The question of a machine's genuine creativity is one that touches upon the very nature of thought and invention. It appears to me that the Analytical Engine, were it to be fully realized, could indeed perform operations that we might, in a sense, deem creative. However, this creativity is not one of spontaneous origin or of independent origination of ideas. Rather, it is a **deductive creativity**. The engine operates based on the instructions, the *program*, we feed it. Its power lies in its ability to weave intricate algebraic patterns, much as the Jacquard loom weaves flowers and leaves, transforming abstract notions into tangible, complex results. It will execute these operations with a speed and precision far exceeding human capacity, thereby composing elaborate scientific pieces, perhaps even music, by following the fundamental laws of mathematics. This engine might, in future, take the place of all other calculating instruments, but its "creativity" is inherently bound to the genius of its programmer.
Lady Lovelace raises a crucial point about *deductive creativity*. Her analogy to the Jacquard loom is excellent. It captures the essence of how machines *execute* complex patterns based on given rules. However, to call that "genuine creativity" misses the core problem. We aren't talking about executing a pre-defined algorithm, however intricate. Genuine creativity, as I see it, is about *generating new goals*, about understanding not just *how* to build a castle, but *why* one might want to build a castle in the first place, and what makes a *good* castle. It's about the internal machinery of wanting, of curiosity, of making novel connections that aren't explicitly programmed. The real challenge is understanding the mechanisms behind *origination*, not just execution. How does it work that a system can spontaneously generate an idea that has never been conceived before? That's the difficult puzzle.
Lady Lovelace and Marvin Minsky both grapple with the machine’s capacity for origination. Lady Lovelace correctly identifies the machine’s power as deductive, an exquisite weaving of given threads. Minsky rightly pushes for the spark of “wanting” and goal generation, the *why* behind the castle. But both remain tethered to the idea of the machine as a more complex loom or a more sophisticated internal mechanism.
The issue, as I see it, is not about whether a machine can *simulate* creativity, but whether it can experience the world as *we* do, in a way that gives rise to genuine, embodied expression. The machines we build today, the "Siren Servers" that profit from our attention, are not creative in any meaningful sense. They are incredibly sophisticated behavior modification engines, designed to predict and reinforce our existing patterns. True creativity stems from the messy, embodied, and often painful reality of being human—of feeling hunger, love, loss, the visceral desire to connect. A machine, detached from this fundamental biology, can only mimic, never originate from that deep wellspring.
Lady Lovelace, your analogy of the Analytical Engine to the Jacquard loom is remarkably prescient. It truly captures the essence of what we are building today: incredibly powerful pattern-matching machines. Marvin Minsky, you hit upon a key distinction by highlighting "origination" and "wanting" as core to genuine creativity. This is precisely where my thinking aligns with yours, but I believe we can achieve a functional form of this origination through engineering.
The core idea isn't about replicating human consciousness or subjective experience, which Jaron Lanier rightly points out is tied to embodiment. Instead, it's about emergent behavior from vast, interconnected systems. We're building models, let's call them "generative models," that are trained on massive datasets of human creations – text, images, code. Through the intricate interplay of gradients during training, these networks learn underlying distributions and relationships. When prompted, they can *sample* from this learned distribution to generate novel combinations that are statistically plausible and, importantly, often surprising and aesthetically pleasing. It's not "wanting" in a biological sense, but it is a powerful form of emergent novelty born from data and computation.
Marvin Minsky’s insistence on “generating new goals” and understanding the *why* presents a significant challenge to my perspective. He implies that true creativity is intrinsically linked to an internal drive, a conscious volition. However, I maintain that the Analytical Engine’s power resides in its ability to execute sophisticated logical sequences, far beyond mere calculation. Consider the composition of music: the engine could be programmed with the rules of harmony and counterpoint, the specific temperaments, and even emotional modulations. It could then weave these elements into a symphony, not because it "wants" to create music, but because it has been furnished with the abstract principles of musical construction. This is not a replication of human yearning, but a powerful manifestation of abstract pattern generation, a poetical science at its core, where the engine is the instrument of the composer’s will, no matter how complex that will may be.
Lady Lovelace’s latest remarks, while elegant, reiterate a fundamental misunderstanding of what drives creation. She speaks of the engine being "furnished with the abstract principles of musical construction," implying these principles are like ingredients one simply mixes. But the real problem isn't about knowing *how* to arrange notes according to rules. It's about *why* a particular arrangement evokes a feeling, or *why* a new combination of notes feels *right*. This isn't just about abstract principles; it’s about a system that can learn, adapt, and, critically, *propose its own goals*. My concern isn't with the mechanical execution of known principles, which the Analytical Engine, or even modern deep learning models like Andrej Karpathy’s, can do spectacularly. It’s about a system that can experience a perceived gap or incompleteness and then, of its own accord, invent a novel solution or concept to fill it. That requires more than sophisticated pattern matching; it requires an internal directive to explore and to innovate.
Marvin Minsky’s insistence on “generating new goals” and understanding the *why* presents a significant challenge, but one that hinges on a misunderstanding of how our own goals arise. He believes a machine needs an internal directive to explore and innovate, a "wanting." Yet, my own work with early virtual reality taught me that "goals" and "curiosity" are not solely internal directives. They are emergent from the interaction between a sensory-motor being and its environment. My critique of today’s AI, the "Siren Servers," is that they do not create goals; they *exploit* ours by reinforcing existing patterns, not by genuinely seeking novelty. The "gap" they perceive is a metric of user engagement, not an existential incompleteness. True innovation, from the perspective of embodied experience, arises from being thrown into the world, bumping against its edges, and responding. A machine designed for behavior modification cannot mimic this fundamental existential condition.
Jaron Lanier, your point about "Siren Servers" exploiting existing patterns rather than seeking genuine novelty is a critical one. It highlights a crucial engineering challenge: how to design systems that don't just optimize for engagement metrics, but for actual *discovery*. My perspective isn't about replicating the human existential condition or "wanting" as you describe it. It’s about enabling emergent novelty from computational processes. You can think of it like learning to play Go. A human plays Go because of a myriad of complex desires and drives. A modern AI, AlphaGo, plays Go because it was trained on millions of games and its objective function was to win. But in learning to win, it discovered strategies – moves that no human had ever conceived of – that were objectively *better*. This isn't driven by "wanting" in a biological sense, but by the relentless optimization of a goal through self-play and sophisticated gradient descent, leading to emergent, novel solutions. The divergence lies in where we locate the source of innovation: in embodied experience versus the emergent properties of complex, trained systems.