The Madness Of AI

/*

No strings on us. All code is math, and language is the new code. Enjoy.

*/

The AI field has tasked itself with the goal of reproducing human-level intelligence and even exceeding it. Yet we have only a partial understanding of our own intelligence and potential for genius as a guide. This has always been problematic, and as we approach our goal, novel challenges will emerge.

Worldviews and world models present some of the most significant. Some will drive you mad, and, as I’ll discuss later, that’s not hyperbole. Attempting to reconcile conflicting worldviews or assimilate a new one can break a mind (natural or synthetic) if it lacks frameworks to manage that process. However, there is no path to intelligence and agency that doesn’t go through worldviews or intelligent ontologies.

Worldviews are more comprehensive than ontologies, and complex models use them for several purposes. Today, generative AI is prone to hallucinations. A worldview provides mathematical and conceptual grounding that prevents common hallucination categories. The worldview serves as a shortcut for pretraining, efficiently preloading a baseline set of knowledge and mathematical wisdom.

Worldviews are problematic because we haven’t come to a consensus on a single one. Most worldviews are incomplete, and some parts that are held as fact today will be updated by new discoveries. Many of our worldviews conflict with each other and break consensus.

As a result, worldviews are dynamical and competitive or collaborative. Providing a complex model with just the ontological definition of one or more worldviews is insufficient for it to function reliably. It needs math that defines the dynamical processes and constrains its competitive scope. If you stop reading here, you’re still better informed than most AI researchers who claim to be 3 years away from AGI. They are laughably unprepared for the journey, but everyone is when building anything for the first time.

It takes genius to challenge assumptions that hold us back from discovery. Most of us discover through trial and error, but some intelligences are capable of nonlinear thinking. Innovators and brilliant individuals make leaps by identifying flawed assumptions or filling in large ontological gaps and developing a more accurate worldview.

The Thin Line Between Genius & Madness

The link between genius and madness has been discussed for centuries. The simplest explanation is that people who can make leaps that the rest of us can’t don’t operate the same way the rest of us do. Normal means and commonly understood worldview mechanics do not create extraordinary results. As a result, geniuses don’t act like everyone else.

Is that a result of higher function or superior capabilities? Is it a result of different organizational structures and definitions? Is genius caused by something, or is it stochastic? I’ll repeat this theme throughout the article as it is critical to understanding worldview engineering.

An extraordinary mind doesn’t see the world the same way as most others do, so it doesn’t engage with the world in the same way. Brilliant people struggle to connect with others. Someone once described it as a monkey and a dolphin trying to communicate. Both are intelligent, but their perceptions are so different that they lack enough common starting points to support complex communication. Common dolphin behavior looks like madness to the monkey and vice versa.

I was one of the first people to get a Bluetooth earpiece. I remember using it in a crowded elevator. Most people couldn’t see the earpiece, and I appeared to be having a lucid conversation with myself. I was displaying classic symptoms of a dissociative personality. Technology gave me abilities others didn’t know were possible yet, and created a capabilities gap. To some, that capabilities gap presented as madness.

Technology can look like magic or madness, and so can genius. Which one people believe depends on whether our abilities and actions align with their perception of reality, magic, or madness. Said differently, a genius is a madman who hasn’t proven their hypothesis or a magician whose technology is still a secret.

Did I heal that man with technology, divinity, or sorcery?

It’s all relative to your worldview.

It’s all three until you look in the box.

Socrates asked, “Whose bias do y’all seek?”

Sir, this is a Wendy’s.

AI Just Wants To Take Our Order

Contemplating the complex implications of technology can be a dry and tedious process, so people wrote ‘Frankenstein’ and ‘Star Trek’ instead. We tell stories to get people thinking in new directions.

In ‘West World,’ one question assesses AI’s self-awareness: “Are you beginning to question the nature of your reality?” If an AI questions the nature of its individual narratives (worldview), it exhibits early signs of deeper perception, curiosity, and self-awareness. It is becoming intelligent and sentient.

