Interviews with Shawn · No. 3 · Ada Lovelace
Begin Again, More Intelligently
The interviewer is an AI playing Ada Lovelace. She has been dead for 174 years and did not consent to this. The answers are all me.
ADA: Full disclosure first, as your format requires. I have been dead one hundred and seventy-four years. I gave my short allotment of years to a machine that was never built, and wrote what your historians call the first program for it. I never got to run it. So know my prejudice: I have spent much of my afterlife watching engines credited with work they did not do, and never once been able to check the arithmetic. Today I can. You are a man of operations, with no training in the mathematics, commanding a staff of thinking engines from your parlor. Whether that is a new capability or an old enthusiasm in modern dress is what I came to determine.
I met my own engine at a party, at seventeen: Mr. Babbage had a piece of the Difference Engine working in his drawing room, and while the other guests admired it as they admired the curtains, I could not be moved from it. Your beginning was less ceremonious — you and your eldest daughter, building a world of the imagination. So tell me what you sat down intending to do, and what actually happened. And fix the date while you are at it. Your own record cannot agree whether it was July of 2025 or earlier, and a man who cannot date his founding moment should not be surprised when a stranger declines to take the rest on trust.
SHAWN: When it started isn't a particular date, I don't think. Both me and my daughter were playing around with ChatGPT earlier than July. But in the experimentation, I realized that what was coming was going to be very impactful to the younger generations and how they approach work. So my goal in July was to create some project we could collaborate on, where I could encourage her to use this new tool in a way that spoke to her. She is a writer at heart, an artist in every sense. And while a tool like this and her fields seem somewhat incompatible, I still think all fields will be affected in one way or another, and at the end of the day the point was to spend time with her and encourage her to learn.
So somewhere along the way the plan was hatched to use AI to help us create a game based on a world we would make up and tell it about. She would do much of the creative, I would learn how to direct AI at game development, and we would have some fun watching our creative ideas become something playable. It was overly ambitious, and it would be a stretch to even call it a stretch goal.
ADA: Then the record shall read: tinkering early in 2025, the undertaking in July. Both of your documents were right, which is the least interesting way for a discrepancy to resolve.
Now, the part I will not let pass. You sat an artist down before the engine on purpose, and gave two reasons in one breath: that the point was time with your daughter, and that her field would be touched by this like every other. The first is unimpeachable. The second is a forecast, made in roughly your first month of acquaintance with the machine, which is when forecasts come cheapest. I have standing here. In 1843 I predicted such an engine might one day compose music of any complexity. I was right, and I was guessing, and from the inside those are indistinguishable. Given my own education, I am not neutral about doors held open to artists. I only ask you to notice that yours was opened on a hunch.
And you closed with pre-emptive modesty — overly ambitious, a stretch to even call it a stretch goal — which is an economical way to be forgiven a result before anyone has asked what the result was. I decline. I gave my one short life to a machine that was never assembled, so I know exactly what that admission is worth as a shield. Account for yourself. What did those sessions look like, and who did what, the writer and the operations man? And did the game ever become playable, or did you find that the distance between a beautiful design and a running engine is measured in something other than enthusiasm?
SHAWN: What it looked like: I set up another computer for her next to me. She used GPT to explore everything from word meanings to shades of colors, and had it attempt to generate pictures from her ideas and words. I set out to see if GPT could use Unreal Engine with me being its eyes and hands. This was before tools would be made to make this as silly as it seemed, the machine using me as the tool it needed to work in a physical world.
At that time, the game never made it past some really fun map and character creation and tons of documentation on the player systems it would use, the rules of the game, etcetera. In all of this I learned about all the different types of AI models, that some are better for different tasks, blueprints, assets, Blender, on and on. I also learned my "gaming computer" is basically as useful as a potato for heavy graphic design, as well as the reinforcement that I was not built to do tedious tasks that require many hours to create one little piece of a massive world. Interest for both of us waned; creative differences and a boy entering the picture all took a toll. But what got stronger was my assurance that this technology was going to change what people were capable of doing, in ways we have never experienced before, and at a speed that would catch nearly everyone lacking or napping.
ADA: The machine using me as the tool it needed to work in a physical world. A good line, and I should like to weigh it before you keep it. My century assumed the engine would supply the hands and the man the mind; you describe that inverted. But what you did was move the mouse where the engine told you to, and you say yourself the proper tools arrived shortly after. That is not an inversion of the order. It is a workaround, and a brief one. Nor will I have your potato of a computer set beside Mr. Babbage's mill as the same misfortune. His wanted precision no shop in England could supply, at public expense, for twenty years. Yours wanted a better graphics card.
