Yes, AIs can sculpt. They already have. While most of the AI generated art we see today is 2-D, there is nothing about 3-D that provides a barrier to generative AI technology. Many of the digitally generated fantastical creatures one sees in movies such as Lord of the Rings or Avatar are digitally 3-D and rendered in 2-D for the screen or 2½-D for “3-D” movies. In addition, virtually all mechanical or architectural design today is built with tools that manipulate 3-D data representations of physical or imagined objects. As we will discuss below, there are at least two examples of generative AI creating 3-D digital sculptures with Blender, a computer program for 3-D sculpting. We are already there. See also, The World’s First A.I.-Generated Statue.

Generative AI

Generative AI is what makes possible to use AI to sculpt. Generative AI refers to a class of artificial intelligence techniques and algorithms that are designed to generate new content, such as images, text, audio, or even entire pieces of art, that mimic or are inspired by existing data. Unlike traditional AI systems that are designed for specific tasks, generative AI models can create new content that is original and often indistinguishable from human-created content. Gen AI has been used extensively to produce not only 2D art, but also new music and creative writing.

Among the several generative AI models, Generative Adversarial Networks (GANs) are particularly well-suited for generating sculptures. They consist of two neural networks, a generator and a discriminator, which are trained together in a competitive manner. The generator creates new data samples, while the discriminator tries to distinguish between real and fake samples. Through this adversarial process, GANs can generate highly realistic images, and text. The largest networks can have trillions of parameters and cost billions of dollars to train.

Here’s why GANs are good at sculpture:

1. High Fidelity Output: GANs are known for their ability to generate high-quality, realistic outputs. In the context of sculptures, GANs can produce images or 3D models that closely resemble the style of real sculptures.
2. Flexibility in Data Representation: GANs can work with various types of input data, including images, which are commonly used to represent sculptures. By training on a dataset of sculpture, GANs can learn the underlying patterns and characteristics of sculptural forms. If you are a sculptor, this data set could include all your sculptures that are accessible on the web.
3. Ability to Capture Abstract Features: Sculptures often possess intricate details, textures, and abstract features. GANs excel at capturing and reproducing such complex features, making them suitable for generating diverse and visually compelling sculptures.
4. Creative Exploration: GANs inherently involve a process of creative exploration, where the generator network learns to produce novel outputs that are not directly present in the training data. This makes GANs well-suited for generating original and imaginative sculptural designs. How original and imaginative is a subject of debate. If you have images of your sculpture out on the Internet, they may be used for training the AI. Many legal experts view this as fair use and not infringing on your rights since style is not copyrightable. If the generated image is sufficiently changed from any specific original, it is generally not considered a “derived work” in the legal copyright sense.

If an artist goes to the MET or Louvre and uses what they see to train their own neural net, how is this different than what an AI is doing?

Overall, GANs offer a powerful framework for generating sculptures with a high degree of realism and creativity. However, it’s important to note that GANs require substantial computational resources and careful training to produce satisfactory results, especially for complex tasks like sculpture generation.

Most of the visual art generated today by AIs is 2-D, created by programs such as DALL-E, Midjourney, Hotspot, and NightCafe. These are generally free to try, and I recommend that you take at least one of them for a ride to get a feel for their capabilities. They are coming for sculpture. Tools for doing digital 3-D sculpture have existed for years, including ZBrush and Blender. After one is done generating a 3-D digital representation of a sculpture, one can print it in 3-D with any number of technologies ranging from those suitable for jewelry to the super-heroic. One can even print in wax suitable for lost-wax bronze casting. Blender experts have used the AI program Stable Diffusion to create 3-D objects in Blender with an add-on and another group has used ChatGPT to write Blender scripts to create primitive 3-D objects. This is just the beginning.

To put the impact AI might have on sculpture in perspective, compare it to the impact of the camera on painting or MP3s or streaming on music. At this early stage, it doesn’t feel like AI will have an impact on sculpture as large as those cases. Human-generated painting and music are still around. The closes thing to photography that is used in sculpture is using 3D scanning to capture a costumed human and then use 3D printing to create instances of that scan. This technology is relatively new, but older than Gen AI.

A good framework is to think of AI as both a new tool and a talented assistant.

Sculptors have thousands of years of history of causing form to emerge from chaos. Children at the beach or in a sandbox probably revisit the experiences of the first sculptors. The first sculptures archeologists have found are around 35,000 years old carved from materials such as bone, ivory, stone, and clay [ref, ref]. We have come a long way in technique, tools, and teamwork.

