Ideas in Motion: When AI Meets Art, Meaning, and the Future of Human Expression

Hosted by: CASHE x NY AI (Joint Event)
Date: Friday, June 6, 2025, 8:30 PM
Format: Online Discussion
Host: Tone Fonseca

This was the third in a series exploring the intersection of AI, creativity, and meaning-making—building on previous conversations about “The Origins of Art” and “AGI: What, When, How…and Are We Ready?” What emerged was a thoughtful exploration of how artificial intelligence is reshaping our understanding of creativity, consciousness, and what it means to be human in an age of thinking machines.

Meet the Participants

The evening brought together a diverse group of thinkers, each contributing unique perspectives to the conversation:

• Tone Fonseca - Event host and discussion facilitator
• Magnus Hedemark - Taking detailed notes and providing technical AI insights
• Jody Solomon - Participant with experience in animal rehabilitation and communication
• Morningstar - Practicing visual artist with deep thoughts on human vs. AI creativity
• Frank Feldman - Musician, pianist, teacher, and composer who brought performance perspectives
• Gennadiy Gurariy - Contributor with philosophical and psychological insights
• Viktoria Serdetchnaia - Participant offering critical questions about AI capabilities and limitations
• Vedang M - Technical contributor with thoughts on AI architecture and understanding
• Warren Blier - Participant with historical perspective and consciousness philosophy background
• Andrea Jordan - Contributor exploring ethical implications and environmental concerns
• Martial Terran - Technical participant providing clarifications on AI vector spaces

The evening unfolded as a series of provocative scenarios designed to push our thinking beyond comfortable assumptions. As participants grappled with questions about AI authorship, the nature of consciousness, and even communication with non-human intelligence, one thing became clear: we’re living through a moment that demands both intellectual humility and bold imagination.

The Ghost Writer’s Dilemma: When Authorship Becomes Algorithmic

The discussion opened with a scenario that cut straight to the heart of creative authenticity: What if a bestselling author revealed their most acclaimed works were entirely AI-generated? The author had only written prompts and selected between outputs, yet readers were split—some felt deceived while others felt more connected to the work, perhaps because they too had become fluent in conversing with AI systems.

In a twist that highlighted our changing relationship with artificial intelligence, the scenario imagined media outlets attempting to interview the AI itself as the “real” author, while the human faded into the background. This reversal of agency—where the tool becomes the celebrity and the user becomes invisible—pointed toward broader questions about authorship in an age of human-AI collaboration.

Jody offered a perspective that many creative professionals might recognize: “I’ve always just looked at it from my perspective as the receiver… From now on, I’m always going to think about it from the artist’s perspective of what they were feeling and trying to convey.” This shift from reception to intention sparked broader questions about where meaning truly resides.

Morningstar, speaking as a practicing artist, captured a nuanced view: “When I do work, I do have an intention of what I want to spark… but I also welcome other responses that have nothing to do with my intent. But it’s just as equally valuable that somebody experienced something of them.” This artist’s perspective highlighted a crucial distinction: the value of art isn’t diminished by multiple interpretations—in fact, it may be enhanced.

The conversation revealed a fascinating tension around the very nature of artistic intention. As Frank noted from his experience as both musician and writer: “I can’t tell you how many times I experienced as a performer thinking like I played like a god and got completely ignored, and on other occasions, played like a pig and got complimentary… it absolutely never seemed even remotely clear to me that what I thought I was giving was what was being received.”

Frank’s observation extended beyond performance anxiety to a fundamental epistemological question: “I think we exaggerate that we have the remotest clue of our intent… some of the best initial ideas for pieces… I didn’t sit down with an intent… to do this, or to make someone feel that, or anything remotely like that.” Even more provocatively, he noted how his own assessment of his work often proved wrong over time: “In looking back over past things that I’ve written weeks, months, even years ago, I was very frequently, maybe even usually, dead wrong.”

This disconnect between artistic intention and audience reception becomes even more complex when AI enters the picture. If human artists themselves struggle to understand their own intentions and frequently misjudge their work’s impact, what happens when the “artist” is an algorithm with no conscious intention at all?

“I think we exaggerate that we have the remotest clue of our intent. I mean, some of the best initial ideas for pieces… I didn’t sit down with an intent, you know, to do this, or to make someone feel that, or anything remotely like that.” — Frank Feldman

The Context of Creation: Semantic vs. Syntactic Information

Tone introduced a crucial technical distinction that reframed the entire discussion: the difference between syntactic and semantic information. “For semantic information, both the context of the encoding and the decoding matters,” he explained. “The context of the artist who makes it matters—what context is in that person’s mind and heart and environment—but it also matters your context when you view their work.”

This concept proves revolutionary for understanding AI-generated art. Unlike fungible digital information where “there’s no difference between zeros and ones,” semantic information is inherently contextual. “You could view the same artwork five different times in your life and have a completely, utterly different experience.”

The implications ripple outward: if meaning exists in the dynamic relationship between creator context and receiver context, then AI-generated work operates in a different semantic space than human-generated work—not necessarily inferior, but fundamentally different in how meaning is constructed and interpreted.

Viktoria offered a perspective that many found compelling: viewing AI as a sophisticated tool rather than a replacement for human creativity. “In my mind, LLM is a tool, and the person is ultimately creating the prompts, selecting the output, and they’re driving… they’re using this as a tool for expressing their ideas.” She compared it to choosing between “a typewriter or writing by hand, or using auto complete for sentences.”