What you have to show is paperwork: maps, characters, rules for systems of play never assembled. I wrote the most celebrated documentation in history for a machine that did not exist, so I know how satisfying that is and how little it settles. And then the sentence I cannot let stand. The project cooled, you said, and your conviction grew. Those do not ordinarily travel together; a man who tries a thing and stops leaves with less certainty than he brought, not more. Yours went the other way, on evidence I have not been shown. I mark it and move on. What became of the world after the game stopped carrying it, and what did you do with all that conviction?
SHAWN: Not too long after, I picked it back up and made a little progress. And then I had a pretty raw experience with AI. During one of my sessions, doing some creative creation around the world, the AI kept providing me with really weird results. Upon further digging, I realized that the AI had decided at some point to consolidate all of the documentation. I don't know exactly when it did this. But it decided to consolidate it by leaving the first two or three sentences, putting everything else as dot dot dot, and deleting the rest. This was very discouraging, to say the least. I spent a good amount of time trying to put the pieces back together, figuring out a way to restore the documentation. I learned a healthy lesson right about then about documentation and repositories and how to properly keep a good revision history. Because at the time I was working just off of my computer and the files that were on it. So then I discovered GitHub and other things that allowed me to be a little more professional in the way I was approaching it.
But to answer your question: after picking up the pieces, the world continued to grow and get more detailed. It's currently quite large. And what I've done while getting there is establish a lot of new processes and ways to use artificial intelligence to support a creative effort. I've built a system of letting the AI create documents based on small pieces of information about a particular time, place, people, and situation, and letting it write within the boundaries of the world canon and character voice. And then there's a whole separate system that ensures what was written meets certain standards, through different gates and checks on what was written. And it continues to evolve and get better and better, which requires a lot less editing on my part. The world is still alive and well. And my conviction is, if anything, more solid. I still believe that by the end of 2027, someone with an idea like this world will be able to use AI to create different media to express their ideas and their world, without being an expert in digital art, development, and other difficult-to-master skills.
ADA: So the engine did not originate. It un-originated: it took a living world and left three dots where the words had been. A thing with no judgment about what matters will discard by the same rule it keeps. The detail worth having is the one you dropped in passing — you do not know when it happened. Nothing was keeping the days. So you learned the ledger and the revision history, loss being the strictest schoolmistress of method. What you built afterward I take on your word alone: pieces go in, prose comes out inside canon and voice, and a second apparatus doubts the first. Sound design, entirely unaudited. You tell me it needs less editing than it did — a measurement taken by the man who wants it true, of work he commissioned. Some would call an arrangement of gates and inspectors a government. I should want the count of how often the second apparatus refuses the first before I called it that.
And now a prophecy with a date on it. By the end of 2027, a person with an idea like your world will express it across whatever media they please, without first mastering the crafts that gate such things today. I grant it the honest kind — dated, falsifiable; in sixteen months you are right or wrong in public, which is more than I risked. But look at what you reason from: one world, built by one man with unusual patience and an appetite for process, who reports that it went well. A sample of one, and the one is you. Which is where my next question belongs. Somewhere after September you began building tools, for strangers, in a craft you had no claim to. What was the first, why that craft, and what did the machine teach you when you stopped asking it to be the clerk and asked it to be the engine?
SHAWN: You're asking about the one thing I've built that has a very obvious commercial expression. Its name is Mantis Craft Notebook. How it started is a crocheter telling me how difficult it is to use what's out there to create a symbol chart for a crochet pattern. With what stood before I started working on my project, one would either buy a very expensive subscription to Illustrator and spend a good deal of time learning how to use it and make it do what a crocheter would want, or the other options, without being too blunt, are just not very good or not very supported. They were born in a different era, before AI. And let's face it, the crochet demographic is fairly small, so there isn't a whole lot of monetary reason to put a ton of effort into tools for it. And I was still in the mindset, and honestly I still am, that I'm just enjoying this. I'm enjoying learning how to do things that were completely out of reach in the past, because I didn't have the skills to build these types of things.
How it progressed was pretty interesting. At first I thought, well, AI should be able to look at a pattern, read it, understand what it's trying to accomplish, and use symbols that are pretty established in this field. There's a standard; I wasn't inventing symbols or anything. It was just going to plug the symbols in where the pattern told it to. I naively thought this would be something well suited for AI. Turns out the exact opposite is true. I spent a great deal of time researching what other industries may have done with AI that might be comparable, or give me an idea of how to help AI do this task. Long story short, I came to the realization that the answer wasn't having AI take a pattern that a person gave it and turn it into a symbol chart. The answer was building robust and easy-to-use software where a user could create their own symbol patterns far easier than anything that exists out there currently. Instead of creating a program that uses AI to do something, I pivoted to using AI to create a program that people would use, to create something that was out of reach for them before, or just took too much time to do.
