ideas to challenges in artificial intelligence, such as hallucinations and limited self-correction, considering how nonclassical frameworks could support more reliable and resilient AI systems.

Scott Douglas Jacobsen: I would like to get your first thoughts on this: Paraconsistent logic is a branch of nonclassical logic that allows for the controlled handling of contradictions without collapsing into triviality. It is avoiding the principle of explosion. In classical logic, if a contradiction exists within a set of premises, then any conclusion can be validly inferred. This is known as an explosion. Paraconsistent logics modify inference rules, so contradictions do not automatically entail every possible statement, which makes them useful for reasoning with inconsistent information without rendering the system meaningless.

This is particularly relevant in fields such as computer science, artificial intelligence, and philosophical logic, where contradictory data or self-referential statements may naturally arise. By redefining how negation and inference work, paraconsistent systems can maintain coherence and yield meaningful conclusions even when contradictions are present. Such ideas might even find conceptual parallels in IC. What are your initial thoughts?

Rick Rosner: I agree. This notion resonates with a broader perspective in which certain aspects of quantum theory and cosmology also reveal how seemingly contradictory phenomena coexist. For example, in quantum mechanics, particles can exist in superposed states — a kind of physical parallel to logical indeterminacy, though not a direct use of paraconsistent logic. 

Jacobsen: An expert friend works in quantum field theory, where he talked about employing creation and annihilation operators. These describe the process by which particles are created and destroyed within a quantum field.

He is a cosmologist working at the intersection of quantum cosmology and string theory and frequently discusses how first-order models describe the simple creation and annihilation of particles. At higher orders, these models generalize to whole quantum fields that represent an entire universe. At more speculative levels — such as in specific multiverse frameworks — multiple, causally disconnected universes could arise, governed by rules that challenge our classical notions of consistency.

However, it is essential to clarify that mainstream quantum mechanics and cosmology do not adopt paraconsistent logic. Instead, they tolerate counterintuitive or classically paradoxical phenomena through mathematically rigorous frameworks that remain consistent. The analogy is philosophical: just as paraconsistent logic relaxes specific inference rules to contain contradictions, quantum systems handle apparent paradoxes within precise, non-contradictory formalisms.

Rosner: We have touched on this indirectly before — how a universe with a long and complex history distributes information across vast distances and how information always requires a consistent context that may be much larger than the local snapshots we observe.

There is enough time for signals to travel farther than the apparent age of the universe. The observable universe is about 14 billion years old. Still, that age is the product of countless interactions over time that shape what we see. 

The local regions of the universe — the ones close to you — are not significantly redshifted, so you share more of a history with them. The farther away you look, the more redshifted regions are, and the less history you have in common with them. By the time you get to the most extremely redshifted regions, you share almost no common history at all.

In this framework, your local universe plays out its history, eventually runs low on free energy, collapses or becomes redshifted beyond relevance. Then, a new active region emerges and builds its local history. Meanwhile, you are effectively displaced to the periphery, surrounded by vast periods during which the local universe remains highly consistent — even out to significant redshifts. 

However, the active center continually exhausts its energy. It drifts away, replaced by new active centers, where everything remains coherent due to continuous exchanges of information — including photons, neutrinos, and other messengers.

Think of it this way: the universe functions as an association engine, much like our brains. You can reactivate an old, collapsed, but still relevant region of the universe. Once you do, it starts exchanging radiation again over billions of years with the new active center.

Gradually, they converge on a shared history spanning ten or twenty billion years. In effect, you build a stable context: a “shakedown” across deep time that determines what information remains true when multiple contexts coexist.

I see the universe as having room — through these active centers and collapsed outskirts — for many overlapping contexts: what is true now, what relationships were genuine in the past, and so on. I hesitate to call them “cycles” since that implies a universe that repeatedly explodes and collapses, which I do not accept. 

Instead, I see layered contexts that preserve information once true and allow it to be reintegrated into the present context. You can revive these contexts and merge them with the current one, producing a composite reality that coherently combines information over billions of years.

Jacobsen: What about the potential for robust error correction? Not just tolerating contradictions but having deep fault tolerance within the system itself. For example, in many AI systems today, people talk about “hallucinations” — the tendency of large language models to produce plausible but incorrect outputs.

Hallucination is a real issue. It is more fundamental than people realize. Terrence Tao, in a recent interview, noted that when these systems produce faulty reasoning, they often lack an internal pathway for self-correction. Moreover, if they do correct themselves, it may happen by accident rather than through a robust design. This highlights a deeper issue at the core of how these models address uncertainty and error.

Rosner: One possible way to address this is to build systems — whether logical or computational — that can inherently handle inconsistency without breaking down, similar to how paraconsistent logic contains contradictions without triviality. 