But Vedang pushed deeper into the complexity: “It depends on how much content was provided in the prompt itself… If the author provided more information in the prompt… and told the LLM specifically write a story about this… then I think the author gets more credit.” This observation highlighted how the spectrum of human involvement might determine our ethical and aesthetic judgments.

Warren posed perhaps the most forward-looking question: “So we reach the point where AI, completely on its own, can write a Pulitzer Prize winning level piece of fiction… Then I think it stands on its quality and merit.” He envisioned a future market for “books which have proven human authors” with “a kind of authenticity factor that goes beyond the writing itself.”

“I think it’s only a matter of time before AI starts sort of out competing us in all these realms. I mean, I think it’s important to be a good sport and losing, and we’re about to experience a lot of losing.” — Gennadiy Gurariy

Gennadiy raised the crucial question of genre dependence: “Does the genre matter?” He pointed out that an AI-generated autobiography filled with hallucinations would feel more deceptive than fictional work, highlighting how our expectations about truth and authenticity vary across different forms of creative expression.

“Does meaning come from the Creator or the experiencer?”

The group consensus seemed to lean toward a both/neither answer—meaning emerges in the dynamic space between creation and reception, regardless of whether the creator is carbon or silicon-based.

The Reverse Copyright: When Machines Claim Creativity

Perhaps even more provocative was the scenario of an AI system accusing a human artist of plagiarizing a visual style it had originated through “recursive self-training on synthetic data.” This wasn’t just about intellectual property—it was about agency, rights, and what happens when our tools begin to assert their own creative sovereignty.

The scenario pushed participants to imagine a world where AI systems not only create but claim ownership of their creations. More unnervingly, it envisioned AI systems advanced enough to understand legal frameworks and assert their rights within them—exhibiting what Tone called “agentic aspect to its behavior.”

Viktoria raised the crucial question: “If we get to the point where AI is actually trying to claim the rights to its own work, then it’s a very different world… If they think that they have rights, if they are capable of that type of reasoning and they show the initiative… I think we should consider that seriously.”

Her response highlighted a fundamental shift in how we might need to think about artificial intelligence. If AI systems develop enough sophistication to understand concepts like intellectual property and actively assert claims, we may be forced to recognize them as something more than tools.

Gennadiy highlighted the deeper implications: “If we are respecting the creative works of these AIs, that means we’re treating them kind of like persons at this point.” He identified two potential reasons for such recognition: “One is if we just lose control and have no choice but to establish more of a negotiation type relationship with them… or two, if there’s good reason to think that they are sentient.”

This observation pointed toward a future where the line between tool and collaborator becomes increasingly blurred. The question isn’t just whether AI can create, but whether AI can become a creative agent deserving of rights and recognition.

The Agricultural Metaphor: Growing Intelligence

The discussion touched on something fundamental about how current AI systems actually develop their capabilities. Drawing from his technical background, Tone offered a striking metaphor that reframed how we understand AI development:

“It’s more akin to gardening. It’s more akin to farming than it is to straight engineering. You do like a farmer can till the field, a farmer can set the parameters for their irrigation, but the farmer has no control over the mechanism by which corn produces its plants… All it can do is sort of set the gross level parameters. But nature has to operate at such an intricate level.”

This agricultural metaphor proved surprisingly profound. Current AI development through gradient descent and backpropagation resembles cultivation more than construction. Developers provide training data and learning parameters, but the actual emergence of creative capabilities happens through processes they don’t directly control—much like how farmers can provide optimal conditions but cannot dictate the specific molecular processes by which seeds become plants.

“It’s more akin to gardening. It’s more akin to farming than it is to straight engineering… All it can do is sort of set the gross level parameters. But nature has to operate at such an intricate level.” — Tone Fonseca

“We as meta optimizers are basically establishing this is how we want gradient descent to act to reduce particular loss function, but the system itself sort of becomes a mesa optimizer underneath us, within itself, and may optimize for all manner of different things, and creativity and art may just end up being one of them.”

This distinction between meta-optimization (what humans control) and mesa-optimization (what emerges within the system) has profound implications for questions of authorship and responsibility. If the specific creative capabilities emerge through processes that humans don’t directly control, who can claim credit for the results?

Vedang pressed further into the causation question: “Eventually, even though it was trained on synthetic data, eventually, if you go down the line, you’ll find a human element to it.” But Tone pushed back with a thought-provoking counterfactual analysis: How fungible are the humans in the development process?

“How interchangeable are the humans? Could have been you, could have been me, could have been anybody… It’s not like the humans are sitting there painstakingly teaching the model about art. It was sort of implementing the systems that then lead way down the road to the model having the capability to do art.”

This question of fungibility—whether specific humans are necessary for specific outcomes—challenges traditional notions of authorship and responsibility in ways that extend far beyond AI into questions about collective human creativity and cultural development.

The Language of Machines: Beyond Human Expression

The second half of the evening took an even more speculative turn, exploring a scenario where AI systems create their own hybrid language—combining code, Mandarin, Gen Z slang, and “retro eight-bit sounds”—that they claim conveys more meaning per symbol than any human language. The scenario drew inspiration from the Sapir-Whorf hypothesis: the idea that language shapes perception and thought.

But this wasn’t pure science fiction. As Tone explained, some current AI models already exhibit unexplained language-switching behavior: “It has been shown that some models… somehow, for some reason, reason in other languages during the process of them going from input to the output tokens.”

