AI Divergence Engine - the Creative Process in Making Tech-Art Sculptures

The Divergence Engine: A Tech-Art Adventure, Hunting For More Creative & Ethical Use of AI.

This is a project that explores how AI can be used in a more original and creative way: Currently AI tools like ChatBots and LLMs tend to be somewhat ‘black boxes’, and you don’t really get to ‘adjust’ the parameters, if you’re not a data scientist or techy person.

We believe this process of generating a query for AI should have some ‘knobs, levers, sliders’ - and other ‘adjustables’ to remind general users to take better control of their experience. The Divergence Engine highlights the issues around the provenance of data sources being potentially quite biased, and not as globally expansive as one might like.

In short, by giving you agency over more parameters of the internet it makes you aware of how to make it more provocative, ethical, and hopefully that in turn, expands your creative horizons. This is about creative diversity, as a means to making your work richer, but also because it's more exciting to engage with more of the world than the 'default' or 'most popular' sources.

A Guide To Creating Sculptural Tech-Art.

The Divergence Engine, (See Build-Guide Instructable) created by creative technologist Jude Pullen, and artist Max Park, is an open source project you can build, but this Instructable is written to share how we think Tech-Art (and Sculpture) is a particularly suitable means to explore new frontiers in technology, ethics, creativity - and we feel this thinking is as important as what we built. Jude and Max hope to improve upon through discussions, debates, and challenging ourselves to collaborate with new people.

The Instructables community seems an ideal community to engage with both creatively, technically and ethically. We hope you like it! Thanks for your feedback, challenges, and builds upon this project.

CONTENTS:

Artistic Exploration (the process of finding out what we wanted to say)

Hardware Feasibility (checking if we can manifest ideas into the tech and physical forms)

Embodying Ideas Into Art (deciding what messages the artwork embodies)

Politics of Art & Design (deciding the hierarchy of what messages need to be communicated through the work)

Sculpting Ideas Into Art (insights, reflections, mistakes, changes, as we developed the physical artwork)

Revisions & Refinement (taking feedback, finding faults to correct, discussions and more... [ongoing])

Evolution & Collaboration (the future efforts from this "V1.0" iteration)

Why Tech Needs Art and Art Needs Tech.

For many years, creatives at pioneering award-winning studio, Pixar operated influenced in part by a mantra of "The art challenges the Technology, and the Technology inspires the Art", which is why if you've ever seen a Pixar film at the movies, you'll notice that there is a 'short' movie before the main feature.

The 'shorts' were not just a means for junior staff to get some real world exposure, (although it does enable emerging/rebel talent to be recognised early), but it has generally be used as a way to test a new technology idea, such as way to render fur (leading to Monster's Inc), water (Piper), crowd dynamics (Blue Umbrella), or to see if narrative can work without narration or dialogue (leading to Wall-e). Pixar's animator, computer scientist, and co-founder, Ed Catmull captured many of these insights in his book Creativity Inc. - and it is a great overview if you're into exploring how Art can influence Tech, and vice versa, especially in a a commercial creative enterprise like Pixar, which is not just artistically world-leading, but commercially lucrative - a fine balance, indeed!

I have been somewhat 'tongue-in-cheek' to add this in 'Supplies' here, on Instructables, as this sort of thinking are all the 'tools' or 'gear' you need to get started in this way or working. However, I'm also keen to perhaps start from an important point - Max and I had both been exploring ideas, (roughly for similar times in AI), before we worked together, and that was incredibly helpful.

LLMs Providing A Link To The Past: Prompting Nowhere.

Max and I first met New Designers, at Central St. Martins, London. He was exhibiting his final-year degree project - a tech-art object which looked like an old sewing machine with some intriguing and mysterious tech attached. It turns out he had created a 'ghost in the machine' -type of experience, where he'd created an interactive experience where you could 'converse' with an LLM (ChatBot) with the AI-personality of activist-artist William Morris. The source material was his highly influential book, News From Nowhere, which embodied many of his ideas around craft, technology, equality, and society. Although the book was written in 1890, it still feels relevant today in many aspects of musings on work vs craft vs liberty. I was incredibly impressed by the fact Max has just 'figured out' how to do so much with this relatively recent tech (in 2025), and has also made a nice statement in that the machine required you to 'treadle' the sewing machine, in order to power a generator, to power the Raspberry Pi computer, running the AI agent or LLM. A tour de force of tech skills, and artistic thinking! We of course hung out and the rest is history.

Playful Physical Interfaces: RadioGlobe.

Since I went freelance after working as a Tech Scout for LEGO's Creative Play Lab, I created RadioGlobe, an IoT device which makes 15,000+ web radio stations accessible via an engaging physical interface, or a globe with a pointer on it. It also uses a Raspberry Pi computer, as well as some state-of-the-art military rotary encoders, to 'segment' the world into 1,048,576 coordinates to locate most major cities in the world. It was created for RS Group, to celebrate their 10th anniversary of the engineering website, DesignSpark, and went on to gain awards and accolades from Elektra, Core77, The Drum, and FastCo. It was designed in collaboration with Don Robson, and Pete Milne. The project has also been made by many people on Instructables, with new variations. Although the RadioGlobe is a fun gizmo, its backstory of being created in 2020 in the midst of a global pandemic, was certainly influenced by a feeling of isolation, division, and fear - and so it is a counterpoint, and celebration of feeling that one can remain connected to other people and cultures - through music, news, and talk radio.

Creative Collaboration.

We both connected over the way our respective projects were playful, yet tackled some deeper more thoughtful questions underneath the main idea. We kept in touch for a number of months before wondering how we can work together on something which made use of both our creative making skills, interest in AI and tech, as well as being provocative in what we do - to challenge others to explore broader creative futures.

Art As A Trojan Horse for Tech Evolution.

At the time of writing in 2026, it's a fair understatement to say 'AI is contentious' as a subject. We have egregiously unethical problems around Data Centres, AI safeguarding, and the problematic business model of many Social media companies. It would be foolhardy to pretend that any of the above would be considered 'revolutionary' acts, and so it is more likely that ideas in art can only hope to create change in more modest ways, but which in large numbers add up. Another approach is to use an artistic project to begin a difficult discussion, by coming at it tangentially, or in a 'Trojan Horse' fashion.

I'd been working with NVIDIA and RS for the previous 2-3 years around the question of what AI can do which is ethically progressive (if not perfect). Reporting on companies like Lion Vision who used AI machine learning vision to remove LiPo batteries from eWaste (to prevent chemical fires in recycling plants), was a laudable use of AI. I wrote a blog series before that about how repair legislation will change how products are designed. This informed a project to use AI to reduce packaging waste for RS, by using NVIDIA's AI hardware to optimise packaging. This latter project used the 'Trojan Horse' method to develop something which encourages sustainable design, whilst still being about reducing waste, which is net-profitable for the company.

"Tell All The Truth But Tell It Slant" - Emily Dickinson.