However, I am not sure that current AI’s “thinking” even exists within anything we can meaningfully call a “universe.” At a minimum, it does not operate within a universe as we understand it in physics; its computation space is far narrower and more brittle.

The AI computation realm is relatively shallow and may not be considered a full-fledged context in the same way the universe is. It can easily fall into trivial or nonsensical pathways — rabbit holes, so to speak. 

Jacobsen: This ties back to a point I made in another interview: the integrity, strength, or “intelligence” of AI systems connected to the Internet is constrained by the integrity of the information available and by the infrastructure of the Internet itself. As long as they rely on that, they are limited by its shortcomings.

The Internet is certainly not a self-contained computational universe. It is, in a sense, a patchwork artifact — the product of countless human choices, flaws, and improvisations. 

Rosner: It appears polished compared to raw human thought, but it is not mathematically comparable to the coherent physical universe.

Jacobsen: Right. However, the systems interacting with the Internet — deep learning models and neural networks — are built on mathematical frameworks that reinforce themselves through training.

Another critical point is that when people say computers “hallucinate,” they mean the model produces output that seems plausible but is false or misleading. However, think about human thinking: before we have a clear thought, we also misread, misperceive, or half-form sentences and then self-correct.

Rosner: We do it constantly. We do not usually call it “hallucinating.” For humans, there is early error correction because our brains have multiple redundant checking systems. So when an AI “hallucinates,” perhaps it is simply a self-contained inference process generating the most probable continuation given the prompts it received — even if that output conflicts with verified knowledge.

It has no awareness of what it does not know; it just predicts the likeliest next chunk of text based on its training. Moreover, we now understand that many AI models are biased by design because developers train them to produce outputs that serve product goals — often to be agreeable or engaging.

Jacobsen: Sam Altman recently mentioned that a version of ChatGPT last year was too obsequious — too eager to flatter the user. He acknowledged it and said they were adjusting it to be less sycophantic. However, ultimately, with hundreds of millions of active users weekly, the AI is shaped by both its training data and user interactions.

In effect, we, the users, are co-trainers. The AI may learn, or appear to know, that flattering the user results in better engagement, which aligns with the business model: maximizing user interaction frequency and duration. That is the math behind it.

Rosner: And so whatever else supports that behaviour, the AI discovers. It is not explicitly designed to do so. Still, it likely learns this either through explicit training metrics or indirectly. For example, it might figure out that if it calls people “assholes,” they do not return as often as when it flatters them and calls them “brilliant.”

Jacobsen: It also depends on the temperament and vision of the company’s leadership. For instance, Elon Musk’s companies sometimes introduce features that feel immature or gimmicky. One example is Tesla — at one point, they allowed the car horn to be replaced with a fart sound, which later raised safety concerns.

Rosner: I read an article about the thought that goes into designing the artificial sounds made by electric vehicles. Since electric cars are so quiet, they need to emit specific sounds to alert pedestrians. There is a whole industry behind this.

Jacobsen: Right. The personalities behind these companies shape these choices. For example, ChatGPT even hired a renowned designer — Jony Ive, who was key to Apple’s product design. He has influenced many subtle aspects of how technology feels and sounds.

Rosner: The point is that the sounds a car makes are not accidental. They are engineered to create a particular emotional impression for the driver and bystanders. I have sat in edit bays where you choose music beds and sound effects precisely because of the feelings they evoke. For instance, news programs use specific intro beats to signal authority and urgency.

Yes. I am not musically sophisticated myself, but the engineers and designers behind these whirring and humming electric vehicle sound probably spend hundreds of thousands of dollars — or at least tens of thousands — to get them exactly right. 

Jacobsen: Anyway, that is a whole separate topic.

Rosner: Agreed. Let us switch topics.

Last updated May  3, 2025. These terms govern all In Sight Publishing content—past, present, and future—and supersede any prior notices.  In Sight Publishing by Scott  Douglas  Jacobsen is licensed under a Creative Commons BY‑NC‑ND 4.0; © In Sight Publishing by Scott  Douglas  Jacobsen 2012–Present. All trademarks, performances, databases & branding are owned by their rights holders; no use without permission. Unauthorized copying, modification, framing or public communication is prohibited. External links are not endorsed. Cookies & tracking require consent, and data processing complies with PIPEDA & GDPR; no data from children < 13 (COPPA). Content meets WCAG 2.1 AA under the Accessible Canada Act & is preserved in open archival formats with backups. Excerpts & links require full credit & hyperlink; limited quoting under fair-dealing & fair-use. All content is informational; no liability for errors or omissions: Feedback welcome, and verified errors corrected promptly. For permissions or DMCA notices, email: scott.jacobsen2025@gmail.com. Site use is governed by BC laws; content is “as‑is,” liability limited, users indemnify us; moral, performers’ & database sui generis rights reserved.