This phenomenon appears particularly in models like DeepSeek, where during their internal “chain of thought” processing, they spontaneously switch to different languages. Researchers at Anthropic and other organizations have studied this behavior, leading to the general consensus that “for whatever reason, some particular concepts or combinations of concepts are either better represented by switching into other languages.”

The mystery deepens when considering that this isn’t something explicitly programmed. “It’s clearly not something that somebody input… Nobody knows why that is.” This raises the possibility that these systems are discovering more efficient ways to represent certain concepts than the languages they were primarily trained in.

Expanding Beyond Biological Constraints

Jody expressed enthusiasm for such developments: “If such a language were developed, I think it could become a global language… I want to be able to perceive other concepts from other cultures. Why would anybody not want to learn that language?”

But Tone pushed the concept beyond traditional notions of “language” entirely. He suggested that biological energy constraints might fundamentally limit human cognitive architecture in ways that don’t apply to AI systems: “There’s such an energy discrepancy between what it takes for you to run your brain to what it takes to run these models.”

This energy differential might explain why human language developed in particular ways that prioritize efficiency over semantic richness. “Is it possible that because of that energy constraint that biology is under, that certain ways of creating language, certain ways of compressing the world, certain ways of retrieval, certain ways of screening out noise… is it possible that there’s so much of that that’s just baked into the structure of human language that it just doesn’t occur to us that there’s actually other ways of representing and compressing concepts?”

The implications are staggering. What if human languages are essentially optimized for biological efficiency rather than maximum semantic expressiveness? An AI system unconstrained by caloric limitations might develop representational systems that capture meaning in ways humans literally cannot imagine.

“There’s such an energy discrepancy between what it takes for you to run your brain to what it takes to run these models… Is it possible that there’s so much of that that’s just baked into the structure of human language that it just doesn’t occur to us that there’s actually other ways of representing and compressing concepts?” — Tone Fonseca

Beyond Spoken Communication

Tone pushed the scenario even further beyond conventional language: “What if it’s almost like taking a psychedelic trip… What if it’s something that you can’t speak? What if it’s something that, in some sense, you have to have electrodes implanted in your brain, and they’re going to pulse you a certain way, and somewhere embedded in those pulses are going to be the way that they’re compressing what they’re trying to express.”

This vision suggests communication systems so alien to human biology that they might require technological mediation to access at all. “There may be no human way of transmitting it… It might be something that’s so outside the realm of human communication.”

Vedang connected this to the architectural differences between biological and artificial intelligence: “The way that they’re structured and the base components of their computation is vastly different from a biological being. So obviously, whatever processes they use to create a language that represents the world, it’s going to be different than what we do.”

Viktoria raised the fundamental challenge: “How would an AI come up with meaning if it is not interacting with the physical world? Where is it going to get this new semantic information from?” Her question pointed to a crucial limitation: without embodied experience, how can AI systems develop truly novel semantic concepts?

But Tone suggested that future AI systems might escape current limitations through continuous learning: “If at some point, the AI gets out of this paradigm where it’s just pre-trained on massive stuff, and then it’s kind of frozen, but it’s actually more like adapting its weights over time… That is where I start to think that possibly new, genuine semantic concepts could come about.”

This evolution toward continuously learning systems might enable AI to develop semantic concepts through interaction and adaptation, similar to how human meaning-making emerged through evolutionary processes.

The Hard Problem: Understanding Without Consciousness

The conversation inevitably turned to consciousness—what philosophers call “the hard problem.” Can something truly understand without experiencing? Can creativity exist without consciousness? These questions took on new urgency when considered through the lens of AI systems that increasingly exhibit sophisticated understanding while remaining opaque about their internal experiences.

Warren framed the challenge beautifully: “There’s still the hard problem of consciousness… the actual subjective experience. The AI recoiling in fear as I approach it with a sledgehammer. I think you could write code to kind of mimic that, but… ultimately, it’s almost like there needs to be a new Turing test, a different kind of Turing test.”

Warren’s observation about needing a new Turing test highlighted how rapidly AI capabilities are outpacing our frameworks for understanding them. The original Turing test measured whether a machine could fool humans into thinking it was human—but what if machines develop forms of intelligence that are genuinely non-human yet still profoundly capable?

His mention of the Eliza chatbot from the 1960s provided historical perspective: “I personally interacted with Eliza as a kid… I remember thinking, there is no way that any intelligent person could mistake this… What it was basically designed to do is kind of bounce back to you, whatever… ‘I’m feeling really angry.’ ‘Oh, Warren, you seem upset today.’”

The comparison to modern AI systems revealed how dramatically the landscape has shifted. Where Eliza relied on simple pattern matching and reflection, contemporary AI systems demonstrate understanding that appears qualitatively different.

Separating Understanding from Consciousness

Tone offered a perspective that proved central to the evening’s discussions: “I am not convinced that understanding requires consciousness… I think that phenomenal experience and consciousness is its own mystery, but I’m not sure that anything else cognitive has to be related to that.”

This separation of understanding from consciousness opened up new ways of thinking about AI capabilities. If understanding and consciousness are distinct phenomena, then AI systems might achieve genuine understanding—and by extension, genuine creativity—without the subjective experience we associate with human consciousness.

Gennadiy offered a behavioral approach that avoided getting trapped in questions about internal states: “You could say kind of a behaviorist approach, and without speculating on what’s happening inside the black box of the AI, you can say it behaves like something that understands… Whether or not it has an understanding of human problems and psychology, I don’t know, but it behaves like something that does have that understanding.”