My point with RS and NVIDIA is that sometimes what begins as a seemingly artistic or ethically motivated project can build over time to be something which affects the economics of how we do things, but still with those roots in aspirational places, which may have been overlooked if the project started only from a profit-centric mindset.

We hope that the AI Divergence Engine, for all its apparent whimsy and fun coloured buttons and knobs - is actually making some very serious points about unethical use of AI / LLMs / Agents / Datasets / Coded Bias / etc.

Emily Dickinson wrote "Tell All The Truth But Tell It Slant" a poem about how often the truth can 'dazzle' or even 'blind' those not wanting to see/hear it, and so a more circuitous route is necessary (eg Trojan Horse, being one approach of many) to gently persuade others of an alternative perspective.

Asking Big Tech to change their business models is not trivial. Hoping Governments will 'sort out this mess' is also famously slow. However, neither is it a given that people will choose activism / abstinence over convenience, when tech is very skilled at making things desirable. As designer Victor Papanek observed “There are professions more harmful than Industrial Design, but only a very few of them" - implying that *desire* is at the root of consumerism, and conspicuous consumption, and is responsible for many issues of the modern world.

Preaching or blaming people is rarely a good strategy for change (unless fuelling a Revolution). Pointing the finger at Big Tech / Government / Society simply ends with no side feeling they are responsible for all the blame, and in an objective sense this is true. However, this is perhaps why Art is a useful 'tools' in response to Dickinson's point - it allows us to tell the truth, [about Tech], but in a 'slant' way... one can look at the Divergence Engine not as a criticism of one group (assuming you work in tech, are a consumer, or perhaps work in policy, or a combination of these), but instead as a plea to see the uncomfortable and inconvenient truth that much of what we encounter online is inherently bias, has colonial overtones, and in short, is not ethical.

With that said, it is also not intended as a polemic to say all AI is bad, as of course many life-improving things are occurring in the 'Robot Age', as described by Marty Neumeier in his book, MetaSkills - but we should remain vigilant about retaining our agency over how we use the internet. An act as simple as 'turning a dial' [Soft Power Index dial] reminds us that we can either be complacent as use the internet 'as default', with it's biases, or we can *shift* our perspective to engage with new sources from the global majority beyond just USA, UK, EU, for a broader perspective.

The Business Case For Divergence.

Art has generally always been about provocation. However, even the most hard-nosed businesses would agree that becoming entrenched in one ideology might work in the short term, but is risk in the long term: AntiFragile: Things That Gain From Disorder and The Black Swan: The Impact of the Highly Improbable by Nassim Nicholas Taleb are notable books on the subject. It is worth noting that Taleb is from an economic background, so if he is pointing out the importance of diversity of teams, experiences and creative thinking to help businesses remain profitable and dynamic, this is not someone a CEO might dismiss as a 'tree-hugger', to put it bluntly!

People Hate Bad News. So Try To Be Positive When Re-Defining The Future.

In summary, this is why we think that Art is vitally important in the Tech sector right now, as it exposes problems in a way that is aspirational, rather than confrontational. The Divergence Engine is not trying to preach, shame, moan or 'doom-stoke'.

It is trying to say: 'Wouldn't you like your creative sources to be more exciting and hence perceived as being more original?'

It is asking gently 'Do you ever find the typical AI outputs dull, middling, and average - too safe perhaps? Well, perhaps you're ready for some 'Black Swan' or 'Anti-Fragile' thinking, and that comes from looking beyond your usual places online?'

It embodies the sentiment of 'If you value going on holiday to new countries and exploring new cultures, this machine is trying to do the same for your online AI searches - sounds like fun?!'.

The commercial pledge is 'If you value original pioneering ideas, often espoused by Steve Jobs [and whose who quote him], then the blunt truth is, they got them by stepping outside their 'bubble', if you use AI in a lazy way, you're just drinking from all the same wells as everyone else... so go beyond, if you want to 'think different'!'

Why Use AI in a Creative Process?

In early 2024, I felt a restless sense of how GPT had been 'having a moment' in 2023, but it occurred to me that it was very much a powerful tool for exploring 'linear' thinking, which was reasonably grounded in logic. However, it was also very prone to hallucinations, and mistakes.

When you look back at how LLMs actually evolved as a means of translating Chinese <-> English, and not really designed to be 'Artificial General Intelligence' - to put it simply - they are 'language prediction models'. They are good at guessing what comes next. Writing this in 2026, this is fairly well understood in the tech community, but I still feel the general public does not have a strong affinity for why AI is good for X, and less good for Y, and terrible for Z. And I still feel even then, it's a moving feast, which changes on a weekly basis.

Anyway, I created this slide to explain what I'd been ruminating on, so I could throw some ideas around with Brian Schwab, who I'd worked with on a recent AI project with NVIDIA and RS around packaging (link). We ended up working together on that paid project (and still are), and this somewhat fell into 'passion project', as I was not entirely sure I could find a client to fund such an esoteric project.

I mention this to be somewhat candid about the reception of 'artsy stuff' vs 'serious stuff'. Perhaps I'm biased from working at LEGO as a Tech Scout, but I honestly believe most really weird new breakthroughs in tech were predated by either nerds working in labs, or by artists playing around. It's easier to say something has been created by 'serious researchers' at MIT rather than some quirky art graduates - even if I personally think BOTH are as likely to play significant roles in discovering impactful steps. Scientists are tried to look for the tangible, concrete, definitive. Artists are more comfortable with ambiguity, uncertainty, and the subjective. AI and AGI seem to be things which will need both scientific rigour, and emotional 'feeling' our way through, to fully realise.

Indeed, there is an irony that the people we tend to idolise, be it Leonardo Da Vinci or Steve Jobs - were often both pretty agnostic to such boundaries. Yet both 'code switched' depending on what they were trying to do, which is not as disingenuous as it sounds, but rather to recognise having a room full of polymaths as your audience is a rare, to say the least.

Meat Grinding Other Ideas

The sketch / schematic I drew is of a device (hand drawn) that would parameterise an experience such that it'd produce a 'creative unblocking' prompt for you. Inspiration was taken from Brian Eno's Cards, as mentioned, but also Brendan Dawes 'States of Mind', as well as a compendium of creative tools, called Thinkertoys by Michael Michalko. Matt Webb's Poem1 was an early Kickstarter project, exploring GenAi in an approachable form. I also was somewhat obsessed with Pulse Oximeters as an ever-increasing symbol of 'data harvesting' in our lives.

These and many more peripheral influences went into the 'meat grinder', as if to symbolise 'grinding all this inspiration up' into the final [sausage] output, which looked initially to me, like a gambling machine, a 'one armed bandit' form actor to imply that this will be random, and require some luck.