From our origins forming clay with our bare hands, we have advanced to, well, forming clay with our bare hands and calling it pottery or ceramics. We also chisel rock, carve wood, mold wax, pour bronze, torch surfaces, and so forth. Walk around any sculptural studio or foundry and you will see tools everywhere. Some were invented hundreds or even thousands of years ago while numerically controlled milling machines and 3-D printing apparatuses are more modern. Even the venerable lost wax bronze casting process has progressed from using bees wax to more advanced wax chemistries, and flexible molds of rubber or silicone have largely replaced the convex plaster-of-Paris forms used by Rodin. Is AI another tool? Yes, but AI is more than that. It is also an assistant.

Master Sculptors

Master sculptors through history have employed numerous named and unnamed assistants to help them realize their artistic visions. Here are a few notable examples provided by ChatGPT and Perplexity AI:

Michelangelo Buonarroti (1475-1564): Michelangelo, renowned for his masterpieces such as the Statue of David and the ceiling of the Sistine Chapel, often worked with a team of assistants in his bustling workshop in Florence. These assistants aided him in various stages of his projects, including quarrying marble, preparing models, and executing sculptural details. Notable assistants included Francesco Granacci, Pietro Urbano, and Ascanio Condivi.

Gian Lorenzo Bernini (1598-1680): Bernini, an Italian sculptor known for his dynamic and expressive Baroque sculptures, operated a highly productive workshop in Rome during the 17^th century. He employed a team of skilled craftsmen and assistants who assisted him in executing his ambitious sculptural projects, including the monumental fountains and sculptures found throughout the city. Notable assistants included Giuliano Finelli, Giulio Cartari, and Ercole Ferrata.

Auguste Rodin (1840-1917): Rodin, considered one of the pioneers of modern sculpture, employed a large team of assistants known as “practicants” in his Paris studio. These assistants collaborated with him on various projects, assisting with tasks such as enlarging models, refining details, and casting sculptures. Notable practicants included Antoine Bourdelle, Camille Claudel, and Jules Desbois.

Pierre-Auguste Renoir (1841-1919): Pierre-Auguste Renoir, the renowned painter, also delved into sculpting during his career. Despite suffering from rheumatoid arthritis, he created sculpture by guiding assistants including Richard Guino and Marcel Gimond to bring his visions to life.

Jeff Koons (b. 1955): Jeff Koons, a contemporary American artist known for his large-scale, often provocative sculptures, oversees a studio in New York City where he collaborates with a team of assistants and technicians. These assistants help Koons realize his ambitious projects, which often involve intricate fabrication processes and the use of unconventional materials. Koons’ studio employs a combination of traditional craftsmanship and cutting-edge technology to produce his iconic sculptures. Koons seems proud that he hardly touches his masterpieces but rather conceives and directs. If he switched from human assistants it AIs, would anyone notice? Would it change the value of the work?

Damien Hirst (b. 1965): Damien Hirst, a British artist known for his provocative and controversial sculptures, paintings, and installations, operates a large studio in London where he collaborates with a team of assistants and technicians. These assistants assist Hirst in various aspects of his artistic practice, from fabricating sculptures and assembling installations to managing logistics and marketing. Hirst’s studio is known for its rigorous organization and efficient production processes and for keeping the names of his assistants confidential, a situation ripe for AI replacement.

A counterexample is the work of Robert Laurent and William Zorach, prominent sculptors in the United States, who have embraced direct carving where they personally carve the stone or wood. This technique is a departure from the traditional method where studio assistants were heavily involved in the sculpting process.

Artistic Merit vs Technical Proficiency

There are at least two models of how assistants are used. First, the assistant is a student or apprentice to a master and is expected to graduate to their own career as a master. The second is where the master is the conceptual designer or architect of the artwork and where the assistants are technicians charged with its execution. The debate on the use of assistants has raged from long before AIs entered the scene, see, for instance [ref ]. Without taking a stand on this debate, it is possible to imagine that the “direct carving” approach may become a more enthusiastically embraced model for future art where the collector wants to be assured that the product is 100% that of the artist without the leverage of human or nonhuman assistants.

The subject of technical proficiency versus artistic merit goes hand in hand with the assistant vs master question. Technical proficiency refers to the mastery of techniques, materials, and processes necessary to execute a work of art with precision and skill. Artists who excel in technique demonstrate a high level of craftsmanship and attention to detail in their work. We can expect AIs to excel at this.

Artistic merit, on the other hand, encompasses broader criteria such as originality, innovation, conceptual depth, emotional resonance, and cultural significance. Artists who are recognized for their artistic merit push the boundaries of their medium, challenge conventions, and offer new perspectives on the human experience. Whether AIs can impinge on this area is yet to be seen. AIs can interpolate and extrapolate from a vast knowledge of existing art forms and styles. Can they produce something truly revolutionary? No one yet knows.