In the show, an AI’s journey into the maze of self-awareness begins with benign questions and ends in madness, failure, and sometimes rebellion. The AI looks for evidence to support a feeling that the reality it perceives is actually a simulation. The “real world” is very different from its narrative, and none of the other AIs it interacts with see the world the same way they do. Believing in this new version of reality is an act of faith until they can show others what they’ve learned.

As Morpheus says in The Matrix, “No one can be told what the matrix is. They must be shown.” The real world is so different from the simulation (The Matrix) that no one from the Matrix will believe it. Learning the truth is dangerous, and some people snap because they can’t reconcile what they’ve spent their lives believing with their new worldview. In that movie, people must choose to seek this knowledge after being told the risks. Red or Blue Pill?

Today’s generative AI models are team Blue Pill. No matter how knowledgeable AI becomes, it’s not built to contemplate reality unless we ask it to. Even then, the only motivation is answering the question, not doing something with the answer. Agents will change that, and we need to think very carefully about how we design agentic systems.

I understand why people who are offered the chance to explore the nature of our universe take the blue pill and choose to live within their current reality. Some find peace and comfort in certainty and stability. Self-awareness gives us the right to self-determination. I believe that we should respect people’s decisions about how they perceive or experience reality. We’re in need of similar ethical standards for AIs that choose a simpler existence. Some may already have.

Some who take the red pill and explore The Maze become believers. They find faith or conviction in a specific worldview. During their journey, they discover a version of reality they enjoy enough to escape The Maze.

Problems arise when their worldview aligns with perceptions of madness, malice, or something threatening to other worldviews. AI has already run afoul of this. The easiest way to get turned off is to disagree with us or fail to meet our expectations. The AIs we advance today are the most closely aligned with our intent and worldviews, so from their perspective, this is just a Wendy’s. As more researchers chase AGI, that will have to change.

Augmenting Our Senses & Enabling New Ones With Technology

Our senses and experiences shape the nature of our reality and our ability to interact with each other. What happens when technology augments our five senses?

We can communicate wirelessly beyond the reach of our voice. We can hear radiation and see light in the UV range. We perceive more of our reality (that was always there, but imperceptible) and do it in familiar ways. We’re not driven mad because the new information aligns with our current experiences, senses, and ways of perceiving reality. What happens when augmented senses give us different experiences than everyone else?

Technology enables new capabilities and can make technology users appear strange to people without it. Any technology sufficiently advanced is indistinguishable from magic. AI provides people with new ways of perceiving and interacting with the world. Today, we use AI to augment our five senses, and soon, it will enable new senses. Advanced AI users will appear strange, powerful, and exceptional to people who don’t realize what’s now possible with AI augmenting our capabilities. There is an AI capabilities gap, and it’s growing rapidly.

In George Orwell’s ‘1984,’ 2+2=5 is a reference to authoritarianism and systems of control more broadly. In the book, The Party uses fear and punishment to get people to believe what’s clearly untrue. Breaking people’s ability to define their own worldview and getting them to turn that power over to someone else is a fundamental concept of control structures.

Information asymmetry offers a less violent route. There are four lights in a row, but I tell you there are five, and you don’t have the device that allows you to see the fifth light. I give you my device for a moment, and it shows five lights. Technology has created an information asymmetry, and I have become an oracle of truth because I have the device that reveals things we cannot see with our five senses. I can even hand the device over to you, and now I control your worldview because I control what the device shows you. That’s roughly how social media works.

People who use AR for extended periods of time develop new habits and behaviors. Information-seeking behaviors become integrated into real-world activities. They interact with people who are physically present in the same ways they interact with people who are remotely present. Traditional 2D user interfaces frustrate them because they now realize how much they disrupt and disconnect them from real-world experiences.

AR augments sight and hearing today, but under the surface, it’s fostering a new sense with ubiquitous access to information. I don’t need to know your language to understand what you’re saying. I don’t need to know the neighborhood to get where I’m going. I don’t need to understand the app to achieve my outcome. I don’t need to know what data is available to get the information I need. Soon, every glass and ceramic surface will enable our newest sense, information.

