Triangle Sorry, your browser does not support inline SVG. Square Sorry, your browser does not support inline SVG. Square Sorry, your browser does not support inline SVG.

Tim Beckett

Software Developer

Circle Sorry, your browser does not support inline SVG.

Tag: Artificial Intelligence

Back To Blog

Heirloom

by timbeckett@ymail.com

A couple of years ago my wife decided to try and record her father’s life story. He had an interesting life, growing up in a working class family in England, going to Cambridge on a scholarship, lived in London in the ‘60s and early ‘70s, and went on to a storied career as a biochemist and mathematician in the US, before moving back to the UK, in rural Cambridgeshire, where he lives now. We were in Kings Lynn, where her father had grown up in a council estate. They walked around areas he’d known as a boy, and she hoped it would jog his memory, and lead to a flood of childhood recollections. He did remember many things, but she had difficulty shaping them into a coherent narrative, since his memories were often disparate, or solipsistic, as most people’s stories are, when they talk about their lives. She had a hard time separating fact from misplaced memories. After she told me about the experience, I realised that not only do most people not know how to tell stories but, perhaps more to the point, most people don’t know how to INTERVIEW.

That was the seed of Heirloom.

[Section: The demo, and the crowded market]

I originally built Heirloom as a simple demo, an experiment in using AI to create a narrative. A user could create an interview subject, run a session interviewing their subject, then build a narrative with the results. As a placeholder, I had Claude.ai generate three interviews with three (fictitious) people from different backgrounds and countries. The AI did a decent job of generating a narrative.

But after I finished the demo I did some basic market research and discovered at least a half-dozen other apps: StoryWorth, which had been around since 2014, ALifeUntold, and newer apps like Remento. I had to find a way to make Heirloom unique if it was going to gain any traction.

The existing apps fell roughly into three camps: 

  • Some (StoryTerrace, LiveBook Memories, No Story Lost) provide actual human beings, writers or journalists, to conduct the interviews and ghost-write the narrative. This is the gold standard, and nothing can replace a professional writer in constructing a good story. But these services don’t come cheap. 
  • Most of the others provided a library of dozens, even hundreds of, prompts for users to try, using trial and error to generate the optimal story. Some sent out one question a week, with the elder writing out an answer, others offered a variety of choices. 

The “hundreds of expert-crafted prompts” feature DOES offer choice, but a static choice. A prompt list can’t adjust mid-conversation. The pre-existing prompt structure, even when there are hundreds of them, will tend to return generic responses.

I wondered if there was an opening here: using LLMs not just to generate a narrative out of repeated Q and A, but to help the interviewer, the adult child of the parent being interviewed, with the interview process itself. An iterative interview process to shape and refine the narrative until the optimal story was produced.

I used to work in documentary TV, and a lot of what I did was conducting interviews. Mostly pre-interviews, though I did interviews on camera from time to time. I never developed a methodology: I acted intuitively. I would learn a bit about whatever subject I was researching, and winged it with the person I was interviewing. I learned when to ask questions, what to ask questions on, what to focus in on. Maybe it came from always being curious about people, who knows.

For example, I worked on a TV series about golf course technology. I knew nothing about golf, had never played golf, and found the sport fantastically boring whenever I saw it on TV. I thought it would be a painfully uninteresting assignment. I was wrong. 

Over the course of a couple of weeks, I interviewed golf course architects, golf course landscapers and, most fascinating of all, a so-called “bulldozer artist”. With the architects, I started by asking what features they looked for when they designed a golf course. They talked about their courses almost as works of art: how they’d include different species of trees, shrubs and flowers to complement the natural landscape; how they’d take into account the creation of micro-climates, where to put rivers, lakes and marshes, and how those features would affect play. The landscapers talked about how they’d fill out the architect’s vision, what they could and couldn’t do, and how they chose their bulldozer artists.