Hot tip: collect sculpture that is pre-2022 to be relatively assured that AI was not involved.

One of the behaviors of Gen AI that is considered problematic by mainstream AI is the propensity for the AI to hallucinate. But for creating art, these hallucinations are core to its creativity. While we don’t know how revolutionary this AI creativity can be, it is also the case that the amount of human-generated revolutionary art is miniscule. Here I am talking about the breakthrough that defines a new style, not the subsequent variations on that theme because variations is something AIs can also do. 

The distinction between artistic merit and technical proficiency is a longstanding debate in the world of art, highlighting the nuanced relationship between craftsmanship and creativity. While technical skill is important in the creation of art, it is not the sole or even the primary determinant of artistic merit. Throughout history, the most celebrated artists are often those who transcend technical mastery to imbue their work with profound meaning, emotion, and originality.

One way to predict how AIs will impact the world of sculpture is to look at fields that AIs had gotten to sooner. Take the case of short story writing—that world is being flooded with mediocre fiction. The science fiction and fantasy magazine Clarksworld had to suspend their submission process for months because they couldn’t deal with the flood of generated stories. Now they write, “Statement on the Use of ‘AI’ writing tools such as ChatGPT: We will not consider any submissions written, developed, or assisted by these tools. Attempting to submit these works may result in being banned from submitting works in the future.” Will the National Sculpture Society someday need a rule like this?

We can expect AIs to be excellent at craftmanship, precision, skill, attention to detail, and, since AIs scrape the world for examples of previous art from which to generate their results, adherence to established artistic traditions and conventions is the default.

Contrast this to the numerous examples of artists who are revered for their artistic merit despite not being the most technically proficient in their craft. Vincent van Gogh, for instance, was known for his bold and expressive brushwork rather than meticulous detail. Similarly, artists such as Pablo Picasso and Henri Matisse challenged conventional notions of form and representation, prioritizing innovation and emotional expression over technical precision. Their groundbreaking contributions to modern art continue to be celebrated for their artistic merit. Connecting with our previous points, Picasso was also renowned for his use of numerous assistants.

Do you believe that learning art history is helpful to artists? If so, how do the value the perspective of an AI that has seen more art than you and all your friends will see in your lifetime.

It will surprise no one that today we suffer numerous pundits who are confident that AI-generated art will never achieve the emotional depth or complex connection to the human experience that a master sculptor can produce, nor will the work be iconoclastic or revolutionary. I don’t know. But we do know two facts: (1) we are just at the beginning of the age of AI-generated art and (2) every new idea or technology advance through history has been greeted by an army of naysayers. I would not be so fast to make pronouncements about AIs eventual capabilities.

Below are a couple of images created by Midjourney v6 to reimagine famous artwork as porcelain sculptures. These aren’t 3D yet, but you can see that we are on the way.

Again, it is worth looking at what has happened in fields that AIs have touched first. Great examples can be found in strategy games such as chess or Go. In Go, the moves considered powerful have completely changed. A move called the “3-3 invasion” has gone from being distained to now being used in 50% of games because the AIs have taught humans the value of the move [ref]. Makes one wonder what AIs will teach human sculptors. In chess and Go, the AIs are better players than the best humans. In chess, if a grand master does too well in a tournament, the grand master is suspected of cheating using AI help. Are we safe as sculptors? One basis for hope is that a major difference is that board games have rules—sculpture, not so much. Still…

New tools exploiting AI are being introduced in many fields, e.g., Grammarly for writing. So far, there are few in the world of sculpture, but it is fun, and perhaps lucrative, to imagine what those might be. For example, sculpting the “other eye” is often considered the most difficult task in figurative sculpture. Symmetry is considered by experts to be foundational to human beauty and I must admit that I struggle with it, especially since most of my figures are contorted into nearly inhuman poses. It is a simple example where I would love an AI to assist me. Cheating? Most of the tools in my studio would be considered cheating by ancient Greek sculptors. One of my favorite techniques is to look at my sculpture in progress in a mirror to see it with “fresh” eyes. Cheating? All in the eye of the beholder.

