All Glossary

AI in Ceramics

This is huge and exploded into the mainstream in 2023. Industry figures are campaigning for trillions of dollars to build dozens of chip factories and hundreds of data centers (and the massive electrical infrastructure these will require). The goal is to create AGI (artificial general intelligence) - the degree to which that goal can be taken seems limitless. Some call it the most complex and ambitious task in the history of humanity. Strangely AI is turning out to be especially good at creativity and the liberal arts. But only sometimes great on technical matters (see examples below about how dead wrong it is about some things). While so-called "Prompt Engineering" is emerging as a new science of learning how to interact with AI to get the best possible results it likely won't last long, we will all have to learn it as part of our jobs.

Ethical questions also abound here. Unfortunately, humankind is riding this horse of AI and we don't really know where it is going! Many have inaccurate and uninformed opinions about how it works, whether it is real and to what extent it is already touching their everyday lives. First - pretty well everyone is likely underestimating AI. We know for sure that it is not going to get worse, it is only ever going to get smarter. Second, the unstoppable AI tidal wave going mainstream is math and it is out of the bag. There is no stopping or regulating it. Third, people who use AI are going to do it in such a way that we won't know. And it will edge its way into our lives from all sides. The smarter it gets the less able we will be able to recognize its role. But one thing that few people are talking about is that AI requires computing power akin to what Bitcoin mining took. Actually, that is wrong. The AGI (artificial general intelligence) goal needs one hundred times that, many trillions of dollars of investment in chip manufacturing, data centers, power plants and network bandwidth. And the software in current models (in 2024) uses 150 billion algorithmic parameters. But that is going to one trillion! nVidia is riding the AI wave like no one else, they could become the world's largest company.

People who feel that AI is just a plagiarism machine base that on the belief that it is not actually learning, it is either just copying or is a word prediction system. However, a fundamental aspect of modern AI (as opposed to old-style expert systems) is that it does learn. Modern AI is far beyond just generating words. It is making music, art, photos, video and movies, writing poetry and so much more. One may argue that it simulates learning and does it crudely and inefficiently. Yes, its outrageous carbon footprint is proof. But remember, it is only ever going to get better, not worse.

Although many object to the use of AI as a creative tool, mountain that most of us struggle to climb is making a website. AI tools can do that completely automatically now. This kind of creativity seems helpful, doing something that pretty well none of us enjoy. And it is fast. And the results are better than 95% of the sites that artists and potters make (or have made) to promote their work. Maybe this is one place where we could tolerate the use of AI.

Related Information

These pots are not real. How is that possible?

I created these images using the "imagine" command at MidJourney.com. To me this is jaw-dropping - I thought AI was just plagiarizing but it really does appear to be learning. Top left: I asked for "stoneware pottery mugs with floral decoration covered by a glossy transparent glaze". For the vase: "Crazed transparent glaze over floral decoration". How does it know what crazing is? Notice the things I forgot to specify for the porcelain mugs: Handmade, one handle, decoration should only be on the outside and handles are not decorated (these errors seem like an indication that this AI has not learned that yet). For the last one, I asked for a piece demonstrating the difference in runniness of two different glazes. How does it know what runny glazes are? How does it know a bowl would be the best shape to demonstrate the latter? These observations and the fact that it can generate very high-resolution images seems to indicate it is actually drawing the pieces, not just showing photos it found online.

This impossible AI porcelain mug suggests it is actually learning

Consider these porcelain mugs I "imagined" at MidJourney.com. At first, they look pretty typical but take a closer look at the one in the middle. The floral design on the one behind morphs into the actual 3D shape on the one in front. This seems like an indication that the AI is actually drawing these, not just plagiarizing them from its scraping activities online. The #aiart, #aiartwork, #aiartcommunity and similar hashtags generate millions of posts (albeit of questionable taste), but they are a harbinger of things to come.