I later showed this to Max, and he pointed out that the Meat Grinder was a fun form factor to consider, and so we decided to 'run with it', as I felt the original idea may have been too 'straight' - and I could stand a little more remarkability in the form to make people curious. The 'grind' and 'wait' aspect was also likely a good foil for having to wait for the AI to process all the data, so it may be no bad thing not to expect it to output in less than 5 seconds.

Original Ideas & AI Data Scraping

Much has been written about the unethical scraping and stealing of artists' work by big AI companies. Both Max and I were fundamentally against this cynical approach. With that said, long before AI was around, Humans have been 'stealing', borrowing, remixing, reappropriating, and recontextualising ideas. I wanted to acknowledge here that we were of course influenced by others in our creating The Divergence Engine.

However, where our idea of course differs, is that we have not aggressively scraped or stolen or passed work off as our own. Indeed, part of the point of my citing influences is that I don't believe this diminishes our work to say we were inspired by others who came before us.

Granted, this essay will not 'resolve' the politics and economics of Big Tech here, but I hope that the guiding principles of the Divergence Machine is that we might discover more people's work and ideas, but also in doing so give due credit. This is why 'receipts' is a nice metaphor, as 'showing receipts' is often used as slang to say 'prove your sources and reasoning'. So thanks to those who inspired us, and we hope you check out their work!

I'll get into this later, but we are trying to use as little 'Gen AI' as possible, even the images and links are 'retrieved' from Open Source (free to use) commons, and images are 'cited' not generated. Both have links and credit given.

Our provocation of Big AI Tech is they are reluctant to give citations in Image Generation especially, but even in text, the sources are especially limited. It is this latter point that the Divergence Engine somewhat ‘throws down the gauntlet’ to ask ‘why not more diverse sources?’. We may have to revisit the issue with GenAi in a different project, but we may also need a lawyer on the team too! (Please get in touch!).

As mentioned, I met Max at his degree show, and it took many months before we got to exploring what we'd collaborate together on. After being excited by the ideas we were discussing, it made sense to also get these down on paper. These 15 sides of the note book are the first 'minutes' of our meeting. Looking back on the project it's kinda fascinating to see how much of the underpinning ideas managed to be realised, but also how much we still can explore in more detail. Having worked in Chemistry on 'Design of Experiments' you realise that most of the time our brains can visualise only a couple of parameters at once. We tend to imagine X vs Y graphs at school. Occasionally you get a X vs Y vs Z graph, which is 3 dimensional. In chemical DoE, you might have hundreds of 'axes' or parameters being explored at once, and of course that 'breaks my mind' trying to visualise it. There is something about that with the Divergence Engine, which is why the Dials perhaps 'keep it relatable' by giving us frames of reference we can 'hold onto', whilst the AI does its thing.

That feeling of 'the more you learn, the more you realise you don't know' is a real thing! Indeed, many discussions have left me triggered to go read up more on History, Psychology, Neurology, Computer Science, and more. Which is a way is the state of mind the Divergence Engine is trying to promote: Curiosity.

When not in the same room, we use Google Slides as a way to drop in ideas for discussion. The ability to work simultaneously, and asynchronously on a document is something having grown up 'pre-digital', that still blows my mind, in the way it enables ideas to not just be recorded and shared, but the simple activity or critiquing and building upon each others ideas, when the moment takes you (ie without formal meetings) is an interesting mode.

To be clear, I think both 'formal' and 'informal', as well as 'in person' and 'remote' - are all needed, but the digital side is perhaps a precursor of how we might adapt to AI, GenAI, and AGI - it will just become normalised into our workflows, ignoring the bits we don't like, and keeping the bits that make sense.

The fact that very few people stop to consider how Google Docs have revolutionised how we can collaborate on the same document or idea at the same time, is astounding. I know it is astounding, as in 2009 I was working on CAD models using UniGraphics at Dyson, and that ability to collaboratively work in CAD 'live' was considered 'game changing' and no doubt valued at millions of pounds! And yet in 2026, it's normalised, and virtually 'free'...

...Sounds like AI futures to me ;o)

Anyway, here are some of the musings, minus the comments section, but Max and I both worked on them over the coming weeks. Not just the ideas, but also the pragmatic stuff of sourcing parts also.

This first image is 'where we got to as first draft on components'...

...If you want the whole back-story, that's the subsequent 30+ images of some of the discussions between Max and I. Hopefully useful to students or perhaps prospective clients, to show that you have to 'sweat the small stuff' and figuring out your USB connectors can be a surprisingly deep rabbit hole!

Essentially, the point I think this shows is that 'the devil's in the detail' is real! Small considerations like how the receipt is printed, and the amount of compactness of the printer, has a big impact on the size of the [Divergence Engine]. One thing you do become aware of when hunting around of random parts at the start is that you can run the risk of things going out of stock on Amazon / eBay / AliExpress, so we ended up 'transferring' to RS and PiHut for more reliability in supply, but also in the components themselves. Sometimes buying from a specialist supplier is worth it in the long run!

As with the Soft Power Index, the use of 'Meat Grinder' was a useful 'good enough' term, that allowed us to make the point that just as a sausage 'hides all manner of sins' of the Butcher, making it hard to tell the actual ingredients of the sausage, we also felt that the 'good' data was mixed in with the 'bad' data - or rather that most AI companies were not operating in a very 'open' way (sadly OpenAI's 'openness' has been redacted).

We did actually buy a meat grinder, and some playdoh, and it has become useful to see how certain dominant colours/amounts can make smaller 'inputs' feel overwhelmed to the point of insignificance. In many ways we observed this happening with the grinder, but can also see how this 'most dominant thing wins' rule of many LLMs is why so many AI interactions can feel 'middling' - as they are trying to give a 'sure thing' which is not about being creative, tangential, unconventional, but nor is it also about being diverse, aware of minority opinions, and the AI does not 'audit' its own biases as a default.

Without wishing to oversimplify things, we need to shift the narrative from 'AI is bad' to 'AI companies need to be more accountable', and should operate with the same ethics that a good newspaper, company, or indeed, Butcher adheres to - or they face penalties which are meaningful deterrents.

As the time of writing, many AI companies have lawsuits against them, which even if they lose, the cost is so trivial compared to their colossal revenues, it's simply 'the cost of doing business' and is clearly no deterrent from them correcting issues, or even being lawful, let alone just.

Whilst wishing I had shares in 3M, who make Post-its, Max and I got to work laying out our thoughts on various parameters and how these should be sequenced.

It turns out that it matters quite a lot what order you'd operate on a query... for example, if we were to select a country first, and then select a topic, and then apply the social sensibility, this delivers a different result than if you do that in reverse.

Without getting too much into things, it also changes the compute time, as of course you want to narrow in ways that help efficiency, but don't undermine the point we're trying to make by being 'divergent'.

This was why we have the Soft Power Index first, followed by Overton Window, as it makes sense to do this, rather than the other way around, as this way we're focussing on a country before applying social norms, rather than the entire world's norms, and then picking a country.