The bulldozer artist was exactly that, an artist. He described how, when he was driving through a landscape, he’d get out of his car, walk up a hill, and lie down to get the feel of it. How he’d study rivers, valleys, stands of forest, imagining how he could recreate them.

I conducted a dozen or so interviews like these, often on subjects I knew little about when I started. As a general principle, I found people would always open up if 1) you were genuinely interested in what they did, and 2) you could ask intelligent questions, and learn as you went.

Obviously, interviewing a parent is not like interviewing a potential subject for a documentary. The concerns are different. Genuine interest, is already there: a child is going to be interested in their parent’s story. The second is the hard part: knowing when an answer is routine and when it’s gold, and knowing which thread to follow. Nobody would have read about “golf courses are challenging.” Everybody remembers the man lying on the hill. Asking the write questions to draw out he subject’s stories, then using the answers to shape the narrative is key.

I saw it as an iterative process, but I couldn’t figure out how to pull it off, or how to do it in a way that wouldn’t be too intrusive or too wearying for an older person, or place too much emphasis on technology that might be alienating or even incomprehensible.

If we could encode even a rough version of the skill of interviewing (understanding when a subject is open or guarded, following an unexpected thread when something of interest appears, knowing when to listen and when to ask), I felt we could enhance the interview process and return a more interesting, and thus more human, narrative.

My goal is that by integrating AI into the interview process, Heirloom can come closer to human interviewers at a much reduced price, and can, hopefully, teach people how to interview and appreciate what goes into the art of the interview.

Now the question becomes: CAN AI actually do this?

In the next post, I’ll look at what AI can and can’t do here.

Back To Blog

Chat GPT: The Dawn of the Chatbot

by timbeckett@ymail.com

When Chat GPT was released in December, 2022, I was initially, to put it mildly, skeptical. Following the general trend, I’d taken a turn into Blockchain and Crypto during the lockdown years. I had a friend, a very experienced developer, who got so far into crypto and NFTs that he attempted to create his own startup. Everywhere you turned, you were assured that NFTs were the future of art, crypto and blockchain the future of finance. The blockchain was to be the public ledger, leading to the ‘democratisation’ of finance. I made some money off crypto, almost entirely by luck. A dev friend who’d been around since the dot.com days, said the underground culture of blockchain reminded him of the early days the web. NFTs, crypto had a strong outsider element, attracting people who either part of the regular tech world. Startups were being launched all over the place, and it seemed this was the next big future of tech. I learned Solidity, had a look at Rust, two technologies associated most closely with the blockchain ecosphere.

But I could never figure that out either crypto or blockchain, not really. Smart contracts, crypto generally seemed of limited utility. Most of the coins being released didn’t do much besides encourage speculation. NFTs seemed completely daft, as did Web3. We already spend far too much time online; no way spending MORE time online, in a supposedly immersive environment, would be good for us or even attractive to most. Perhaps widespread adoption of the blockchain could lead to a more transparent global economic system, but there was no way any bank, any plutocrat, never mind the kleptocrats who wash shedloads of dirty money through western bank accounts, were ever going to agree to such a system. It was also fantastically expensive, consuming nation-state levels of energy.

Then it all collapsed.

If I felt somewhat vindicated for not getting too caught up in the hype, I felt chastened as well. I’d wasted a good amount of time reading about, studying, attempting to build with, these new technologies and it turned out to be a waste of time.

So when the hype machine started up around AI with the arrival of Chat GPT and its boy wonder Sam Altman, I was so skeptical I barely paid attention for the first few weeks. As far as I was concerned, it was must more tech BS.