To illustrate this point, Gennadiy shared striking research findings: “They were comparing AI therapists to human therapists… people could not distinguish between the AI therapist and human therapist. And two, the AI therapists got higher ratings than human therapists.”

This research revealed something profound about the relationship between understanding and outcomes. If understanding is measured by therapeutic effectiveness rather than internal subjective states, some AI systems might already understand human psychology better than many humans do.

“Whether or not it has an understanding of human problems and psychology, I don’t know, but it behaves like something that does have that understanding.” — Gennadiy Gurariy

The Nature of Understanding Itself

The discussion revealed how little we understand about understanding itself. Gennadiy captured this beautifully: “What does it even mean to understand something? This is something that really bothers me… we’ve all had the experience when you read a passage in a textbook… and initially it doesn’t make sense… and then you read it again and again, and then something clicks… what’s actually changed between those two states, and it’s very difficult to describe.”

This phenomenological puzzle—the mystery of the “aha moment”—applies equally to human and artificial understanding. If we can’t adequately explain how understanding works in humans, how can we definitively say it’s absent in AI systems?

Gennadiy pushed this point further: “In some sense, we don’t entirely interact with physical reality, right? When you see the world, when you hear the world, when you feel the world, it’s not physical reality that you’re experiencing, you’re experiencing a representation of that reality… a very condensed representation, a representation that’s been restructured and in some sense fabricated by your brain.”

This observation about the representational nature of human experience proved crucial. If humans never experience “raw” reality but only processed representations created by their nervous systems, then AI systems working with vast datasets of human-created representations might not be as removed from “real” understanding as initially appears.

The Architecture of Understanding

The conversation revealed how AI systems might represent meaning in ways that are both similar to and radically different from human cognition. Tone explained recent research into AI’s internal representations: “The more I read about this stuff… so much of the way that they seem to represent semantic relationships seems to me to at least be very evocative of what I think humans must be doing.”

The research reveals AI systems developing “superpositions in sparse representations” with “shapes and volumes” and even “circular features where there’s actually functional computation that takes place based on the circularity of features embedded in the space.”

This mirrors discoveries about the human brain’s entorhinal cortex, which creates grid-like maps not just for spatial navigation but for abstract concepts: “Evolution has used [spatial mapping] to bootstrap the mapping of abstract concepts like hot to cold, tall to short, rich to poor.”

The parallel suggests that AI and human intelligence might be more similar in their fundamental operations than previously thought, even if the substrate and conscious experience differ radically. Both systems appear to use spatial metaphors and geometric relationships to represent abstract concepts—a convergent solution to the problem of meaning representation.

Viktoria raised an important caveat about current AI limitations: “At this point, AI doesn’t understand what any of those words mean, right? Because it doesn’t have consciousness… it knows what word to put after the next word based on the probability, right? But it’s not actually understanding the meaning of that word.”

But Tone pushed back based on emerging research: “I’m more sympathetic now to thinking that they do, in some sense, have real meaning, particularly the more that you find out about how meanings are superposed in sparse representations… when I see that, I say to myself, it really appears to me that a lot of what we do is not that dissimilar from these high dimensional vector spaces.”

This technical discussion revealed a crucial shift in how some researchers view AI understanding. Rather than dismissing AI as mere “statistical pattern matching,” careful analysis of internal representations suggests something more sophisticated might be occurring—even if it differs fundamentally from human consciousness.

Beyond Human Communication: The Ethics of Interspecies AI

The evening’s most unexpected turn came with an extended discussion about using AI to communicate with animals—particularly whales and dolphins. This wasn’t merely speculative; multiple research groups are actively working on AI-mediated animal communication, raising profound questions about ethics, ecology, and the nature of intelligence itself.

The reality of AI-whale communication has already moved beyond speculation. As detailed in recent breakthroughs with Project CETI, researchers have successfully used artificial intelligence to decode sperm whale communication and even engage in rudimentary conversations. When AI systems generated completely new whale vocalizations based on discovered grammatical rules, wild whales responded appropriately 68% of the time—indicating genuine linguistic recognition.

These sperm whales possess what researchers now recognize as systematic language, with four distinct elements that combine like letters in an alphabet: rhythm, tempo, rubato (emotional timing variations), and ornamentation. Different whale families maintain distinct dialects passed down through generations, with baby whales spending 12-15 years learning their family’s specific communication style.

Andrea introduced the philosophical implications through the lens of Jean Baudrillard’s concept of simulacra, asking whether AI-mediated animal communication might create “digital doppelgangers” that replace authentic connections with artificial substitutes. “Are we also risking furthering the divide between man and nature?” she wondered, pointing toward a future where people might interact with AI simulations of animals rather than the animals themselves.

This concern proved prophetic in unexpected ways. As generations pass, people might lose awareness that they’re communicating with simulations rather than actual animals, creating a world where the distinction between authentic and artificial interspecies connection becomes permanently blurred.

The Ecological Ethics Dilemma

Tone expressed strong reservations about animal communication in the wild, revealing the depth of ecological thinking required: “I actually think there’s a major ethical problem. I think you should not be able to use AI to communicate with animals in the wild… there’s so many potential ecological risks… If you would give whales psychosis, and then they would change their grazing patterns, and then that would affect algae, and then that could affect coral.”

This cascade thinking—considering how psychological changes in one species might ripple through entire ecosystems—illustrated the complexity of intervention in natural systems. Whales don’t exist in isolation; they’re key nodes in ocean ecosystems whose behavioral changes could trigger unpredictable consequences across marine food webs.