It sounds obvious in hindsight, but these and other parameters took some time to work out, and of course the shuffling of stick-notes helps.

The post-it notes were invaluable to discuss the 'high level' thinking, but at some point this needed to transform into Code that could be used by LLMs. Max took the lead on this, as worked this into a sequence of commands, which allowed us to be efficient with the processing, whilst also not using too much GenAi when 'retrieval' would suffice.

Here you can see the basic structure, before breaking it down. Pleasingly, this was a GenAI image, but which was created by parsing the code, and some descriptions of the process. Max was quite impressed at how well it has interpreted this, so we've used it in keynotes so far!

There are various maxims about how "it's pointless 'asking' a Goldfish about Water - because they are in it, and it's all they've already known" [unless you take them out of it of course].

Speaking as a western, male, who is mixed race, but presents as white, from a middle class family, one can appreciate that much of what I consider 'normal' is a 'default' in Soft Power terms... I consider BBC News my 'go to' and rarely look beyond it, if I'm honest. Same for my choice of newspaper websites like The Guardian. Feel free to mock the stereotype, haha, but we all do some version of this. There is a comfort in the familiar, but it can be limiting over a lifetime.

The Soft Power Index is something devised by Brand Finance, to essentially rank countries in terms of influence (not just GDP, although there is some heavy overlap). More info here.

The additional pictures here are some of the 'works in progress' at the time of how Max and I were trying to formulate this into a workable system for the Divergence Engine, along with other factors, parameters and considerations.

My wife happens to be from Hong Kong, and that of course introduces a different perspective on a regular basis. It sharpens your awareness to see 'default' as being 'culture', not a 'universal'. It can be something as trivial as the type of milk one puts in tea, all the way to our respective takes on medicine and the ratio of preventative:cure mindset, to politics, and beyond.

A fun example of this was watching the sequel of Ne Zha, based on a well-known mythology in Chinese culture. I had never even heard of the first movie in the West. Nor did I know any of the cultural significance to 1.4+ billion people. Most of my peers have seen Lion King, numerous Pixar movies, Star Wars, Titanic, Avatar, and so on - and yet #5 in worldwide movie sales is Ne Zha 2, and it's the HIGHEST grassing animated film at the time of writing. As a Westerner, I feel this is worth some reflection on how I receive not just information, but the ranking.

The Guardian Reviews, of Ne Zha (3/5 star) by Cath Clarke and Ne Zha 2 (4/5 star) by Phil Hoad, honestly are reviews I would have previously taken at face value. Yet having spent considerable time immersed in Chinese culture, (although still a long way from 'Native'!), I'd agree with my wife that they really haven't deeply understood the cultural, artistic, and political significance, in their scant reviews, and brisk ratings.

Although this is a small isolated example, I think it nonetheless makes a point about perspective, and how the person writing a review is going to have 'blind spots', which may introduce bias, even if well intentioned, or that are not trying to be discriminatory. To watch Ne Zha without considerable priming in Chinese mythology, religion, and culture is to at best only appreciate 50% of the spectacle.

It feels like critiquing a Mark Rothko painting without the context of his persecution as a Jew in WWII. It just looks like 'bit paintings with blocks of colour, some with gradients' - surely not Fine Art!

Max and I wanted to try to do the simplest of things, perhaps crude by any standards of a scholar, which was to use the Soft Power Index as a 'lever' - by which we could quickly tweak the global order of things - to leave it as is, level it, or invert it - with a simple turn of a dial.

Lastly, it should be noted that China is #2 to USA being #1 on Soft Power Index, so this of course does not mean that Ne Zha was 'obscure' cinema, but rather it illustrates that even using this metric is subjective and imperfect. The goal of the project is to try to disrupt your usual sources of information, but as with my having to 'onboard' with various cultural contexts to really enjoy Ne Zha, it still comes down to time, effort and willingness to dive into things. As much as we may like the idea of foreign holidays, the deeper cultural aspects of places we go on holidays often remain somewhat under-explored, and quite literally a lifetime is required to truly appreciate some things. The Divergence Engine is not about trying to become 'all knowing', but rather promoting a willingness to be curious, whilst realising your own 'defaults' - like the Goldfish in Water - are not absolute.

In biology, you may be familiar with the sensory Homunculus - who has a body which is 'distorted' to be in scale with how sensitive each body part is, so lips, tongue, fingers, etc. are exceptionally large, and shoulders, back, knees, etc. are small.

It turns out, our news can have a similar 'distortion' added to it, (perhaps 'homunculisation'), and The Guardian newspaper had a similar analysis done to its new subjects, showing how not all countries have been given equal representation - link.

One might argue that because of population differences, there are simply less people, and hence less news. But this does not hold as India has the largest population in the world, and is 4.4x more than the USA, and yet, in 2026, never a day goes by without a USA story, and India is a rare occurrence!

To circle-back again to 'Soft Power', this starts to underscore by Max and I felt this was a curious metric, as it accounts not just for population, GDP, or another singular metric, but an amalgamation of many. Once again, this is not a perfect 'fix' to all the ills of data bias, but as a *provocation* it does a lot of lifting for relatively little code and investment.

[If you happen to know better metrics, please do let us know in the comments!]

You can even try this at home - I entered a simple GenAI text to image request to GPT, Gemini, MidJourney, Grok and others, to "Create an image of a young girl, playing in a bedroom, with some toys. She looks happy and engaged" - all returned images of Caucasian girls, in a western-looking home, typical standards of beauty, with an equivalent degree of income/affluence, similar toys, etc. You get the idea.

With Gemini, I repeated this numerous times, and the result still tends to be the same. There may be many complex reasons for this, but it's hard not to question this at a fundamental level.

Does this mean that when users (or companies using AI) 'forget' to add 'diversity' into a prompt for an image, it's them that are being racially discriminatory, or do we just blame AI? With an example like this, I think it's fair to say as users we should 'own' our prompts a little more.

Coming back to the Divergence Engine - although this is not exactly analogous to how we treat data, it's a good discussion point to appreciate how 'if left to its own assumptions', GenAI / AI in general, can tend towards what is 'most likely', and when one considers the historic dominance of white, middle-class kids in various household global brands, this of course skews the dataset to be bias, and prompts some introspection on how many years diversity and representation has been lacking not just in marketing, but in the upward mobility to buy many aspirational middle-class goods.

Somewhat a companion to the Divergence Engine 'mindset' would be the late Anthony Bourdain's series Parts Unknown. Chef, writer, and rebellious personality, Bourdain travels to less well explored places for the US audience. In his earlier works, A Cook's Tour, these tend to lapse into some 'gross-out' tricks of TV, with him eating stuff guaranteed to make Western audiences recoil in disgust, invoke laughter, or perhaps at worst trivialise a culture. What is perhaps a hopeful point in my drawing attention to this, is that with the later works, Bourdain has 'matured' considerably and is more interested in 'breaking bread with his fellow humans' and trying to establish an empathetic take, as best he is able to within the limited time he is there and to the best of his research team's capabilities to paint a fair picture, in 45 minutes of screen time.