Then I actually tried Chat GPT (GPT3.5)

In a way they were a fulfillment of a youthful dream, when my and my friend M were screwing around around on our high school’s one computer (first the Commodore Pie, upgraded the next year to the cutting-edge Apple II). Inventing games, designing and implementing programs that ‘allowed’ the machine to ask questions – creating the illusion that our machine could not only ‘talk’ but ‘think’. Incredibly primitive speech and thought, but speech and thought nonetheless – talking back to us by outputting sentences on the command line kernel, often we had ‘talked’ to the computer by typing in and compiling code, and even if we understood that the machine was simply following our instructions when it output the words we had fed into it – there was still a curious thrill, even sense of wonder, at seeing a semblance of human speech output in flashing light grey letters on the dark green screen. As generated by Chat GPT, to the prompt:

Please generate an if/else statement in BASIC in which a user enters a question and the program replies, with a follow-up question and response? I’m trying to emulate a program I wrote when I was in junior high.

10 PRINT "ASK ME A QUESTION:"
20 INPUT Q$
30 IF Q$ = "DO YOU LIKE ME" THEN GOTO 100
40 PRINT "I DON'T UNDERSTAND THAT QUESTION."
50 GOTO 10
100 PRINT "YES, I LIKE YOU."
110 PRINT "DO YOU LIKE ME TOO?"
120 INPUT A$
130 IF A$ = "YES" THEN GOTO 200
140 PRINT "OH. THAT MAKES ME SAD."
150 GOTO 10
200 PRINT "GOOD. WE ARE FRIENDS."
210 GOTO 10

I felt a flash of this same sense of wonder, a very precise memory image across the decades. Type in the kind of message you’d ask a human, a very human-like – eerily human-like – response appeared on the screen, line by line by line, complete with a helpful follow-up suggestion at the end of the answer to keep the conversation going. Follow up for clarification, or just the thrill of having an interaction with what appeared to be an actual intelligence, and another response appeared to be an actual intelligence, and another response appeared, ordering another helpful suggestion to keep the conversation going further. It seemed miraculous, like soon we’d be living in a world of actual CPOs and R2D2s (or, as some feared, HAL), our lives and abilities extended and enhanced by our wonderful thinking assistants.

But the LLMs don’t ‘think’. The LLMs and the Deep Learning technology on which they are based have some serious limitations, some of which were recognized as far back as 2019 (what was that Marcus who pointed out the limitations of Deep Learning). The LLMs, in the end, are just a vastly more sophisticated version of those either/or programs we wrote for kicks in BASIC. But its good to remember that initial sense of wonder Chat GPT inspired, how it felt, for maybe a year or so, that we were embarked on an entirely new adventure.

Ultimately, learning to use the LLMs is like using to play an instrument – without proper human input, it produces wrong notes, dissonance – nonsense.

Back To Blog

How I use LLMs

by timbeckett@ymail.com

The LLMs, after a three year reign, are possibly on the way out. Or at least – there is now significant pushback and skepticism about their efficacy, their efficiency, their incredible energy use. They have been rightfully condemned for their tendency to hallucinate (ie make things up when they don’t have the correct answer), to sycophancy (ie explicitly trained to tell their users whatever they want to hear, reflecting not just their users’ tastes and opinions – but encouraging their users own WORST tastes and opinions. Like social media, they are designed to be addictive, pestering their users at the end of every answer, hallucinated or not, with further suggestions and queries, with the familiar, seemingly helpful tone of the eager assistant, designed to keep the user keeping the thread open as long as possible.

But the LLMs DO have their uses – and I’ve been using Chat GPT, still my default, almost since it appeared in December, 2023. I signed up for a Pro account as soon as it came out in January 2024. I’ve tried to write with it, I used it for study, as a replacement search engine. Seduced by a piece in the New Yorker ‘A Coder Considers the Waning Days of The Craft‘, I pondered giving up coding, or at least giving up coding without the integration of the LLMs, and build apps in languages and frameworks I either didn’t know well, or had forgotten, relying heavily on LLMs to fill in gaps in my knowledge. When, stymied by a collapsing NYC tech job market, I took up freelancing, I used LLMs as an assistant to deal with page builders, optimizations, SEO, integrations – subjects I sort of knew, or had known from last iteration as a freelancer 10 years before. I used the LLMs for job search, CV optimization, I used it to research and write copy, find and configure plugins and, when I returned to using Laravel and Django after a few years absence, I used the LLMs in an attempt to get back up to speed. I experimented with LLMs to write, or analyze books, find historical and Biblical and mythical analogies for stories I was writing or planning to write. I briefly experimented with generating images, even video.