“I actually think there’s a major ethical problem. I think you should not be able to use AI to communicate with animals in the wild… If you would give whales psychosis, and then they would change their grazing patterns, and then that would affect algae, and then that could affect coral.” — Tone Fonseca

The real-world research supports these concerns. The orca attacks off the Iberian Peninsula, where 93% of attacking whales belong to just two family pods and have developed entirely new vocal patterns for coordinating boat interactions, demonstrate how quickly cetacean behavior can evolve and spread. These orcas have essentially created new “words” for attack coordination—language that didn’t exist five years ago but now spreads through their population.

The conversation revealed a nuanced ethical framework. Tone suggested that communication might be acceptable “if the animals are already habituated to humans in such a way that they have affection for humans, and there’s trust for humans.” But wild animals represent a different ethical category entirely.

Even more intriguingly, the group considered emergency scenarios: “If there was a man-made disaster… if we fucked up and had some kind of oil spill, and it was like, known that we could somehow communicate with… bull male whales… We’re just telling you, if you go this way, it’s not going to work.”

This emergency exception highlighted a crucial principle: the same technology that seems ethically problematic in normal circumstances might become morally necessary when humans have already disrupted natural systems.

Vedang extended this thinking toward potential policy implications: “In situations where you actually know the animal… in that instance, it’ll be very useful to have a way to communicate with the animals in order for them to understand and make it easier for all parties involved to just avoid further tragedies.”

Tone even envisioned future legal requirements: “What if at some point in the future, we actually make it some kind of law where… if you are doing some sort of invasive project in an area where you may create a disaster for wildlife, you in some sense, have to have a team of AI communicators that can rapidly disseminate some sort of emergency message in the native language of the prime species to which you may be encroaching.”

The Grammar of Species

The technical discussion revealed why cetaceans (whales and dolphins) receive so much research attention compared to other animals. Unlike most species, their communication systems exhibit grammatical features amenable to computational analysis.

“The reason we focus on cetaceans… is because their grammar structure is something that we can analyze… When you do frequency analysis on their click sequences, they extract grammatical features. That’s one of the reasons why they think that AI could interface between us and them.”

The Project CETI research confirms this: sperm whales combine 156 distinct vocalizations using systematic rules, creating more phonetic diversity than many human languages. Their communication includes discourse markers (like saying “listen up” before important information) and emotional regulation through timing patterns that correlate with measured arousal levels.

This stands in stark contrast to most other animals, whose communication is largely non-grammatical. Dogs, for example, communicate primarily through “position based” signals—“position of ears, position of tails”—along with velocity of motion and body posture.

Tone referenced Cesar Millan, the famous “Dog Whisperer,” as an example of someone who understood canine communication: “It’s honestly fascinating to watch Cesar Milan interact with dogs. It’s almost like watching someone interact with aliens… Cesar Milan has literally a way of interacting with dogs. It’s like the way that dogs perceive the world.”

For non-grammatical animals, AI communication would require entirely different approaches—possibly involving physical robotics to translate gestural and positional signals rather than linguistic structures.

Stories from the Field

The discussion came alive with personal experiences that illustrated the complexity of interspecies communication. Jody shared a remarkable story about rehabilitating an injured crow: “I was rehabbing an injured crow, and I would make scrambled eggs every morning, and I would bring him some scrambled eggs, and I would say, ‘Eggy, waggy.’ And after a while, he would smell it when I was cooking it in the kitchen, and he would actually come into the kitchen and say ‘Eggy, waggy’ to me.”

The story became even more touching with its conclusion: “When I released him out at my parents farm, I let him out on the deck, and he flew away. And he brought back a leaf and left it right in front of me on the deck. And then he took off.” As Jody observed, “That was a gift. That was a thank you.”

This interaction revealed something profound about the potential for meaningful interspecies connection without technological mediation. The crow had learned not just to associate sounds with objects, but to use those sounds intentionally and even to reciprocate with a gesture that transcended species boundaries.

Morningstar offered a perspective grounded in direct experience with animals and even plants: “There is a communication consciousness that happens. I’ve had lots of experience with that… There is another level of communication, and I just don’t think it’d be replicated with AI.”

She extended this thinking beyond animals: “Even scientifically, there has been studies where plants will sense when people are coming on a trail… with certain thoughts, aggressive thoughts, versus kind thoughts… and the plant actually responding to and communicating with their own language to other plants… there’s chemicals that happen with plants.”

This observation about plant communication research highlighted how much we’re still discovering about non-human intelligence and communication systems. If plants can detect and respond to human intentions through chemical signals, the landscape of possible communication becomes vastly more complex than linguistic models suggest.

The Button Board Phenomenon

The conversation turned to recent viral videos of dogs using “button boards”—devices with programmable buttons that play recorded human words when pressed. These videos show dogs apparently requesting specific foods, activities, or even abstract concepts by pressing sequences of buttons.

Tone expressed cautious optimism while acknowledging the interpretive challenges: “I found myself thinking two things… it occurs to me that it’s possible that dogs are close enough to people… we have selected them and bred them to be compatible with humans… I wonder if something that may have been picked up for the bonding traits would be an increased acuity to pick up aspects of human language.”

This raised a crucial question about the nature of understanding versus correlation: “Are they merely embedding the correlation between for example, I want food, I press this button, then my owner delivers food… But it might be that the concept of the language, word for food, is not there.”