ChatGPT is completely wrong about the cause of glaze crazing!

Today, ChatGPT is parroting common wrong suggestions about the cause and solution of the serious issue of crazing. Yet it trained on thousands of internet pages about the subject! Crazed functional ware is defective, and customers will return it. So fixing the problem is serious business, we need correct answers. Consider ChatGPT's suggestions: #1 is wrong. There is no such thing as an "incompatible mix" of ceramic materials. Crazing is an incompatibility in thermal expansions of glaze and body, almost always a result of excessive levels of high-expansion K₂O and Na₂O in the chemistry of the glaze. The solution is reducing them in favor of other fluxes (the amount per the degree of COE mismatch). #2 is wrong, firing changes don't fix the incompatibility of thermal expansions. #3 is wrong, refiring makes the crazing go away but not the stress of the mismatch, it will for sure return. #4 is completely wrong. Firing higher takes more quartz grains into solution in the melt and should reduce the COE (and mature the body more which often improves fit). And melt fluidity has nothing to do with crazing. Furthermore, if a glaze does not run off the ware, it is not overfired.

ChatGPT woke me up about making my own frits.

I asked ChatGPT this question and got a very thoughtful answer that seems to confirm what I have observed in smelting material mixtures into ingots in alumina and zircon lined slip cast crucibles. The mixtures are melting well in the test crucibles (the upper one was fired at cone 4, the lower one at cone 6). But there are issues. There does appear to be phase separations. And bubble froth at the top. For some compositions alumina works better as a liner, for others zircon. We are getting closer to trying larger multi-kilogram batches and ball milling so time will tell.

ChatGPT is surprisingly wrong about the causes of glaze crawling.

ChatGPT trained on the entire internet and yet gave 100% wrong answers and neglected the key thing that causes 90% of crawling! How can the internet be so wrong? Consider the suggestions it gave: -Dust or oil on the bisque: This almost never happens. Besides, glaze is a mix of dust and water! -Too much feldspar in a glaze can cause it to shrink excessively during firing: No, high feldspar causes thermal expansion/contraction of the fired glass, not physical shrinkage of the melt. -If the clay body surface is not roughened, the glaze may not adhere properly: No, glazes don’t crawl any more on porcelains than other bodies. -If the glaze is too thick in some areas and too thin in others, it can crawl in the thin areas: No, it crawls where thick because that’s where it cracks during drying. -Over-firing or under-firing: No. Glazes fired to the ideal temperature crawl just as much. -If the pottery is dried evenly or the drying process is too rapid: No, rapid drying of glaze on bisque is important to prevent cracking. This crawling happened because the glaze cracked along the inside of that corner during drying. Such cracking is by far the number one cause of crawling, the melt pulls back from either side of the crack. The specific gravity of the slurry was too high, the resulting greater thickness right at the corner gave the shrinking glaze power to pull a crack. Adding water to bring the SG back down to 1.4 and then Epsom salts to gel it to thixotropic gave the slurry much better dipping and drying properties and totally solved this issue.

An entire website created with a one-sentence request

This was made at 10web.io. I simply wrote it a couple of sentences describing a website that explores the connections between ceramics and fashion and one that could be used to promote the making of garments having ceramic elements. In about a minute it generated this! It is pretty weird, but a great starting point. It is built on top of WordPress so I get the normal dashboard with all the authoring tools (and optional add-ons). None of these people exist! Or things. This site adapts to phones, tablets or PCs. And it is fast.

Another pottery website generated entirely by AI

This was made at 10web.io. I simply wrote it a couple of sentences describing a website that sells pottery mugs made to celebrate that special pet dog. It is built on top of WordPress so I also get the normal dashboard with all the authoring tools (and optional add-ons). Neither this business or any of these dogs or people exist! But the idea exists in my mind and the AI is a starting point to making it real.

By Tony Hansen Follow me on

Got a Question?

Buy me a coffee and we can talk