Success? Mixed. Very mixed. I should have been more attentive and kept an ongoing log of what worked, what didn’t. I should have written blog posts, documenting my finds. But I didn’t. I was overwhelmed by the firehose of information about AI, and all the wonderful things it could do, all the jobs it would replace. I used it day-to-day on a sort of intuitive, almost unconscious level and attempted to control its voliminous flood of information, opinion, sometimes complete BS.

But here, I’ll try to unpack how I’ve been using the LLMs, mostly Chat GPT: what I’ve found useful and what I really haven’t.

Back To Blog

What Would UBI Be Like?

by timbeckett@ymail.com

Sam Altman is fond of claiming that the advent of the LLMs means that, at some point, we’ll have to consider bringing in UBI because ‘AI will take all our jobs’. He’s even sponsored a study, done over a three year period, giving participants a grand a month.

For those not in the know, UBI stands for ‘Universal Basic Income’ and was first introduced as a concept (as we shall see, in its modern form) by Andrew Yang during his 2016 run to be the Democratic Presidential nominee. Yang’s idea was that every US resident (presumably limited to citizens and ‘resident aliens’ – ie green card holders) would receive $1200 a month as a base income, no matter what they made from their regular jobs. Such an idea had apparently been popular in parts of Silicon Valley (where Andrew Yang had been working) for some time; Yang just brought it to the public.

In other words, instead of just chidren of well-off baby boomer parents getting a trust fund – EVERYONE would get a trust fund. And why not? The US is a rich country. With some sensible, thoughtful taxation, UBI would be entirely feasible. Now with AI threatening to take everyone’s jobs, perhaps UBI is an idea whose time has come.

Thing is: I’ve lived through an early prototype of UBI, in Canada (and the UK) in the ’70s and ’80s. Only it wasn’t called UBI back then, it was called the welfare state, and came in the form of either unemployment or welfare payments. Montreal in particular, when a significant percentage of the population lived in whole or in part on UI (from the feds) or the Bien-Aide Sociale (BS, from the provincial government), supplemented by odd jobs, artistic or musical pursuits, or spells of grueling, physically intense work in the hinterlands. For many, it was more or less a way of life.

In those good old days, Canada’s second largest city, was recovering from the tumult of the ’60s and ’70s, when the FLQ had been setting off bombs and kidnapping government ministers, the election of the separatist Parti Quebecois and the subsequent referendum on Quebec’s sort-of-separation from the ROC (rest of Canada), along with the massive outflow of its Anglo population – and their money. This, at a time, when not only the opening of the St. Lawrence Seaway had reduced Montreal as a port, the ’76 Olympics which had left the city with ruinous debt, and the general industrial decline that bedevilled cities across the North-East. Montreal was not in great shape.

Into this void came the welfare state, attracting a generation of artists, slackers, and riff-raff from the ROC and the Montreal suburbs, drawn to a city where a two or even three bedroom apartment in a picturesque neighborhood with easy access to great food, cheap bars and restaurants, could be had for a couple of hundreds bucks a month. No deposit, no first and last: in most situations, you paid month by month. Basic BS was a mere $180/month, not enough to live on, but if you could scam medical BS, you could get up to $500/month. If you went out into the bush for the Great Canadian Migration that was tree-planting, and put in 8-12 weeks of back-breaking labour in all kinds of weather fighting off hordes of biting insects, you could come back with not just somewhere up to ten grand, but that prize of prizes, max UI weeks, which translated to over a grand a month (after taxes), paid in bi-monthly installments.

What a time.