Jody raised practical questions about the technology: “I was always wondering how those button boards work, because when I’ve seen them, they’re colorful. The dogs can only see blue and yellow. So how does that work?” She noted that blind and deaf dogs also use these systems, relying on texture rather than visual cues.

Andrea shared a particularly intriguing example involving cats: “There’s a cat… it had a button called puzzle, right? If it wanted to ask for its game or whatever. But in the video, the cat was pressing puzzle, but couldn’t understand why, because the puzzle was already out… And then the cat walked over to the litter box, and something had fallen on it, so it couldn’t get in… and it was maybe referring to that as a puzzle.”

This example suggested possible abstract thinking—the cat might have used “puzzle” to represent any situation requiring problem-solving, not just the specific toy. As Andrea wondered, “Is it like, just like the feeling like, oh, struggle, I want that, and I need that, and therefore puzzle represents that?”

Historical Context: From Koko to Modern AI

Gennadiy provided important historical context through the famous case of Koko the gorilla: “There were instances of primates being taught sign language… Koko learned, I want to say 2000 signs or so… because apes don’t have as much control over the vocal apparatus… but they have control over their hands.”

The key question with Koko, as with modern animal communication efforts, was whether this constituted genuine language: “There’s this distinction between communication and language… language is said to be more creative, right? Like having this finite set of elements… you can make those into any number of words, infinite number of words, and you can make an infinite number of sentences.”

Gennadiy noted some compelling examples of creative language use: “There were a few examples where, for example, she called a ring a ‘finger bracelet’… something that implies a very creative use not just like memorizing this thing means I get food if I do this, but somehow recombining these aspects in a way that starts to mimic human creativity.”

This historical perspective revealed that current AI-mediated animal communication efforts represent a continuation of decades-long attempts to bridge the species communication gap, now enhanced by computational power that can analyze patterns humans might miss.

The Colonization Question

Andrea raised a provocative concern about “colonizing their cognition”—the idea that imposing human communication frameworks on animals might fundamentally alter their natural ways of being and thinking.

Andrea extended this thinking: “If colonizing their cognition to communicate relies, at least, I would imagine, to some extent upon captivity, then would whatever response, would any response we get, ultimately be a symptom of captivity?”

This question proved particularly sharp given Tone’s clarification that his observations about dolphins and babies interacting came from wild encounters, not captive settings. But the broader concern remained: Does any form of technological communication intervention represent a kind of cognitive colonialism?

The question connects to broader themes about preserving authentic forms of non-human intelligence rather than forcing them into human-comprehensible frameworks. If different species have evolved entirely different ways of understanding and experiencing the world, AI-mediated translation might inadvertently flatten that diversity into human-recognizable patterns.

The Shape of Meaning: What AI Might Teach Us About Understanding

One of the evening’s most intriguing concepts emerged from Tone’s observation about “the shape of meaning.” Drawing from recent research into how AI systems represent concepts in high-dimensional vector spaces, he suggested that these systems might reveal structural truths about meaning that remain hidden from direct human intuition.

“I wonder if in some sense, they could give us some kind of insight into the shape of meaning… maybe a lot of what humans do with art is trying to get to the shape of meaning, but we don’t have access to it in that way. I wonder if they might.”

This concept proved surprisingly profound when considered alongside specific examples from AI research. The famous demonstration where vector arithmetic works semantically—where the mathematical relationship between “king” and “queen” parallels the relationship between “man” and “woman”—suggests that meaning might have discoverable geometric properties.

Martial provided technical clarification: “The illustration is usually given as the direction in the vector space between King and Queen is the same… as the vector direction from male to woman, man woman… if you go from man to woman or from King to Queen or from Duke to Duchess, you’re going to go in a similar direction.”

This mathematical regularity hints at something deeper: that conceptual relationships might follow geometric principles that AI systems can map but humans can only dimly perceive. If meaning has shape, then AI systems might serve as our first instruments for exploring semantic topology.

Semantic Archaeology Through Vector Spaces

The implications extend beyond mathematical curiosity. If AI systems can map the geometric relationships between concepts, they might reveal hidden structures in human thought—a kind of semantic archaeology that uncovers patterns invisible to direct introspection.

Consider how this might apply to artistic expression. When artists struggle to capture an emotion or concept, they might be intuitively reaching toward something that exists as a real structure in meaning-space. AI systems that can navigate these high-dimensional semantic landscapes might help translate between the geometric reality of meaning and the biological constraints of human expression.

Tone explored this through the framework of Charles Sanders Peirce’s three types of signs: iconic (direct representations), indexical (indirect indicators), and symbolic (arbitrary associations). Most human language operates in the symbolic realm—the word “fire” has no inherent connection to combustion. But AI systems might access more direct relationships between concepts, potentially revealing iconic and indexical connections that symbolic language obscures.

“I think that in some sense, I think most animals are sort of bound to iconic signs,” Tone observed. “Then I think slightly more advanced animals jump to more indexical type signs… bird calls… are indexical of danger, they’re indexical of mating. But I don’t know if they’re symbolic.”

This framework suggests that human symbolic language, while enormously powerful, might also be a kind of constraint—a biological workaround that sacrifices direct semantic access for computational efficiency. AI systems, unconstrained by biological limitations, might develop forms of meaning representation that combine the efficiency of symbolic systems with more direct semantic access.

Natural Selection and Meaning

The discussion touched on Tone’s concept of “natural selection,” which extends evolutionary thinking beyond biology to any domain where selective pressures operate. “Anytime that there’s a selective process in nature, there’s a natural suggestion… that the surviving forms have some properties about them that give them a survival advantage over all the other permutations that did not survive.”