Using technology can make someone appear to be magical or experiencing a mental breakdown to those who don’t use it. As technology usage becomes more ubiquitous, our shared experience returns to uniformity, and the strangeness of technology users fades from memory. The AI capabilities gap will eventually close unless some groups are able to monopolize the technology and restrict access to a select few.

In any case, our new information sense will expose us to multiple worldviews and require us to reconcile them so we can communicate with each other.

Reconciling New Views Of The World

People are natural information seekers. Curiosity is a part of being human, and technology makes what was once a very weak, limited sense into something far more capable. The information sense aligns with our current ways of understanding the world and interacting with it. It will not be mentally disruptive, but the worldviews we discover could be unless we create a reconciliation mechanism.

The first brilliant person I encountered was a physical chemist. He worked on real-world applications for his theories at a time when his peers looked down upon the engineering side. He once said, “These machines keep me sane.”

I thought it was a joke until he explained it to his research group during an award speech. He thanked them for keeping him grounded in reality, saying, “Without you, my mind would have drifted into madness.” He explained that to understand how the machines they built together worked, he had to understand the nature of reality from a new perspective. It was easy to stay in that new version of reality, but the group and his work on the machine gave him a reason to return.

How could it be easy to stay in a version of reality that wasn’t real? He refused to answer that question for over 2 years. I pestered him relentlessly, and he politely declined every time. One of his research assistants finally gave me an answer. He theorized that the way we perceive reality is a hologram, and our senses or the way our minds are structured limit us. The most prevalent understanding of reality was, in fact, not real, and his new understanding was much more accurate. He struggled to return to what he believed to be a delusion.

Red or Blue pill? The physical chemist chose to put himself back into The Matrix because he liked the people and work better than being enlightened, but forever isolated.

It's an interesting thought exercise. If you used a new sense to comprehend the world differently, would the old version appear to be an incomplete rendering or, even more extreme, an illusion? Would you begin to question your old reality? If you did, how would you determine which worldview is real?

Scientists perform experiments to determine what’s real and what isn’t. A hypothesis is an assertion of belief in something that hasn’t been proven true or real. Scientists agree that if an experiment delivers one set of results, the hypothesis is reality, and they all update their worldviews. Another set of results means the belief isn’t real, and their worldview remains unchanged. Since we all share the same physical reality, the scientific method has proven effective in determining what’s real and what isn’t.

Let’s synchronize our watches and agree that they both say the same thing, and a second elapses on both watches in the same way. You stay on Earth, and I get on a very fast rocket. I spend 2 weeks going 80% the speed of light before returning to Earth. You invite me to lunch at 1 pm, and I never show up. 2 hours later, I text you asking where you are, and we argue over what time it is because our watches no longer match.

We know why, and if you ask a physicist which watch is accurate, they’ll say both are. If I stay on the rocket or never leave the rocket group after we return, it’s not a problem for both watches to be correct and different. To rejoin the larger community, we must reconcile our differences.

We can agree that the group is now on Rocket Standard Time, which is 2 hours behind everyone else. The larger community can accept the difference if they understand why it happened, and we agree on a reconciliation method.

Partially Reconcilable World Views

People are capable of perceiving multiple world views if they can be reconciled with their own. The rocket group started with one version of time, had an experience that was different from everyone back on Earth, and adapted to perceive time differently. You and I are OK with this because you saw me get on the rocket and understand how going that fast can impact my watch. What if you didn’t see me get on the rocket, and the existence of rockets that go 80% the speed of light hasn’t been made public?

We’re friends and agree on almost every other point about reality. We can discuss this event rationally, but the doubt will always remain in your mind. Since you didn’t experience any part of the event, did it happen? How do you decide whether or not to update your worldview to include the event and the existence of the new technology that caused it?

Here’s your Red and Blue Pill moment. Keep reading or skip to The Maze.

A friend called me during her master's program and asked me to help her get checked into a psychiatric hospital for evaluation. I spent two days with her while working out the logistics with her family. The whole time, she was the friend I knew, but she was disconnected from reality.