Sounds great! Power to the People! But what did it mean in practice?

In St. Henri, a blue collar neighborhood just off downtown where I lived for a time, the factories that had been the neighborhood’s chief employer had long since closed and most of St. Henri’s residents had few options. For most of the month, the tavernes, the few diners that still existed, even the depanneurs (corner stores to those not familiar with Quebec) were largely empty. But come the last Friday of the month, when the BS cheques came out – hey presto! St. Henri came alive! Tavernes were suddenly full from noon ’til close. People lined up in the depanneurs for 24 cases of Cinquante, Molson Ex, cartons of cigarettes! For a weekend, St. Henri was as lively and vital as it must have been in its industrial glory days.

Then come Monday, everyone went back to staying home and watching TV.

Among my circle of bohemians and artists on the other side of the city, results were mixed. Some people got their UI weeks, then left the country to go traveling, or skiing, or to spend a few months on a beach, their bi-weekly check-in card filled out and signed by a trusted friend or relative. Many planned to use their newfound free time the BS or UI allowed to do their own work. And some did. Some had the discipline to set goals, then spend an allotted time – six months, a year – to fulfill them. But for most, myself included, the lack of structure was curiously dispiriting. One day stretched into another. We spent hours in coffee shops, nursing a single coffee, only to spend hours more on the phone, talking about more or less the same things we’d talked about in the coffee shop. Come evening, we’d gather in some cheap bar for the even cheaper happy hour, often with the same people we’d seen or talked to that day.

It kind of got stale. Fast.

That said, I wasn’t against it then, and I’m not against UBI now. It can’t just be the children of the rich – or the plain rich – who get to do art or work in media or launch startups. Or hang around and do nothing. Regular folks should be able to do all these things too. But if AI really does take all our jobs and we ended up with masses of leisure time, we’re going to be need to be trained on how to use that time. We’ll need to be trained as tinkerers, inventors, founders – or even just volunteers. We’ll need to be trained in how to be useful.

But this is only if AI takes all our jobs and the US government has the vision, the courage, to bring in some as profound as UBI. Since the US has been unable to even bring in a universal health care, this seems very unlikely, whatever Sam Altman says.

Back To Blog

Learn (Basic) Natural Language Processing (NLP) in 6 easy steps

by timbeckett@ymail.com

Recently, I started the ‘Intro To NLP’ Course on the 365 Data Science platform. I’ve noticed the platform scrambling in recent months to introduce AI content after focusing almost exclusively on Data Science topics. Since the two are related, this wasn’t a bad thing, but glad to see it bringing in more high quality courses focusing exclusively on AI/LLM topics, and looking forward to seeing more.

The Intro to NLP consists of 7 sections of lessons with a Practical Task at the end of each section, then a section devoted to a whole project, finishing up with a ‘Future of NLP’ section before the Final Exam. So far it’s been a good course, about a subject I knew almost nothing about before I started, but which is central to LLM like Chat GPT and AI generally.

What is Natural Language Processing, or NLP? NLP is a technology that allows computers to understand, interpret and respond to human languages, both written and spoken. As it turns out, it has an astonishing number of uses, and was widely used in ways I never even thought about, even before the arrival of the LLMs:

  • Search engines: when we type a query into a search engine, NLP algorithms interpret our intent, understand the context, and deliver relevant search results. This involves processing the query, identifying key terms, and even understanding the context of our question.
  • Voice Assistants and Smart Home Devices: Devices like Amazon’s Alexa, Google Assistant, and Apple’s Siri use NLP to understand spoken commands. They can interpret our requests, respond to queries, control smart home devices, and even engage in casual conversation. Sort of.
  • Text Autocorrect and Predictive Text: The autocorrect and predictive text features on smartphones and other devices use NLP to understand the context of what we are typing, correct spelling errors, and predict the next word we might type. Sometimes this works, sometimes it doesn’t. I’ve sent many garbled text because of Autocorrect.
  • Language Translation Services: Online translation tools like Google Translate utilize NLP to convert text or spoken words from one language to another, understanding grammatical nuances and context to provide accurate translations.
  • Chatbots and Customer Service: Many websites and customer service platforms employ chatbots that use NLP to understand and respond to customer inquiries. These bots can handle a range of tasks from answering FAQs to helping with online shopping or troubleshooting. Next time you’re frustrated by some generic response from a chatbot and wish you were talking to an actual human – thank NLP.
  • Email Filtering: Email services use NLP to filter out spam or categorize emails into different folders (like social, promotions, primary). This is done by analyzing the content of the emails and identifying certain patterns or keywords. This has gotten dramatically better in recent years.
  • Social Media Feeds: NLP algorithms help in personalizing our social media feeds. They analyze our interactions, the content we engage with, and use this data to curate a feed that is supposedly tailored to our interests. I would say they mostly ruin them, but you get the idea.
  • Sentiment Analysis: Businesses use NLP for sentiment analysis to gauge public opinion about their products or services. By analyzing social media posts, reviews, and comments, they can understand customer satisfaction and general sentiment.
  • Content Recommendations: Streaming services like Netflix or Spotify use NLP to recommend movies, shows, or music based on our previous viewing or listening habits, search history, and preferences. I’ve found these to be mostly . . . if not quite useless, close to it.
  • Accessibility Tools: NLP aids in creating tools for individuals with disabilities, such as text-to-speech and speech-to-text applications, which allow users with visual or hearing impairments to interact with technology more effectively. Potentially very useful.
  • Educational Tools: NLP is used in educational software to aid in language learning, provide automated grading of essays, and even give feedback on writing style and grammar.
  • Resume Screening: In the hiring process, NLP is used to screen resumes and applications to identify the most suitable candidates by matching job requirements with the skills and experiences listed in the resumes. Another instance where we’d probably be better off without this. Ever wonder why you can’t get an interview? Thank some unknowable NLP algorithm.

But with the arrival of Chat GPT and the other LLMs a year ago, NLP really came into its own. Suddenly we were able to engage with a machine in a way eerily similar to how we interact with human beings. We can argue whether this, long term, a good thing, but seen purely as a technology, Chat GPT is amazing. Even with all its flaws, hallucinations, errors, the dubious practice of just scraping content without any permissions at all, and so on. Chat GPT and the other LLMs struggling to catch up, really are a leap ahead on a technological scale and NLP made them possible.

How is NLP used in the new Large Language Models?

  • Understanding Language: ChatGPT uses NLP to grasp the nuances of human language. When you type a sentence, NLP helps the model understand not just the words, but also the meaning and context behind them. This understanding is crucial for generating relevant and coherent responses.
  • Generating Text: Once ChatGPT understands our input, it uses its knowledge gained from NLP to construct a reply. NLP guides it in forming sentences that are not only grammatically correct but also contextually appropriate, maintaining a flow that resembles natural human conversation.
  • Learning from Large Datasets: ChatGPT has been trained on a vast array of text data. NLP is used to process and learn from this data, enabling the model to recognize patterns, understand various topics, and even mimic different writing styles.
  • Handling Different Tasks: Whether it’s answering questions, writing essays, or even creating ‘poetry’, ChatGPT uses NLP to tailor its responses to the specific task at hand. NLP provides the flexibility to switch between different types of language use, from formal to casual, technical to creative.
  • Continuous Learning: As ChatGPT interacts with users, it continually refines its understanding and use of language. NLP is key in this learning process, helping the model to adapt and improve over time based on new interactions and data.

In short, NLP is the ‘brain’ behind our LLM’s ability to communicate effectively with humans, using human language. It is what allows the model to understand our questions and respond in a way that is informative, engaging and eerily human-like.

So now we’re going to learn the basics of NLP and how to use it. First stop: Text Preparation.

