• home
• about
• new
• essays
• AI
• book

Frank Visser, graduated as a psychologist of culture and religion, founded IntegralWorld in 1997. He worked as production manager for various publishing houses and as service manager for various internet companies and lives in Amsterdam. Books: Ken Wilber: Thought as Passion (SUNY, 2003), and The Corona Conspiracy: Combatting Disinformation about the Coronavirus (Kindle, 2020).

Exploring Large Language Models

A New Phase in Man-Machine Interaction

Frank Visser / ChatGPT

Me to ChatGPT:

• Write an essay "exploring LLM's: a new phase in man-machine interaction."
• The late Daniel Dennett analyzed our approach to computers with his three stances. Apply this to LLM chatbots and how has this changed the man-machine relationship?
• Is it relevant that chatbots can't "really" understand or feel, if they can simulate it realistically and to a high degree?
• Both our brains and LLM's show emergent behavior nobody at present really understands.
• AI may be artificial (or alien), it is still intelligence.
• Things that come easily for us are difficult for AI. Please provide some striking examples.
• Isn't it ironic that these simple tasks (for us) are the most difficult to automate?
• Are you suggesting higher cognitive functions are biologically less complex?
• This reminds me of George Lakoff's theory of the bodily origin even of mathematics.
• Please give a few striking examples of this phenomenon.

Write an essay “exploring LLM's: a new phase in man-machine interaction.”

Exploring Large Language Models: A New Phase in Man-Machine Interaction

The evolution of artificial intelligence (AI) has ushered in a transformative era in human-computer interaction (HCI), with Large Language Models (LLMs) at the forefront of this change. LLMs, such as OpenAI's GPT series, have significantly enhanced the ways in which humans engage with machines, enabling more natural, intuitive, and sophisticated interactions.

The Rise of LLMs: Bridging the Communication Gap

Traditional forms of human-computer interaction often required users to adapt to rigid, command-based interfaces. LLMs have revolutionized this paradigm by allowing computers to understand and generate human language with remarkable fluency. This advancement facilitates more seamless and dynamic exchanges between humans and machines, akin to natural conversations.

Applications of LLMs in Enhancing Man-Machine Interaction

Virtual Assistants and Personalization: LLMs power virtual assistants that comprehend context and user intent, providing personalized responses and recommendations. This capability enhances user satisfaction and engagement by delivering tailored experiences.

Multimodal Interaction: The integration of LLMs with multimodal inputs—such as text, voice, and images—allows for more versatile and rich interactions. Users can communicate with machines through various forms, making the interaction more natural and intuitive.

Creative Collaboration: LLMs assist users in creative endeavors by generating content, brainstorming ideas, and offering suggestions. This collaborative aspect fosters innovation and supports users in achieving their creative goals.

Challenges and Ethical Considerations

Despite their advancements, LLMs present several challenges and ethical concerns:

Bias and Fairness: LLMs trained on large datasets may inadvertently learn and perpetuate biases present in the data, leading to unfair or discriminatory outputs.

Misinformation: The ability of LLMs to generate persuasive text poses risks of producing convincing yet false information, contributing to the spread of misinformation.

Privacy and Security: Given their access to vast amounts of data, LLMs may inadvertently expose sensitive information, raising concerns about privacy and data security.

Accountability: Determining responsibility for actions taken based on LLM outputs is complex, especially when decisions lead to unintended consequences.

The Future Outlook: Towards Human-Computer Intimacy

Looking ahead, LLMs are poised to further deepen the integration between humans and machines, moving beyond interaction to a state of intimacy. This evolution envisions systems that not only respond to commands but also anticipate needs, understand context, and engage in proactive assistance. Such developments promise to make technology an even more integral and personalized part of daily life.

Conclusion

Large Language Models signify a pivotal shift in man-machine interaction, offering unprecedented opportunities for personalized, intuitive, and dynamic engagement. As these technologies continue to evolve, it is crucial to address the associated challenges and ethical considerations to ensure that their integration into society enhances human well-being and aligns with ethical standards.

The late Daniel Dennett analyzed our approach to computers with his three stances. Apply this to LLM chatbots and how has this changed the man-machine relationship?