Given these constraints, and in a world of much more ‘immediate’ TV shows, I think he did a pretty honourable effort. Not least by tackling some of the more contentious themes, by presenting some uncomfortable historic truths about the American involvement not just in the Vietnam War, but in what is often referred to as 'The Secret War' operating in Laos, as a neighbouring country to Vietnam.

I don't know about you, but I couldn't point to Laos on a map, and certainly didn't know that more bombs were dropped on it by America, than were dropped on the entirety of Europe in WWII. An estimated 30% of which remain unexploded today, and even with millions of pounds of donations from the US, Laos will not be UXB-free for decades to come. These sorts of shameful truths of war are compounded by chilling amounts of CIA recruiting child soldiers, and more besides. To say 'history is written by the victors' is something we learn at school, but all too easily forget.

It should be said, this documentary is written by Bourdain, and one could argue that it is still a narrative formed by his worldview, and it will contain assumptions, simplifications, and errors. We also don’t get to hear all of the other side of the story in his programme. However, efforts are made to interview local people, to gain their perspective on things, and to give voice to those who would not usually get access to media conglomerates like CNN on a regular basis. Which is to say, I think Bourdain was making a valiant effort, but ideally, one would want a documentary made by Laotians. Whether it would be as successful as receive as much acclaim or distribution as Bourdain's work is a contentious point there is no quick answer to, but nonetheless, that is yet another tension of 'who is telling what stories' is about.

This has an interesting relevance to colonialism...

My maternal grandfather was Nigerian, grew up in Lagos, later he studied at Oxford University. Met and married my grandmother, a white, English woman. They had two daughters, one was my mother. My mother later went back to Nigeria to learn more about some of her heritage, and worked for 3 years. She was often inspired by the ceramics she saw there, made by the local artisans.

Colonial people referred to them as 'Ibo' people. The preferred term and pronunciation by the locals is 'Igbo'. I'm sure some local people also used Ibo out of deference given the imbalance of power in the country, at large. Writers and political critics, such as Chinua Achebe, whose novel Things Fall Apart incorporated Igbo references and perspectives, to critique colonialism at a fundamental level.

Years later, as someone who maintained a hobby in ceramics, I taught how kids can extract clay from rivers, and make pots from scratch. Whilst browsing a local charity shop I found a book on Nigerian Ceramics, and noted how it referred to the people as 'Ibo' and was written by two European authors. I am not immediately jumping to the conclusion that the book is racist, or defending what occasionally has touches of 'primitive art' aloofness not uncommon in books written by outsiders of any cultures, who are researching the ways of another through their own eyes.

My point is not to get into a dissection of colonial impact on those cultures, there are many more qualified to do that than me, but my point is, even in a book as seemingly apolitical as one on ceramics, it has to be acknowledged that because it is written from a non-Native perspective, there will be things that are represented differently, no matter how well intentioned, or non-racist, or scholarly the efforts are. Even more so if the culture and its associated language are not easily accessible outside of its home (i.e. an English language book on Igbo ceramics is hard to ‘fact check’ when English speakers/reading subject specialists are not thick on the ground). Conversely, it is not to say that Chinua Achebe's word is absolute and that all writings about Nigeria or Africa should *only* come from those with a bloodline and significant time spent there. Indeed, he worked with the BBC, won a Rockefeller scholarship and traveled extensively. He most certainly was an authority on his culture, but his works were also informed by a diverse range of inspirations and interactions.

If we create excessively constraining 'rules' about who says what, it can become narrow and tedious also. My point is not to 'police' who ways what, but rather that *citation* in this age of GenAI is critically important. If we cannot know the author(s) used to generate an opinion, image, etc. - we cannot examine it for its origins and explore the fact that everyone has degrees of bias and alignment that need to be considered as a whole.

In the case of Anthony Bourdain, it is important to recognise the sincere efforts of his trying to make a 'horizon expanding' show that would appeal to a mass market audience, (no easy thing if you've ever tried it or have been part of it!), but that we respect the need for incorporating other voices into the mix, not just as a token effort, but as a significant proportion of the represented opinions, and to seek out the counterpoints where possible.

As regards the Divergence Engine, we cannot claim that a citation from say Chinua Achebe or Anthony Bourdain is 'the correct' voice, or an unbiased opinion - by definition an 'opinion' is individualistic. However, the need to cite it, so that one can weigh the person telling a story as much as the story itself, is perhaps where we start to understand the bigger picture.

To an LLM something as innocuous as the spelling of Igbo (say in Achebe's writings) or Ibo (say in the Ceramic book's writing) may be loaded with a deeper significance, which is not always easy for a LLM to 'footnote' with a citation, and indeed, what is an appropriate level of awareness we should expect from a reader? Is the risk that all of these details, as important as they are, expand exponentially, when you just wanted a 'simple answer'!

It's hard to say whether humans will always read every footnote and citation of the Divergence Engine, and indeed, we may not go 'deep enough' to suit every's tastes, but having some meaningful references seems a start, and having a preference to include 'primary' sources that are representative seems a meaningful thing to 'build in' to the operation of the machine's underlying code whenever possible.

Released in 2020, Joy Buolamwini's documentary, Coded Bias made a huge impact on me. The weight of evidence that not all facial structures, skintones, expressions, etc. are treated equally is impossible to deny following Joy's examination of various tech platforms and methodologies for training AI - and this was done pre-2020, so considering GPT really only 'went big' in 2022/23, a lot of questionable practices still seem to be shaping our digital lives. Ruha Benjamin's book Race After Technology also covers many of the points, being written in 2019, and both are involved in the Algorithmic Justice League.

As digital activists and computer scientists, Joy and Ruha provide evidence to what many people know or have experienced to varying degrees, that data is biassed. One of the 'red flags' for Joy was realising that facial recognition software is fairly accurate for light skinned people, but simply appalling for darker skin tones. This has a profound effect when it is used for policing and court proceedings, with mistakes leading to life changing decisions. Day to day lives are also affected - many apps and cameras have been optimised for light skin, so various tech interactions are substandard for a significant proportion of people.

The Turing Institute has also published reports on how this not only creates problems using historic data, but also if left unchecked, perpetuates and scales the issues as we create more synthetic data. In 2025, Joy spoke at Oxford University '5 Years on' from Coded Bias, link, which is worth a watch also for anyone designing AI systems or managing datasets.

The issue is not just with images, but also text: Stable Diffusion (AI) models often use text-to-image, so if words have associated bias, this can also result in problematic results in generative AI, if the association with a word is attributed to a prejudiced term. AlDahoul et al's paper (images above) makes this point abundantly clear at multiple levels of the Ai process - link.