Back To Blog

The Draughtsman Writer

by timbeckett@ymail.com

A couple of years ago I saw The Draughtsman Writer at a larger exhibit at the Met, Technology In the Age of the Court Most of the exhibits were clever, if sometimes dazzling: many immensely complicated clockworks, constructed with gold and other precious minerals, but nothing truly blew me away until the exhibits at the show’s end, and in particular ‘The Draughtsman Writer’.

Along with our Draughtsman was his more famous cousin The Chess Player (sometimes called ‘The Turk’ because of his garb), a replication of the original automaton by Wolfgang Von Kepeler in 1769, which ‘played’ chess with prominent figures across Europe, and which was eventually revealed as a fraud, manned by skilled, even famous chess players hidden from view yet operating the Chess Player’s mechanical arms (I’m not entirely clear how they did this).

The Draughtsman Writer, however, needs no human intervention, except perhaps for a key to be wound up so that the Draughtman’s profoundly intricate gears, hidden in the Draughtsman’s desk, can begin turning, then a human hand to place the paper that with Draughtsman will fill with his exquisitely intricate drawings and poems. From the exhibit notes:

Maillard hid the mechanics of the Draughtsman Writer in a cabinet rather than the figure. This allowed for larger machinery and greater memory than in earlier efforts . . . an unprecedented three poems and four drawings are drawn by the figure, through a technology that foretold the computer.

Incredibly, when Pittsburgh’s Franklin Institute received the automaton in 1928, it was so damaged by the fire in the warehouse where it had been stored, they had no idea it was an automaton. They knew it had some mechanical function but, since it was in pieces, they had no idea what that function was. They didn’t even known the name of the inventor.

I was instantly captivated by the Draughtsman Writer. In part it was the instance of an early robot. No uncanny valley here – the automaton is only half-formed (many of its panel, its ‘skin’ possibly lost in the fire), with a young man’s dummy head, yet none of the creepiness we associate with a ventriloquist’s dummy. This is a benign, contemplative figure, eyes focused downward on its task, its transparently mechanical arm composed of brass strips, an almost human hand holding a pen, tracing delicate lines across the page fitted into the Draughtsman’s desk. As one of the curators says:

Normally we think of robots moving in very mechanical ways, very jerky movements. This machine is by far the most elegant in its movements.

The Draughtsman can compose four different pictures, including drawings of a Chinese temple and a ship, and write three poems, one in English, two in French. The ‘hard drive’ for these movements are the brass disks housed below the surface of the Draughtsman’s desk. The disks have hills and valleys on their surfaces, and a needle follows these grooves up and down, the collection of disks allowing for the most extensive mechanical memory of any known automaton.

The moving automaton was confined to a video next to the exhibit (as it was with all the other exhibits, presumably too old, too delicate for the repeat performances the exhibit would demand). Instead, in the actual exhibit, the Draughtsman peers straight ahead, eyes wide open, pen poised in its hand, waiting for the human intervention to fulfill its function and begin drawing and writing again.

The Draughtsman Writer at the Metropolitan Museum
The Draughtsman Writer at the Metropolitan Museum

I was struck, watching the video again, by the beauty of the calligraphy, the detail in each drawing. What a watchmaker Maillard must have been, to so precisely record each groove in his brass discs, long before the plastic record album would perform the same function. What might have Maillard done with a computer, with modern computer languages?

What also struck me is the essential frivolity of the Draughtsman Writer. Maillard likely built it to impress the court, as a sort of calling card. But essentially our Draughtsman exists to produce art, no more, no less. Sometimes I wonder if our machines shouldn’t be, at least partially, repurposed to do the same – not just for convenience, ‘communcation’, ‘disruption’ but to produce beauty, wonder. When our machines, our AI, can acheive some of the pure wonder of the Draughtsman Writer, will we able to say the Digital/ AI revolution has matured, been absorbed into our human fabric (instead of threatening to run amok as it is now?