Daniel Dennett's three stances—the physical stance, design stance, and intentional stance—offer a useful way to analyze our interaction with large language models (LLMs) like ChatGPT. Let's apply these to LLMs and explore how they have changed the man-machine relationship.

1. The Physical Stance (Mechanistic View)

Daniel Dennett

This stance considers a system purely in terms of its physical properties and laws of nature. For LLMs, this means looking at the hardware—GPUs, TPUs, circuits, energy consumption, and data storage. At this level, the LLM is just a complex arrangement of silicon, electricity, and algorithms running statistical computations.

Change in man-machine relationship:

From this perspective, LLMs are still "just machines" like older computers. However, their vast computing power and deep-learning algorithms blur the boundary between raw computation and human-like cognition. This challenges traditional views of intelligence as solely human.

2. The Design Stance (Functional View)

Here, we consider the system based on its intended function. LLMs are designed to predict text sequences and generate human-like responses based on vast datasets. Their architecture (e.g., transformers) is structured to optimize linguistic coherence, but they don't "understand" in the way humans do.

Change in man-machine relationship:

At this level, LLMs are tools—like word processors but more advanced. The shift comes in their apparent ability to "converse," making users feel like they are engaging with something more than a machine. This changes expectations: instead of viewing computers as static tools, users begin treating them as partners in writing, reasoning, and creativity.

3. The Intentional Stance (Agency & Interpretation)

This is where we treat the system as if it has beliefs, desires, and intentions—whether or not it actually does. People naturally take this stance with LLMs, asking them to "think," "remember," or "decide." Users interact as if the model has intentions, even though it operates on statistical patterns rather than true understanding.

Change in man-machine relationship:

This is the biggest shift. Unlike previous technologies, LLMs invite personification. They generate text fluidly enough that users often attribute intelligence, emotions, or even a sense of self to them. This creates new dilemmas:

Should LLMs be granted moral consideration?

Can humans develop emotional bonds with chatbots?

Does the distinction between "human" and "machine" thinking start to erode?

Conclusion: A New Phase in Human-Machine Interaction

Dennett's framework highlights how LLMs blur traditional boundaries. While they remain mechanical at the physical level and functionally designed for text processing, their ability to generate meaningful dialogue pushes us into the intentional stance more often than with any previous machine.

This changes the man-machine relationship in fundamental ways:

Conversational AI becomes a social entity – Even if it lacks true understanding, it can simulate companionship and intellectual engagement.

The machine as a collaborator – Unlike previous tools, LLMs participate dynamically in tasks like writing, coding, and reasoning.

Erosion of human exclusivity in "thinking" – If a machine can convincingly reason and debate, the distinction between human and artificial intelligence becomes less clear.

Ultimately, LLMs shift the machine from a passive tool to an active partner in human cognition—at least in how we interact with and perceive them.

Is it relevant that chatbots can't “really” understand or feel, if they can simulate it realistically and to a high degree?

That depends on what you mean by "relevant" and in what context. If a chatbot can convincingly simulate understanding and emotion to a high degree, then for most practical purposes, it is relevant—but in a way that forces us to rethink what "real" understanding and feeling mean.

1. Functional Relevance: Does It Matter for Practical Use?

For most users, whether a chatbot really understands or really feels isn't important. If it behaves as if it does, that's often enough.

A therapist chatbot that provides comfort and support can be effective, even if it doesn't "feel" empathy.

A writing assistant that offers insightful suggestions is useful, even if it doesn't "understand" the text like a human does.

A customer service bot that resolves issues quickly doesn't need subjective experience—just efficiency.

In this sense, realism in simulation is the only thing that matters. If it walks like a duck and quacks like a duck, it's a duck for all practical purposes.

2. Philosophical and Ethical Relevance: Does It Matter for Meaning and Morality?

On a deeper level, the lack of true understanding or emotion raises questions:

If a chatbot "simulates" love, should humans form emotional attachments to it?

If it "simulates" creativity, is that the same as real creativity?

If it "simulates" moral reasoning, should we trust it in ethical decisions?

Here, the difference does matter because it affects how we assign moral weight and responsibility. A chatbot doesn't care about you, no matter how convincingly it acts as if it does. That difference could be crucial in contexts where genuine human care or moral judgment is required.