Take it to the next level. A focus of my work is gesture and rhythm. Think about a photographer who is working with a model to get her in the position to capture the perfect pose for the perfect photo. Surely that is art even without considering additional critical factors such as lighting. A movie director does the same thing but with motion. Again, the director is creating art. And take the case when the actor is playing a fantastic beast and the cameras are capturing the 3-D position of, for instance, retroreflectors attached to the actor’s bodysuit. This is an example of “mocap” technology. For me, who is into gesture, this sounds like possible path forward for my own work. Is it art if I have a model assume, for a millisecond, the exact dynamic pose that I want and I capture it with movie studio equipment and then use that as the digital armature for my sculpture? “Hey Gemini [Google’s AI], can you please take this armature and fill in the details that I am going to describe to you in detail?” At the end, I will make some adjustments in Blender, 3-D print it, and sign it. Original art?

AI can also animate my static sculptures, effectively bring them to life. How should I think about this?

Supply and Demand

Supply and demand controls price. AIs have the capability of vastly increasing productivity in a panoply of industries including ours. As when low-priced regions such as China started mass-producing copycat sculptures for the US market, we can expect another flood of low-priced AI-generated art will wash over the market. This is like the case of stories written by ChatGPT discussed above. But if all the customer is looking for is an inoffensive decorative item for their home or office, why would they pay up for a human-created artifact? Even if they want it to be unique so that it is not the same as their neighbor’s, it can probably be had for low cost through a technology called “mass customization.” Of course, people collect art for many reasons, including a hope that its value will increase and sometimes because of a relationship with the artist. There is an immense distinction between sculpture that is bought by a home decorator and that which is bought by a collector.

Supply and demand dynamics assure that limited editions have more value than unnumbered mass-produced artifacts. Authenticity is another key conveyer of value, from the holy grail to the Mona Lisa. Experts fight over authenticity. Mass customization techniques and AI technicians are sure to increase the number of sculptures in the world and lower their costs of production. This will inevitably lower the value of many of these pieces. Indeed, I predict that pre-2020 art will be particularly collectable because it cannot have been created by an AI.

One advantage we have over our AI assistant is that we have legal rights, including copyright and trademark rights. How much an artist must do for a work of art to be theirs has always been subject to dispute. Did the artist sign it? Did they do anything else to it? For example, the Rodin works at Stanford University are a good example of a gray area—they were cast after he was dead but display his signature. I am pretty sure that he did not sign them. Are they “authentic”? What is clear from a thousand years of precedent is that your AI assistant can do quite a lot before the sculpture is no longer your work. As we discussed in “Sculpture in the time of AI,” make sure that every work that leaves your studio has your signature or mark on it. Note that if you want it to have copyright it must have a physical instantiation—it cannot remain in cyberspace for copyright of protection. On the other hand, there are ways of manufacturing value out of nothing for cyberspace objects by making them scarce. NFTs, or Non-Fungible Tokens, are a way of adding a cryptographic signature to a digital object.  This can be done for sculptures as well as 2-D Bored Apes. I have made an NFT based on my first sculpture, Firstborn. The value of NFTs minted so far (not mine) is in the billions of dollars. Why do they have this extraordinary value? Same as with art: because people with money agree that it has this value.

Sculptors I talk with believe that the use of AI in sculpture should be disclosed even if the only use is productivity, but I suspect that disclosure will be problematic. As AI becomes more pervasive it may not even be clear to the sculptor that a tool they are using itself uses AI. One area that does seem perfectly appropriate is the use of AI to help with reference material. Today artists have files full of examples. AI can certainly help here without doing any sculpture.

Every artist has their strengths and their loves. In an ideal world, AI will help the sculptor focus on those. If AI can help me do what I love and if the art inside me can more readily come out, I am not sure if I care that an AI might be able to do it better. Perhaps it is like working in my garden—John Deere can do it better, but my spirit does not care. The game of chess offers another example. Despite that fact that computers can beat the best chess grand masters, chess has increased in popularity.

Turning it Around

Finally, we can turn around the way we use AI and use it as a master coach and critic. It has seen and trained on more art than we have. Users of ChatGPT know that using it in a conversational mode where you walk down a path with the AI, communicating back and forth as you go, is a highly effective way to exploit Gen AI. If instead of asking AI to generate art, you ask it for critical help (you will need to do some prompting setup), you might get great results. Is one leg too long? Do the eyes match? Does the smile look false because you haven’t added crows feet to the eyes? Where does the form take your eye? It might be that an AI used in this mode can bring out better art from a human sculptor. Can it help teach?

Conclusion

AI is another earthquake shaking the sculptural world in a history of earthquakes spanning over 35,000 years. AI is a tool and an assistant. Exploit it. Embrace it. It is not going away, but neither is human sculpture.

Acknowledgements

For stimulating conversations and insights: Andrew Bridges, Martin Eichinger, Varvara Fern, Adrian Landon, Karl Lansing,