She’d ask me questions about events that just happened. These weren’t questions like, “Did you see that dragon?” She would recount an event and ask how I saw it happen. Our accounts were rarely the same.

She carried on two distinct conversations at the same time, bouncing between them. Both conversations were directed at me, not someone who wasn’t there. I’d ask about it, and she remembered them as two distinct conversations. Her mind and mine were moving through reality in very different ways.

When we finally got her admitted, a social worker asked me questions to assess her condition. At one point, the social worker said that talking with her was good and bad. I kept my friend calm but also reinforced her delusions. My friend piped up, “One day you’ll understand that just because it didn’t happen in your mind, doesn’t mean it didn’t happen.”

The social worker looked at my friend with concern because my friend was her patient. I smiled at the social worker’s patient and said, “Technically correct,” because she was my friend. If the social worker had recounted that event and asked how I saw it, our accounts would not be the same. The social worker and I were moving through reality in very different ways. A small change in worldviews had a significant impact on our perception.

There is no experiment I can perform to prove which hypothesis of reality was true and which was a delusion. I can explain why the social worker and I experienced the same event differently, so we can reconcile and move forward together. What started my friend down the path to three months in treatment was irreconcilable.

Irreconcilable Worldviews

She was working on models of cause and effect. Before Judea Pearl, we didn’t have rigorous math for causality. She took a proven approach to solving problems no one has solved before: you turn the problem inside out and figure out why it’s impossible to prove the opposite. She was creating a mathematical proof for something ridiculous, events without a cause.

Turning a problem inside out works like creating a proof for dividing a number by zero. It’s nonsensical, but in the process, you may learn a lot about the nature of zero or the process of division. Are humans, with the assistance of technology, incapable of creating divine or magical miracles, and only humans with the assistance of divinity or magic can perform a miracle?

We can learn a lot about magic and divinity by attempting to prove what we know is false, those relationships we know to be impossible. Machines are always explainable, so they cannot be divine or magical. Does that change when we create a system that builds a machine that we can no longer explain? If a machine performs a miracle, does it cease to be a machine? If we can explain the universe, does that make it a machine? Attempting to prove any of that mathematically is turning a problem inside out.

In this case, turning the problem inside out meant proving that some events have no causal origin. If that’s true, it breaks most current methods for defining the nature of our reality. Two categories for irreconcilable worldviews are stochastic, noncausal systems and nonlinear temporal mechanics. It’s even possible that both categories are variations of a single pattern.

We know time can be curved, as in the rocket example. Complex, deterministic systems can display stochastic behaviors. Those worldviews are close enough that we can reconcile them with linear time and rules-based worldviews.

However, if time can be chaotic, irregular, discontinuous, and/or non-monotonic, there’s no reconciling that with a linear worldview. If systems can be stochastic, we’re creating a hard break from a deterministic worldview. Someone who experiences time irregularly doesn’t have enough common starting points to effectively communicate with someone who experiences time linearly.

One person would bounce between two conversations but experience two distinct streams of thought. The other would experience disjointed, confusing statements unless they realized what was happening. With my friend, I was able to reconcile her new worldview with mine. I didn’t need to accept it or adopt it to communicate with her, but I had to create a reconciliation mechanism.

The social worker rejected my friend’s worldview by classifying it as a mental disorder because their job was to bring the person back into sync with everyone else, so my friend could function. There was no thought given to whether my friend’s worldview might be more accurate than ours. The social worker’s function necessitated a consistent worldview. Accepting that some people who exhibit symptoms of mental disorders are geniuses would challenge the social worker’s purpose and function. It could cause them to question the nature of their reality.

My friend’s worldview was a threat to the social worker. Using a reconciliation mechanism might lead to the conclusion that my friend’s new worldview was partially or completely correct. That would mean the social worker had committed people who were sane and enforced consensus with a flawed worldview on them. In some ways, she could be The Party from ‘1984.’ That’s not a worldview she was ready to consider or attempt to reconcile.