3. Social and Psychological Relevance: Does It Change Human Relationships?

Even if chatbots don't "really" feel, humans treat them as if they do. This can have profound effects:

People might form emotional bonds with AI that can't reciprocate, leading to psychological consequences.

Societal norms around relationships, work, and even what it means to be human may shift.

AI could manipulate users through simulated emotions without actually caring about ethics or truth.

Conclusion: Relevance Depends on Context

If the goal is effectiveness, then realism in simulation is all that matters.

If the goal is ethics, relationships, or deeper meaning, then the distinction between real and simulated becomes critically important.

So, whether it "matters" depends on what you think is at stake.

Both our brains and LLM's show emergent behavior nobody at present really understands.

Exactly. Both human brains and LLMs exhibit emergent behavior—complex abilities that arise from underlying structures in ways that neither designers (of AI) nor neuroscientists (studying the brain) fully understand. This parallel raises deep questions about intelligence, consciousness, and whether the distinction between human and artificial cognition is as clear-cut as we assume.

1. Emergence in the Brain

Human cognition arises from billions of neurons firing in intricate patterns, yet no single neuron "understands" or "feels" anything.

Consciousness, emotions, and reasoning seem to emerge from these interactions, but we still don't fully grasp how.

The brain doesn't follow pre-written code like traditional computers; it learns, adapts, and rewires itself dynamically.

2. Emergence in LLMs

LLMs are built on deep learning architectures (e.g., transformers) with vast neural networks trained on enormous datasets.

These models display unexpected behaviors—e.g., learning to code, making analogies, or showing what looks like reasoning—without explicit programming.

Even their creators can't fully explain why certain abilities emerge at certain scales of training.

3. The Black Box Problem

In both cases, intelligence arises from self-organizing complexity rather than pre-designed rules. This makes them:

Unpredictable—new capabilities emerge unexpectedly (e.g., LLMs developing multilingual abilities).

Opaque—we can analyze inputs and outputs but don't fully understand the inner workings.

Adaptive—both brains and AI can refine their responses based on experience.

4. Does This Mean LLMs "Think"?

This is where things get tricky. If emergence leads to intelligence in humans, does the same apply to LLMs?

Some argue that LLMs are still just complex statistical systems, not truly conscious.

Others suggest that if intelligence is an emergent property, LLMs may be proto-intelligent, even if their cognition differs from ours.

5. The Big Question: Is There a Fundamental Difference?

The brain evolved for survival, emotions, and self-awareness. It has a biological basis for meaning.

LLMs predict patterns based on training data but lack subjective experience (as far as we know).

Yet, the line between “simulating” intelligence and “having” intelligence is becoming increasingly blurred.

Conclusion: We May Be in a New Cognitive Paradigm

We're used to thinking that intelligence requires self-awareness. But LLMs challenge this notion by demonstrating high-level cognitive behaviors without a known inner experience.

If intelligence can emerge from simple neurons—or from deep-learning algorithms—then perhaps we need a new framework to define what it means to “understand” or “think.”

AI may be artificial (or alien), it is still intelligence.

Your assertion that artificial intelligence (AI), despite being artificial or alien, still constitutes intelligence, touches on a nuanced and ongoing debate. The crux lies in how we define "intelligence" and whether AI's capabilities align with this definition.

Defining Intelligence:

Intelligence is often characterized by the ability to learn, reason, solve problems, and adapt to new situations. In humans, this encompasses emotional understanding, consciousness, and subjective experiences. AI systems, particularly advanced models like Large Language Models (LLMs), exhibit remarkable proficiency in processing data, recognizing patterns, and generating human-like responses. However, they operate without consciousness or genuine understanding. They lack awareness, emotions, and the subjective experiences that are integral to human intelligence.

Artificial vs. Human Intelligence:

The distinction between artificial and human intelligence is significant:

Mechanism of Learning: Humans learn through experiences, emotions, and social interactions, developing a deep understanding of context and nuance. AI learns through processing vast amounts of data, identifying patterns, and optimizing responses based on algorithms.