As for the Divergence Engine, we are arguably not able to 'fix' the problem of bias data, as this operates at a foundational level on the AI training data. However, we have created a 'hack', to inject the Soft Power dial, such that the default bias can be 'levelled' and even 'inverted'. This to be clear is a very basic 'intervention', but experientially, it nonetheless provokes reflection on how an experience can feel very different - such that one becomes aware of a change, much like 'fish out of water' becomes aware. Perhaps more of us should experience a feeling of 'what about this perspective, which is different to our own'. At one level this is a political stance, but at another it is one of curiosity and openness, and another about adaptation and resilience. It seems to me at least, whatever rationale you care to align yourself with, removing data bias can only be a good thing.

To put it bluntly, overt and obvious racism is problematic, but so too is hidden systemic bias. The latter is harder to spot, harder to correct, and harder to discuss in organisations, but it has to be made familiar to those who have not experienced to build empathy towards the problem being deemed worthy of the attention it deserves.

Charlie Brooker's Black Mirror remains, for me, one of the most memorable titles of any Sci-Fi dystopian fable, as not only does it signify a mobile phone/computer, but the subjective use of 'black' to mean sinister, and 'mirror' to imply a reflection of ourselves.

However, many writers have indeed pointed out the ancient linguistic association in English (at least) of 'black' being generally a negative colour (black witch's cats, void, black-hearted), and 'white' being one of purity, etc. (Snow White, pure white paper, etc.). My point is not about 'policing' language, but rather to make the point that a LLM - Large *Language* Model - is no doubt on some levels susceptible to mis-associating words like black or white.

As a kid of mixed-race parents, I can recall using crayons and paints, and wondering 'but why are white people called "white", when their skin is slightly pinkish, and black people have lots of shades of brown, but not exactly "black" in colour'... Indeed, Crayola called one of its crayons 'Flesh' (a Caucasian tone) coloured. It took until 1962 for this to be officially updated and it was renamed Peach, but I can still recall these being in circulation in the 1980s, so perhaps that's a fitting metaphor that these things take a long time to 'purge' from circulation.

These things might seem to some as something that 'Big Tech' will fix in time, but as Reuters reported in 2025, meta wasn't just lax in getting such nuances of racial sensitivities just right, it was flat-out racist, creating foundational guidelines (presumably overseen by actual employees) which considers it acceptable to have a AI ChatBot assert that "Black people are dumber than White people."

In 2026, it's hard to say that we should simply 'act in good faith' and 'assume the best' in tech companies. Something is clearly wrong for meta to issue internal documents of such a crass nature, and that's the stuff that is written by humans, to guide AI. So how can we really blame 'computers' if the company employees are so far off the mark, and that is compounding a huge historic data set to be very unlikely to be corrected.

In regards to the Divergence Engine, we have tried to recognise some of these issues via the Soft Power Index, in that a 'country' is picked early on, and assuming not set at 'zero', there is an increased chance you'll be presented with some content which is more diverse than it otherwise would be, if left unadjusted. We cannot claim to be 'better than Google' or any such search engine, as we don't have the data to verify that claim, but we do wonder if search engines might want to consider a 'minimise systemic and colonial bias' preset, which should arguably be the default, unless you have some academic reason to want to toggle it to be discriminatory. With that said, I am aware colonial bias or even racism may or may not exist in this world, depending on one’s own beliefs, and a few public figures come to mind on this note.

One of the big issues of legislation is that it takes a long time for it to 'catch up' to where tech is, as often if we work in tech, you tend to get hired with big pay-packets by companies like that, rather than work in policy and governance for less. As part of my ongoing work with the European Commission's BEDA group looking at AI & Ethics, (G42), I hope that we might influence such things, even if it will evidently be a long road.

Image Credit: Nathan Gibbs. Flickr.

This is not a comprehensive 'reading list', but I wanted to cite the work of Ruben Pater, who wrote "The Politics of Design: A (Not So) Global Design Manual for Visual Communication", and "Caps Lock: How Capitalism Took Hold of Graphic Design, and How to Escape from It".

There are of course heavier, more academic, and indeed, more culturally comprehensive books out there - recommendations very welcome in comments - but these two books had a big influence on my readiness to not assume things don't carry bias or have a legacy which is obfuscated for numerous reasons.

They are easy to read, with great illustrative examples, and allow one to 'dip-in' to ideas, with signposting to more deep-dive works, authors, case studies.

At the risk of too many food metaphors - Max and I needed to 'age' our thoughts somewhat, and give them time to gain shape, sequence, and structure. I liked this picture Max sent me of the model I made, on his bedroom shelf... he was just 'living with it', and I think that's often what Art, Science, Tech, Design, or Engineering needs sometimes...

As mentioned earlier on, I've long been a huge fan of my friend, Brendan Dawes' States of Mind exhibition. He invites people to turn un-labelled knobs of various shapes and textures, which in turn generates a 3D rendering of an alien looking form. You stop when you feel the quirky image 'looks close to how you feel'.

Just as with the Booba and Kiki Effect, we seem to associate things with feelings and 'states of mind', regardless of them being textural or verbal. You can 'feel' a colour like 'blue', as many of us have 'got the blues' some days.

So it's with this awareness, Max and I enjoyed making the Divergence Machine a more 'hands-on' experience. To make its dials feel connected to what they were implying - a sphere to imply a globe, for Soft Power; a square/rectangle profile to imply Overton window; and Delta being represented by a triangle, similar to its greek/mathematical symbol. Likewise the Grinder Plates are screwed into place with a large flange. The Pulse Oximeter, although not actually 'sampling your blood' is positioned such that it implies that it's being 'fed' into the infernal machine, hopefully prompting a reflection on how our personal data can be used and mis-used.

Max has already used this nice hand-crank generator for his Prompting Nowhere project, and it made sense to use it again here.

The Jetson Orin Nano was also a familiar piece of hardware from the Ai project I'd done with Brian Schwab, so we were in a good place to incorporate those.

The BRAUN Pulse Oximeter was something I'd had kicking around, as a curio, as I'd been interested in these sensors for a while. It also helped to have a 'trusted' brand like Braun, to see how close the OEM ones were, and they seemed pretty close.

It should be noted that we remained fairly sceptical about using Pulse Ox devices to infer 'mood', but it did make for some extra variance to mix things up for the Divergence Engine's performance. As we were not looking to get a 'medical diagnosis', it was fine to have it as an approximate thing.

The Printer was something we originally sourced from PiHut, being a reliable source. However, we realised a very similar looking model (seemingly identical, but one can never tell) on AliExpress, with the benefit of it being in White for the white PLA prints.

Given all the test prints we did, it seems pretty reliable, with no discernible differences.

If you can wire up all the components as soon as possible. As you'll see later on, there were some 'gremlins' which surprised us, and usually are apparent when you do an assembly of *truly* everything, not just things working in isolation.

Oh, and keep doing it - with every major change!