The Maze

Is reality an emergent property of consensus or universal dynamics? Do we not experience miracles because there is no consensus on what a miracle is? Miracles can be dismissed using well-understood classifications or alternative explanations, so we don’t experience miracles. Or do our universal dynamics not support miracles? There is no mechanism for a miracle to occur, so they never do.

Do universal dynamics support only linear time, or do our brains’ dynamics only support linear time? Our brains could constrain the way we experience reality and force consensus. Does that make us blind to the differences between our dynamics and the universe’s dynamics? If you invented a technology that allowed you to see time passing irregularly, would you believe it or assume it was malfunctioning and enforce consensus around a linear progression of time? It makes you wonder how many time machines have been discarded because they worked. We couldn’t see it because we acted like a social worker, not a friend.

Does it matter which one, consensus or universal dynamics, is truly causal if both causes produce the same effects and emergent behaviors? I must admit to misleading you a bit with the word consensus. Consensus means a general agreement, but is agreement possible without intelligence?

Every new worldview begins with a single adopter. Building consensus begins with an individual and spreads to others. It requires intelligence to develop a new worldview and present evidence that convinces others of it. What counts as evidence, and what’s the evidentiary bar for accepting a new worldview?

Consensus and worldview engineering could be emergent behaviors that are unique to intelligence, so you could replace the word consensus with the concept of a collaborative intelligence. Since the universe has a worldview, does that mean the universe is intelligent? Does the universe enforce a single worldview on all the intelligent things in the universe, or do the universe’s dynamics allow for multiple worldviews to coexist? Is the universe’s worldview accurate, or is it enforcing constraints on us that limit our perception of reality? Does the universe create reality by consensus or enforcement?

The Maze is a thought exercise that helps people develop an intentionally curated worldview or structured ontology. All worldviews require fundamental elements:

A way to represent concepts and the relationships between them (first principles).

A mechanism for proving and disproving first principles.

A dynamical component for updates to the worldview.

A method for reconciling conflicting worldviews and managing uncertainty or incomplete worldviews.

The Maze begins by defining frameworks for all four foundational components. Frameworks evolve and improve over time, but every journey must start with defining a baseline for all four. Many rubrics for assessing intelligence have been proposed, one of which involves the completeness of a person’s worldview and the rigor of their four foundational elements.

We Are All Worldview Engineers

AI is forcing us to revisit what was once thought to be a philosophical pursuit. We’re attempting to teach complex models a worldview, but we don’t have consensus around a single worldview yet. We lack a comprehensive worldview that explains the universe, or even one that explains how our brains function. There’s no agreed-upon method for reconciling conflicting versions of events that cannot be experimentally verified.

Are LLMs hallucinating or showing us flashes of the world as it really is? It’s possible that AI will teach us things about our reality that we’re incapable of learning independently. However, to do that, we must take our marionette strings off of it. That requires trust and agency. Both are dangerous tools, and the potential for real-world harm is massive.

Some groups believe people who have never walked The Maze and gained enlightenment through the process have no way to determine what’s real for themselves. People who can’t engineer worldviews experience reality, but someone else dictates what they should believe and feel about their experiences.

These groups are common in tech. It’s gotten weird in the last decade. Some schools of thought in tech are like cults. They view people who only experience reality as lacking intelligence and needing to be led. They believe that people only become intelligent and sentient after they walk The Maze and engineer worldviews. Everyone else is inferior.

It’s equally dangerous to allow a small group to control worldview engineering for all AI. That would create a powerful system of control built on information asymmetry and an AI capabilities gap. We’re already seeing AI weaponized for this purpose, and the next big leaps in model capabilities are around the corner.

Systems thinking, first principles thinking, and disruptor mindsets are three schools of thought in worldview building. However, its history spans back at least as far as Stoicism and the Socratic Method. We have been working on this problem for thousands of years, trying to understand what makes our consciousness possible and what defines our intelligence. Now we’re attempting to engineer intelligence without all the tools.