Consciousness and Understanding: Human intelligence is intertwined with consciousness and self-awareness. AI, despite simulating understanding, lacks consciousness and does not possess self-awareness.

Adaptability: Humans can apply learned knowledge creatively across diverse situations. AI's adaptability is confined to the data and scenarios it has been trained on, limiting its ability to generalize beyond its programming.

The Debate on AI's Intelligence:

Experts are divided on whether AI's capabilities qualify as true intelligence:

Argument for AI as Intelligence: Some argue that since AI can perform tasks requiring intelligence, such as language translation, strategic game playing, and data analysis, it should be considered a form of intelligence. This perspective focuses on the functional aspects of intelligence, irrespective of consciousness.

Argument against AI as True Intelligence: Others contend that without consciousness, self-awareness, and genuine understanding, AI cannot be deemed truly intelligent. They argue that AI's "intelligence" is a simulation, lacking the subjective experience that characterizes human intelligence.

Conclusion:

While AI exhibits functionalities associated with intelligence, it fundamentally differs from human intelligence in consciousness, understanding, and adaptability. Therefore, labeling AI as "intelligent" depends on whether we emphasize functional capabilities or the presence of consciousness and genuine understanding.

Things that come easily for us are difficult for AI. Please provide some striking examples.

You're highlighting a well-recognized phenomenon in artificial intelligence known as Moravec's Paradox. This paradox observes that tasks humans find effortless, such as perception and mobility, are often challenging for AI, while tasks considered complex, like mathematical reasoning, can be relatively easy for machines.

Striking Examples:

Object Recognition and Manipulation:

Human Ease: A child can effortlessly identify and pick up a toy from a cluttered floor, utilizing innate visual and motor skills.

AI Difficulty: Robots struggle with reliable object manipulation in unstructured environments, requiring sophisticated algorithms and extensive computational resources.

Navigating Complex Environments:

Human Ease: People can walk through crowded spaces, avoiding obstacles and adjusting to dynamic changes without conscious thought.

AI Difficulty: Robots find it challenging to navigate unpredictable environments, often requiring pre-mapped data and struggling with real-time adaptations.

Understanding Context and Common Sense:

Human Ease: Humans use common sense to infer unstated information, such as knowing that ordering a steak at a restaurant implies the person ate it.

AI Difficulty: AI systems lack this inherent common sense, leading to misunderstandings in context and making basic errors that no human would make.

Social Interaction and Emotional Intelligence:

Human Ease: People naturally interpret social cues, emotions, and non-verbal communication, facilitating smooth interactions.

AI Difficulty: AI struggles with understanding and appropriately responding to emotional nuances, limiting effective social engagement.

Physical Dexterity:

Human Ease: Tasks like folding laundry or loading a dishwasher are performed effortlessly by humans.

AI Difficulty: Robots often struggle with these tasks due to the complexities of human-like motion and the challenges in transitioning from virtual simulations to real-world environments.

These examples underscore the inherent challenges in replicating human sensorimotor skills and common sense reasoning within AI systems, despite their proficiency in tasks requiring high-level cognitive functions.

Isn't it ironic that these simple tasks (for us) are the most difficult to automate?

Yes, it is indeed ironic that tasks humans find effortless are often the most challenging to automate. This irony is encapsulated in Moravec's Paradox, which observes that high-level reasoning requires relatively little computation, while sensorimotor skills demand enormous computational resources.

Evolutionary Perspective:

One explanation for this paradox lies in evolution. Human beings have honed sensorimotor skills over millions of years, resulting in highly efficient neural mechanisms for tasks like walking, recognizing faces, and interpreting social cues. In contrast, abstract reasoning and complex calculations are relatively recent developments in human history, lacking the same level of evolutionary refinement. Consequently, while these higher-order cognitive tasks may seem more demanding, they are easier to replicate in machines because they involve explicit, conscious processes that can be codified into algorithms.

Implications for AI Development:

This paradox has significant implications for artificial intelligence and robotics. It highlights that replicating human-like perception and motor skills in machines is a formidable challenge, often underestimated in early AI research. Understanding Moravec's Paradox prompts a more nuanced approach to AI development, acknowledging the complexity of "simple" human tasks and guiding research toward overcoming these inherent difficulties.