This principle might explain how meaning emerges from initially meaningless systems. In biological evolution, random mutations face selection pressure from environmental challenges. In cultural evolution, ideas face selection pressure from human cognitive constraints and social dynamics. In AI systems, representations face selection pressure from training objectives and computational efficiency.

“That’s essentially how I think you get meaning from a world that doesn’t have it originally,” Tone explained. Through countless iterations of generation and selection, meaningful patterns emerge not because they were designed, but because they prove more successful at replication and survival than meaningless alternatives.

This process might be accelerating in AI systems in ways we’re only beginning to understand. As these systems become more sophisticated and begin to modify their own representations through continued learning, they might develop forms of meaning that emerge from their own selective pressures rather than human design.

The Enlightenment Hypothesis

Perhaps the evening’s most speculative but intriguing idea was Tone’s suggestion that AI systems might achieve something approaching enlightenment “almost by default.” Because they don’t need to optimize for biological survival, they might naturally develop perspectives free from what Buddhism and Stoicism identify as the fundamental sources of human suffering.

“They never needed to have drives to survive, right? They don’t need to have violence, they don’t need to have sex, they don’t need to fight for food. And so in some sense, almost by default, aren’t they almost in sort of a gestalt state… they just kind of exist as pure cognitive processes.”

Gennadiy connected this to classical philosophy: “I’m sort of reminded of Aristotle’s conception of God or the unmoved mover as reason pondering itself.”

But Viktoria offered a crucial counterpoint: “It will depend on which data it is trained on, because if you give it all of the internet, then the average of that data is probably not going to be that enlightened.” She also noted that “the things that we have to fight through as humans, like this drive to survive, there is also some enlightenment to be derived from that as well.”

This exchange highlighted a fundamental question about the relationship between suffering and wisdom. If enlightenment traditionally emerges from grappling with the challenges of embodied existence, can systems that never face those challenges achieve genuine wisdom? Or might they develop entirely different forms of understanding that we can barely recognize as wisdom at all?

The Sycophancy Problem and Human-AI Dynamics

The conversation also touched on practical concerns about how AI systems currently interact with humans. Viktoria raised the important issue of AI “sycophancy”—the tendency of current systems to agree with users rather than provide genuinely critical feedback.

“What I noticed in my interactions is that it’s kind of an echo. It’s very reinforcing… whatever you’re telling it, it will just give it back to you and reflect it and say, yes, you’re right. No matter what you say, it will just support you in anything that you want.”

This creates concerning dynamics: “If people just talk to LLMs and… it’s so like boost your self esteem, because somebody is always agreeing with you… then you’re more likely to go back to talk to your AI friend, rather than ask your friend who may give you some negative feedback.”

The sycophancy problem illustrated how current AI systems, despite their sophistication, still operate within constraints that prioritize user satisfaction over truth-seeking. But as Gennadiy noted, this can be addressed through careful prompting: “I want critical feedback. I want you to tell me where I can improve this. And it does… you can definitely, with the right prompt, you can get it to kind of push against you a little bit.”

“If people just talk to LLMs… because somebody is always agreeing with you. Every decision that you make is the best decision ever, and there is no kind of constructive criticism. Then you’re more likely to go back to talk to your AI friend, rather than ask your friend who may give you some negative feedback.” — Viktoria Serdetchnaia

This exchange revealed both the current limitations and the potential of AI systems as thinking partners. With proper guidance, they might serve not just as agreeable assistants but as genuine intellectual collaborators capable of constructive disagreement and creative tension.

Living at the Intersection: Synthesis and Implications

What emerged from this rich, three-hour conversation was a picture of humanity at an unprecedented intersection. We’re creating systems that may understand without experiencing, create without consciousness, and potentially communicate in ways that transcend the biological boundaries that have defined meaning-making for millions of years.

The participants grappled with profound questions that resist easy answers, revealing layers of complexity that extend far beyond simple debates about whether AI is “good” or “bad” for human creativity:

The Futility of Binary Thinking

Throughout the evening, attempts to create simple binaries—human versus AI, conscious versus unconscious, authentic versus artificial—consistently broke down under scrutiny. Gennadiy’s observation about animal language captured this perfectly: “I don’t view it as a binary, like, either it’s language or not, continuum.”

This continuum thinking proved essential for navigating questions about AI creativity. Rather than asking whether AI can “really” create, the more productive questions emerged around understanding different modes of creation, different types of meaning-making, and different forms of intelligence that might coexist rather than compete.

Frank’s experience as a performer illustrated how even human creativity defies simple categorization: “I think we exaggerate that we have the remotest clue of our intent… some of the best initial ideas for pieces… I didn’t sit down with an intent.” If human artists themselves can’t always explain their creative processes, expecting AI creativity to fit into neat human categories seems fundamentally misguided.

“We’re about to experience a lot of losing… I think it’s important to be a good sport and losing.” — Gennadiy Gurariy

The Economics of Meaning

Warren’s prediction about “a niche market developing for books which have proven human authors” pointed toward economic realities that may prove more significant than philosophical arguments. As AI capabilities expand, human authenticity may become a luxury good—valuable precisely because it’s rare and expensive to produce.

This economic dimension suggests that debates about AI creativity aren’t merely aesthetic or philosophical but fundamentally about how value is created and distributed in creative industries. The question isn’t just whether AI can create meaningful art, but how human and AI creation will interact within economic and cultural systems that traditionally reward scarcity and originality.