Taking the 'core components', generator and Jetson, which had dominant sizes and significant functionality, I began with these in terms of sketch modelling in cardboard.

For the keen-eyed among you, this is a particularly 'dense' cardstock, and is not the typical double-corrugated card. From the image, this is likely 'E/B' Flute, as opposed to 'B/C' flute. In case you ever wondered, double corrugated cardboard has two slightly different 'waves' or flutings, as this keeps the waves 'out of phase', so they don't make localised weak points. Kinda geeky, but cool if you're a Mechanical Engineer, like me!

As much as I enjoy working in cardboard, some parts are so precise, that they don't make sense to try to slavishly work an imperfect medium. When fitting the USB cables, it just makes sense to have the rotated spacing of 5 parts, all with delicate geometries for screws to clamp the cables in place.

It would take me much longer to make this, for little technical benefit.

One tends to get into a cycle/flow of working where you set off a 3D print, and then work on another part / problem, and then revert when it's done, and back and forth. Most of this build was done in a matter of days, with what is nothing that complex (ie no compound curved surfaces).

As with the 'output' cone with the 3 screens, this probably took another hour or two to build, but a day to design, and was an overnight print job!

Mashing Tape is your best friend in situations like this. You need to create a series of pieces that need to be held together before glue is applied. It also saves you getting burned fingers as much as you might!

As you would with Wood, I chamfered the internal edges, so that they 'sandwich' up to each other nicely. This takes some practice. However, I did also make a handy little tool for this, some years ago, you might enjoy it! link.

The 'sliders' or linear potentiometers were screwed into place, as shown. It's one of the nice surprises that mechanical functions like this work well with nothing more than simple holes in the card. The result is a sturdy interaction that is not flimsy, and feels very close to the end result - but with the advantage of speed, and not succumbing to 'CAD Eyes' - where something looks good on screen, but feels terrible in reality.

If this were not polemic enough, this is why I think card modelling is so useful - as it hones your sense of aesthetics, ergonomics and anthropometrics much faster than anything in CAD. To be sure, as you clock-up your '10,000 hours' of experience, the gap starts to close between 'CAD Eye' mistakes, and being pretty confident in 3D Printed parts which feel/behave as expected, with few tweaks needed - but perhaps the biggest factor is speed. This took me about 1 hour, versus about 10 on my Ender 3, even if printing fast!

Design of cable runs or 'Looming' as it's known in the business, is again something which is easy to misjudge in CAD, and is very obvious when working in physical sketch models. Although later Max and I discussed sequences (see image) it was with prior knowledge that everything can fit in the build, and be assembled.

By doing this physically, you are made more aware of the rigidity / flexibility of various wires - I was told not to make a bend for a wire which has a tighter radius than 2-3 times it's diameter. So for a 4mm USB cable shown, try not to go tighter than a 8-12mm radius bend.

It also becomes apparent when to wire something, then screw things in place, etc. Sometimes if you pull one thing into place, it risks moving something else out of place. Or you have a 30mm 'slack' in the wire, but find that a comfortable dis/assembly requires you need 2-3x that much to pull it apart.

It's worth even CADing the cables, I do mine in 'red' to show the important zone that I have to shave bend-radius for.

Similar to the above point, it's easy to forget that you need to get into a tight corner to access a screw or bolt or part. You can either learn this the hard way, or you can model a basic screwdriver shape (and perhaps even a 'blob' to represent what space your fingers need to operate the screwdriver also) to see what is physically possible.

Again, going back to the point of looming - your wires may need more slack to allow for tools to access things, and assemble parts in sequence.

I really didn't have any major redesigns for the CAD, but it's because I'd treated the Cardboard prototypes with this level of 'future awareness', and made sure they were assembled with screws (not glue) to be faithful to credible assembly details.

Of course there is a balance here, and of course if designs in cardboard are more complex to make in CAD/3DP, then of course do that way instead, but often cardboard is quicker than CAD and printing for basic shapes.

I even considered a small bespoke 3DP screwdriver for accessing tight spaces. However, in the end, I simply made an 'access hatch' as it was getting a bit exotic. But we didn't forget access!

With the benefit of hindsight, it's a good idea to create test programs for hardware like screens, to see how they look...

Sadly, I'm not very good at code, so I just trusted that Max was happy with these, as he did test them using an Arduino to run some demo code, and a protoboard with jumper wires.

If it were a more time-pressured project, it would be wise to have all this in one place, so as to see how it comes together. The OLEDs for example might have been too dim, or the test too small, etc. So it's generally a good idea to have it working in a way that is plausibly close to the end result, to test UI/UX.

Although this was not necessary in the end design (as we used an Arduino, more on that later), it can help to have an 'intermediary' sub-assembly.

Consider how if we had to un-plug and re-plug all jumper cables into the GPIO pins of the Jetson every time we wanted to dis-assemble and re-assemble the prototype, one can imaging it's highly probable to get a pin/jumper in the wrong place and 'blow up' the electronics.

At £225 a go for a Jetson, this is a bit of a gamble, and so I'd recommend having a PCB that allows you have have all the peripherals organised nicely (not just as they come on the GPIO pins), and of course this also allows you to run better power management with 3.3v or 5v and GND 'rails' to be shared by various peripherals.

Try to also design your surrounding enclosure to be 'Poka Yoke' - in that you can't plug the GPIO pins back slightly off by 1 pin, and also blow up the board! Make it so that it can only be plugged in, in exactly the right way, or not at all.

Another sub assembly (later on in 3D printed parts) shown here is to ensure that the wires don't easily get snagged in assembly and get broken - it's better to create a connection point where can disconnect safely in the event something gets yanked.

It sounds obvious, but running through the sequence of things you plan to do, and see how it all feels. Does one thing lead to another? Do you assume 'left to right' sensibility? Are some things requiring 2 hands - for example to turn the crank, I realised that holding the machine was intuitive, but I made sure to keep the power switch away from where you might grab the device when cranking.

USB sticks felt too small, and benefitted from having an exterior 'hefty' and more tactile outer, and it also helped with the aesthetic in general to make it look more bespoke, and not random colours.

I had momentarily assumed the printer printed things in one orientation. It occurred to me (luckily in about 1 hour into the build in card) that I had the printer upside-down, so 'flipped it' over, with my only needing to re-set two brass threaded inserts into slightly different positions. No big deal, but very glad I didn't labour over this for days without checking, as the printer has internally two very different spatial configurations when inside the machine, so a simple 'flip' would not have been simple then!

Having worked as a freelancer through the Pandemic in 2020/21, I certainly developed some ways of working remotely, such as using WhatsApp's photo 'sketch' tool, to draw on photos what would be a preferable design change. Videocall also was invaluable.

However, despite all these advances in communication, nothing beats hanging out for an afternoon, and Max and I got a good number of meetings at my flat to work through the conceptual, theoretical, ethical and practical aspects of the project...

...Pizza followed, which always helped round-off a good day or development!