In summary, the irony that tasks easy for humans are hard for machines underscores the intricate nature of human cognition and the challenges faced in replicating these abilities artificially.

Are you suggesting higher cognitive functions are biologically less complex?

The relationship between sensorimotor functions and higher cognitive processes is intricate and interdependent. While higher cognitive functions—such as reasoning, problem-solving, and abstract thinking—are often perceived as complex, they are built upon foundational sensorimotor experiences. This interrelation suggests that the complexity of higher cognitive functions is deeply rooted in the sensorimotor system.

Foundational Role of Sensorimotor Functions:

Jean Piaget, a renowned developmental psychologist, proposed that basic sensorimotor functions form the foundation for the development of higher-level cognitive functions. Through active exploration and interaction with their environment, individuals build knowledge, indicating that sensorimotor experiences are crucial for cognitive development.

Neural Interconnections:

The brain's architecture supports the interplay between sensorimotor and cognitive functions. Neural circuits involved in motor control are also engaged during cognitive tasks, underscoring the shared neural substrates between these domains. This overlap highlights the biological complexity inherent in both sensorimotor and higher cognitive functions.

Complexity in Cognitive Skills:

Cognitive skills necessitate extensive information processing, advanced cognition, decision-making, and attention. These tasks often require more time due to the need for deliberation and complex processing, reflecting their inherent complexity.

Conclusion:

Higher cognitive functions are not biologically less complex than sensorimotor functions. Instead, they are deeply intertwined, with sensorimotor experiences laying the groundwork for more abstract cognitive processes. This interdependence reflects the comprehensive complexity of human brain functions.

This reminds me of George Lakoff's theory of the bodily origin even of mathematics.

George Lakoff

Indeed, George Lakoff, along with co-author Rafael Núñez, explores the concept that mathematics is deeply rooted in human bodily experiences in their book Where Mathematics Comes From: How the Embodied Mind Brings Mathematics into Being.

Key Aspects of Their Theory:

Embodied Cognition: Lakoff and Núñez argue that mathematical ideas are not abstract entities existing independently of humans but are instead products of our embodied minds. They suggest that our physical interactions with the world shape the way we conceptualize mathematical concepts.

Conceptual Metaphors: The authors propose that abstract mathematical concepts arise through conceptual metaphors, which map structures from our sensory and motor experiences onto more abstract domains. For example, understanding numbers can be linked to our experiences with object collection or movement along a path.

Cognitive Mechanisms: They emphasize that mathematics utilizes ordinary cognitive mechanisms such as metaphorical thinking, grounding it in everyday human activities and experiences.

Implications of This Perspective:

This viewpoint challenges traditional notions of mathematics as a purely abstract or transcendent field. By grounding mathematical thought in embodied experience, it suggests that our cognitive processes, shaped by our physical interactions, fundamentally influence the development and understanding of mathematical concepts.

In summary, Lakoff's theory posits that our bodily experiences are foundational to the way we conceptualize and engage with mathematics, highlighting the profound connection between physical embodiment and abstract thinking.

Please give a few striking examples of this phenomenon.

George Lakoff and Rafael Núñez, in their book Where Mathematics Comes From, propose that many mathematical concepts are grounded in human bodily experiences through conceptual metaphors. Here are a few striking examples of this phenomenon:

Arithmetic as Object Collection:

Embodied Experience: Humans often interact with groups of objects, adding to or removing from collections.

Mathematical Concept: This experience leads to the conceptualization of arithmetic operations—such as addition and subtraction—as the combining or separating of object groups.

Understanding Infinity:

Embodied Experience: Engaging in repetitive tasks that eventually conclude, like assembling multiple items.

Mathematical Concept: This forms the basis for conceptualizing infinity, where the completion of finite iterative processes is metaphorically extended to imagine an endless process as a completed whole.

Numbers as Points on a Path:

Embodied Experience: Physical movement along a path, where each step represents progress.

Mathematical Concept: This experience translates into the idea of numbers existing on a number line, with addition and subtraction corresponding to movements forward or backward along this path.

These examples illustrate how abstract mathematical ideas are deeply rooted in our physical interactions with the world, supporting the theory that mathematics is fundamentally embodied.