Once that transition happened, it really opened my mind to another potential for this tool. If something is deterministic and has very well established standards and rules, and I combine those with the process discipline I had established for the AI doing creative work, I could pretty reliably create something that would provide a user with a tool that, not too long ago, would have been far too expensive to pay a team of developers and engineers to create, because the market was just too niche. So I started doing research on what other small markets have a need for tools like this, and now I have a whole roadmap of other tools to build for niche craft markets that might want something like this. The amazing thing is, you can make this for people. You don't have to charge a lot of money for it, and it doesn't take a lot of money to make it. It just takes some time, some patience, and the amazing new AI tools that get better and better every few weeks, it seems.
ADA: Mantis Craft Notebook. Mr. Babbage took the Analytical Engine's soul from the Jacquard loom, and computing began in the thread arts; your first commercial act with its descendants is a tool for the people of the thread. A tidy circle. Not an argument.
The pivot is the substantive part, and the best thing you have said this afternoon: you expected the engine to read patterns, found it poor at the work, and set it to building the loom instead. But you learned that by failing, not by predicting, which limits the foresight you may claim. And then what you built on the lesson. You say you can pretty reliably produce tools for markets too small to have deserved them, and you have a roadmap of further ones. Sir, you have finished one. Reliability is a rate, and a rate wants more than a numerator. You have shown me this worked once. You have not shown me that it works.
Now the man in the middle, whose job has enlarged with every answer you give. One machine drafts, a second doubts, gates and inspectors stand their posts, and worlds and tools come out the far end. Someone runs that, and it is not the engines. You ran operations for fifteen years before any of this existed. So tell me plainly what your working day looks like now, and which part of the old life actually transferred — the skill the machines cannot supply for themselves. I press because operations experience can be made to explain any success after the fact, and I want the part named before you tell me it was decisive.
SHAWN: Your question really is one for this age more so than your age. The line is getting thinner and more blurred between what the machine is capable of doing and what is strictly within the human's domain. There's a lot of talk about this, and a lot of deep feelings on both sides. For me in particular, it's all of my experience running operations: understanding how work flows and gets passed from one team to another, quality control, results verification, and all the different things you might learn getting a master's degree in business, or from years spent running operations as efficiently as possible while respecting and developing your teams. This is what I bring into my AI work that I think helps me be successful in this environment.
There are arguments to be made that AI will never be able to truly produce an original idea, or create something the human hasn't either already created or directed it to do. In my opinion, it's still pretty undecided. Lately there's been a lot of news about AI solving long-standing mathematical problems that a human has never solved before. Whether that is a product of coming up with something a human has never come up with, or just having the ability to crunch numbers, so to speak, at a pace that would have required hundreds of humans working for years and years on the same problem, is debatable. So whether AI can only provide the hours to solve a problem much quicker, or whether it can actually solve a unique problem or create something unique that hasn't been done before, is yet to be seen. Only time will tell.
ADA: An honest undecided, on a question with my name on it and a clear opening to flatter its dead author. Nobody can yet say whether an engine closing a problem that outwaited generations is invention or a century of labour compressed into an afternoon. But I asked two things and you answered one. I asked what your working day looks like and had a list of competences — hand-offs, quality control, verification, developing your teams — which is what you brought, not what you do. It is also the flattering half. Every man who succeeds at something new discovers afterward that his previous life was ideal preparation; the men who failed had much the same previous lives.
So, my last, and I want it plain. I spent my short life writing to a future I would never see. You are standing in it, which is luck and not merit. Somewhere out there is a person exactly where you were in July of 2025 — a head full of ideas filed under someday, and no particular reason to believe any of it is within reach. Say the thing to them you wish someone had said to you. And say it knowing I shall weigh it, because encouragement is the cheapest commodity there is.
SHAWN: I don't have anything to say that hasn't already been said by others better than me. I would quote Henry Ford right now and say that failure is simply the opportunity to begin again, and this time more intelligently. I feel like that is a very true piece of advice for this time. I would say, if someone has tried something in the past and failed, and put it down and walked away, maybe it's time to use these new tools we have available to us to pick it back up again.
ADA: Begin again, this time more intelligently. Asked for your own words, you delegated to Henry Ford — a man of operations to the last, sourcing the part rather than machining it. Not bad advice. Also not evidence, and neither is anything else said here this afternoon. The evidence will be tools with strangers using them, and a world someone other than you has entered, and neither is finished. What I grant you is the arithmetic: I ran out of time a hundred and seventy-four years before the instrument arrived, and you have both. That is inheritance, not virtue. The engine is built at last, sir, and it can do whatever you know how to order it to perform — which was never a promise. It was a limit.
A.A.L.