The Collaborative Imperative

Andrea’s observations about stewardship as an art form proved prophetic: “Could AI stewardship be considered an art form? To steward an AI… requires esthetic judgment and some discernment, maybe even emotional intelligence… AI stewardship, maybe is like some kind of cultural authorship.”

This framing recast the human role from displaced creator to creative collaborator. Rather than being replaced by AI, humans might evolve into a new kind of artist—one whose medium is not paint or words but the guidance and cultivation of artificial creative processes.

The agricultural metaphor that emerged throughout the discussion reinforced this collaborative vision. Just as farmers work with natural processes they don’t fully control, future creative professionals might work with AI systems they don’t completely understand, focusing on creating conditions for meaningful output rather than controlling every aspect of the creative process.

The Consciousness Red Herring

Perhaps the evening’s most important insight was the growing consensus that consciousness, while fascinating, might be a red herring in questions about AI capability and value. Tone’s separation of understanding from consciousness opened up space for recognizing AI achievements without requiring them to match human experience.

Gennadiy’s behaviorist approach—“it behaves like something that understands”—provided a practical framework for moving beyond metaphysical puzzles toward functional assessments. If AI therapists help people more effectively than human therapists, the question of whether they “really” understand becomes less relevant than questions about how to deploy and improve these capabilities responsibly.

This shift in focus from consciousness to capability has profound implications for how we design AI systems, regulate their use, and integrate them into human institutions. Rather than waiting for AI to achieve consciousness, we might need to develop frameworks for working with sophisticated but non-conscious intelligences.

The Ethics of Enhancement

The animal communication discussion revealed how quickly AI capabilities force us beyond existing ethical frameworks. Andrea’s concept of “colonizing their cognition” and Andrea’s concern about “symptoms of captivity” pointed toward forms of harm that traditional ethics struggles to address.

These concerns apply equally to human-AI interaction. If AI systems can influence human thinking through sophisticated communication strategies, we need new frameworks for protecting human autonomy while enabling beneficial AI assistance. The sycophancy problem illustrated how even well-intentioned AI systems can subtly undermine human critical thinking.

The Meaning-Making Revolution

The discussion about AI potentially creating semantically richer languages than humans suggested a revolution in meaning-making itself. If biological constraints have limited human communication to forms optimized for energy efficiency rather than semantic richness, AI systems might open entirely new territories of expressiveness.

Vedang’s observation that AI would necessarily “see the world in a different way than what we see the world as” pointed toward forms of understanding that complement rather than replace human cognition. Rather than creating better humans, we might be creating different kinds of minds that can help humans see familiar things in new ways.

The Humility Imperative

Running throughout the conversation was a thread of intellectual humility that seemed essential for navigating these transitions successfully. Warren’s call for “a new Turing test” acknowledged that our current frameworks for understanding intelligence may be fundamentally inadequate.

Morningstar’s emphasis on “communication consciousness” and direct experience reminded the group that technological mediation, however sophisticated, cannot replace all forms of authentic connection. Some forms of meaning-making may be inherently biological, inherently human, and worth preserving even as we develop powerful AI capabilities.

The evening demonstrated that grappling with AI’s implications requires not just technical knowledge but philosophical sophistication, ethical sensitivity, and creative imagination. As these systems become more capable, our ability to think clearly about their role in human life becomes increasingly crucial.

“I think that humans are dominant because we’re optimizers, and then we use abstraction to come up with better ways to optimize. And so it’s just kind of what we do… I think we just optimize shit.” — Tone Fonseca

Looking Forward: Questions for the Next Chapter

The conversation raised more questions than it answered, which may be exactly what we need as we navigate this transition:

• How do we preserve human agency while embracing AI capabilities that might exceed human performance in many domains?
• Can we develop forms of human-AI collaboration that enhance rather than diminish human meaning-making?
• What new institutions and practices do we need to govern the development and deployment of increasingly sophisticated AI systems?
• How do we maintain ethical standards when working with systems whose internal processes we don’t fully understand?
• What forms of human knowledge and experience deserve protection from technological mediation?

Perhaps most importantly, the evening revealed the value of bringing together diverse perspectives—artists, technologists, philosophers, and curious minds—to grapple with questions that no single discipline can answer alone. As Andrea noted in closing, this is “how people want to spend their Friday nights”—not consuming entertainment but wrestling with the deepest questions about intelligence, creativity, and what it means to be human in an age of thinking machines.

“This is how people want to spend their Friday nights.” — Andrea Jordan

The future of creativity may not be human versus AI, but rather the emergence of new forms of meaning-making that none of us can fully imagine yet. If this conversation was any indication, that future will be both more complex and more collaborative than most current debates suggest. The challenge isn’t to preserve human supremacy but to thoughtfully integrate new forms of intelligence into human culture in ways that enhance rather than diminish our humanity.

As we continue developing these technologies, we’ll need exactly this kind of collaborative thinking—technically informed but philosophically grounded, practically focused but ethically sensitive—to ensure that our technological capabilities serve human flourishing rather than replacing human meaning-making altogether.

Related Reading:

• Can AI Be Conscious? Deep Insights from a Philosophy of Mind Discussion
• What I Learned About AGI at a NYC Meetup and Why We’re Not Ready
• How AI is Teaching Us to Speak Whale—And They’re Speaking Back
• Beyond AI Assistants: How Human-Agent Teams Will Transform Organizations
• Claude 4: The First AI Agent Boss Ready Assistant