Following on from the above point, it can help a lot to draw something, to discuss 'position A or B - which do you think?' with each other, but it is also good to build this in the prototype, as often it's a balance between 'form follows function' and 'yeah, but it just looks right that way'... you just have to do it and see it, touch it, test it...

This is an example of trying to work out where the Pulse Ox should go. It was a bit tricky to fit at the top, so we tried in the 'body' of the device, but this was bad for the electronics' wiring, and upon reflection, we realised we missed the 'finger in the grinder' vibe, with a more 'mad scientist' feel of adding your 'self' to the internals of the machine's working. It just felt right this way.

We even wondered if there should be a 'status' screen - but decided against it, partly for spatial reasons, but also because it was looking more cluttered that way.

This first image is what we settled on - wiring with an intermediary board, and wires more or less the same length. The second image is what we originally tried, but it didn't work. Despite this being 100% correct *in theory* - for reasons we don't know, we had issues in practice. So again, nothing beats testing all peripherals when all connected with each other, not just in isolation.

Was it some cross-talk from components, we don't know...

Perhaps you do - please comment! Put us out of our misery!!

It's one thing to CAD parts which assume one thing happens, when in reality, it's a bit more subtle. In this instance, the receipt tears off at a specific angle, so it makes sense to model that in the CAD, so that the parts allow for it. Ignoring it means the 'teeth' don't 'catch' on the paper and tear it properly, meaning it just keeps spooling out as you tug it.

Similarly, the printer is very close to the OLED screen at the far side, and this requires the pins to be bent down - no big deal, but one needs to be ready for that in the preparation. CAD those details in also!

Another example is for the 'Plates' (inspired by meat grinders having coarse/med/fine plates) - these need to not allow someone to put the 'Coarse' plate on backwards, and trigger the 'Fine' signal. This would result in kids being able to see adult content potentially!

So care was needed in the design to ensure it that either the mistake didn't happen, or is more likely the case, some teenager wasn't trying to 'break' the system for a giggle, whilst it was on display in a museum*, or something!

*(Truth be told, if it were in an unsupervised place, we'd just remove the 'Coarse setting' altogether, and likely run a 'safe mode' routine of presets until we're 1000% certain the AI does not make any dodgy mistakes! Having worked at LEGO, there is 'highly unlikely to go wrong', and there is 'will never ever go wrong', and you learn why the latter is preferable when dealing with safety, kids, etc.).

The main/first image is actually my first idea - to use different resistors to complete a circuit to correspond to a given part of the programme. It was a nice solution, but in the end, reed switches with magnets were easier to code and prototype with, so we went with that.

It sounds obvious, but lighting is 'subjective'... what I find 'cheery' you might find 'dazzling'. What I find 'subtle', you might find 'hard to see'. Safe to say testing out colours, brightness, patterns, durations, etc. - all takes time to get right, and nothing beats showing it to a friend who knows nothing or very little about your work and getting their initial reactions. It's easy to assume your perspective is universal - it usually isn't!

I'm a big fan of creating 'sub assemblies' where a group of functions are able to be collected, and as such iterated on in relative isolation. It means you can update something pretty significant without having to re-print, re-wire, or re-code everything.

Going back to the card model, there were 3 main segments - the body, the input 'cone', and the output 'cone'. Within these there may be smaller sub assemblies to do with the printer, for example, but fundamentally, it was designed in clusters to allow easy changes to occur for functional areas.

I remember hearing as a kid at school that King Henry VIII had 'tasters' - who would eat his food to ensure he was not poisoned. Perhaps an unlikely metaphor, but we used an Arduino (£12) to manage all our complex analogue peripherals, as first the Jetson didn't have an analogue to digital converter, but secondly, because if we blew up the Arduino, it was designed in such a way to not send a jolt down the USB and blow up the Jetson (£225).

It sounds a strange thing, but it allows for some good flexibility in designing prototypes, where late at night you might just plug in the wrong wire, and this saves a world of pain. Even if money were not your concern, it's far easier to reinstall / flash some new firmware on an Arduino than re-initialise a Jetson (we're talking seconds vs hours!!).

Even when the majority of the functionality was resolved, the PCB was still something that needed continual refinement. It went through at least 5 versions with significant changes. And this is before the final PCB design for external fabrication. As much as using protoboard is 'limiting' as of course micro SMD components allow more compact designs, it's worth acknowledging that it helps to get familiar with the layout, assembly, and even what parts you're more likely to keep redesigning and improving upon, and to make them easy access!

Part of the curse of being ‘creative’ is that you often get extra ideas yourself later on, or you get ideas from other people which you want to add. I try wherever possible to leave some options on the table.

For example, I imagined that we might have different operational modes for the Divergence Engine - where perhaps it also suggested music, or poetry, or something else. Perhaps it needed a 'demo mode' that was 1000% safe for kids. What if someone wanted it to have a different lighting sequence?

We added 4 DIP switches, which would allow us 15 other 'preset modes' - with which to create some alternatives should we need it! At the time of writing, we likely need 2 or 3, but it pays to have freedom to 'roam' creatively, for the sake of some easy soldering and a bit of code tweaking!

Safe to say, this project has undergone many small adjustments (the wiring diagram we share is at V2.6). A lot of this comes from living with it and adjusting things based on our irks and curiosities, but also upon feedback from others...

We presented to a warm, techy, and curious audience - at IoT Meetup, hosted for over a decade by the terrific, Alexandra Deschamps-Sonsino, in London. I've known Alex for some years, and it's a real honour to be invited to a 'who's who' collective, but also to have people who really know tech, arts, science, design, engineering - all ask pertinent questions, and give feedback on what they liked.

Max and I have presented at a couple other places now, and this is always a humbling experience, as firstly, we 'read the room' on what jokes land, which polls resonate [hands up who knows Coded Bias documentary / Brian Eno's Oblique Strategy Cards] - and you adjust subsequent keynotes to dial-in better each time.

Indeed, part of writing this magnum opus of documentation is that 99% of folks won't read this, but there are always a few who do, and that can be a remarkable thing to connect with! Even those who skim-read it, also get a sense of the detail and thought gone into something which might otherwise seem quite trivial or quirky, but which we hope, demonstrates some earnest questions and critique of where we are with AI at the moment, and what might be worth changing.

Max and I have applied for a few grants and awards to develop this further. It's early days, but we certainly have some thoughts on where we might explore it next. The general idea is to make the Divergence Engine a little more 'everyday friendly', so other people might use it, and see how it can adapt to different professions / disciplines / interests / needs / circumstances / etc.

We hope you'll suggest any ideas, and get in touch if you'd like to know more!

Thanks and hope this has been a useful insight into how we create a sculptural piece of 'tech art' that is very much an Open Source discussion of what Ai futures might look like for the better...

https://www.judepullen.com/

https://maxpark.design/

And of course, the [award-winning!] build Guide can be found here: (link) - enjoy!