For as long as we’ve had computers, they’ve produced predictable outputs. But AI – in the form of large language models – represents a new kind of unpredictable computing. The key to implementing useful AI solutions is making the most of both paradigms. One of the oldest known computers is the Antikythera mechanism, an ancient device for calculating astronomical events. Given certain inputs, it computed positions based on logic hard-coded in its gears. Traditional software is kind of like that: it determines what to do based on pre-defined conditions. You give the computer input and get predictable outcomes. If a program produces unexpected results, it’s either because the programmer introduced randomness or because there are bugs. Both can be replicated by mirroring the exact conditions that led to the outcome. Because of this, traditional computation is deterministic. Modern AI, such as large language models, represents a new computing paradigm. If you’ve used ChatGPT or Claude, you know you seldom get the same results given the same input. Unlike traditional programs, LLMs don’t follow explicit instructions. Instead, they generate responses by weighting probabilities across a vast network of linguistic relationships. There can be many paths to possible likely responses. This is a new kind of probabilistic computing. Much of what we value about computers is due to their predictability. That’s one reason why so many people find LLMs baffling or objectionable: probabilistic behavior breaks our mental models for how computers work. Probabilistic computing is good for some tasks but not others. Brainstorming is a good use case since you’re explicitly asking for divergent thinking. On the flip side, math requires deterministic approaches. LLMs can do it by offloading computations to deterministic systems like Wolfram Alpha. Prompt engineering is an attempt to constrain probabilistic processing to make LLMs behave more predictably. But it only goes so far: you can’t force LLMs to behave like traditional programs. A better approach is building deterministic software that uses AI at particular junctures for specific tasks. An example is my approach to re-categorizing blog posts: a deterministic program iterates through files, offloading pattern matching to an LLM. The LLM is used only for stuff probabilistic systems do well – the inverse of the Wolfram Alpha approach. This new paradigm offers unprecedented opportunities. But taking advantage of probabilistic systems requires adding some determinism to the mix. You can’t ask ChatGPT to re-organize a website, but you can build scaffolding using traditional approaches that take advantage of what each does best. If you work with content, it behooves you to learn how to combine AI’s probabilistic approach with the traditional deterministic approach. That’s what I’ll be teaching in my hands-on workshop at the IA Conference in Philadelphia in late April. Join me there to learn how to do it.

In week 11 of the humanities crash course, I revisited one of the most influential philosophies of the ancient world: Stoicism. I was already familiar with this material, so I also took the opportunity to revisit another text: the Book of Job. Heavy stuff – and I paired it with an even heavier movie: a modern classic which I’d not yet seen but has become a new favorite. Readings Gioia recommended two readings: Marcus Aurelius’s Meditations and Epictetus’s Enchiridion, a short manual of advice. I’d read Meditations twice before and was familiar with several of Epictetus’s aphorisms. Both are central works of Stoicism, an ancient philosophy that remains highly relevant. Because I was familiar with the material, I chose to also go beyond the bounds of Gioia’s syllabus by revisiting one of my favorite books of the Bible, the Book of Job. Not only did I read Job itself, but also Mark Larrimore’s The Book of Job: A Biography. The selection wasn’t accidental. Job deals with similar questions as the Stoics, but arrives at different answers. In this section, I’ll provide an overview of the readings, and will get into the parallels and differences in the reflections below. Meditations was written between 161 and 180 CE by the Roman emperor Marcus Aurelius. The book wasn’t meant for publication. Instead, it’s a series of notes-to-self: Marcus is nudging himself to stay on the path of virtue. It’s the same path Epictetus (who was at one point a slave – a very different status from emperor!) lays out in the Enchiridion. Namely, we’re responsible for our experiences of reality. Some things we can control, either fully or partially, and some we can’t control. Trying to control things that are out of our control leads to unhappiness. The Stoics saw the universe as rationally ordered; they aspired to alignment with nature. We must embrace things as they come. This can be very painful – e.g., the death of a loved one. The Stoics saw challenges as opportunities to refine our character, and prescribed exercises such as negative visualization, journaling and reflection, and voluntary discomfort. The Book of Job also focuses on suffering. Job is a rich man blessed with a great life and a thriving family. He worships God. But Satan suggests that God test him: perhaps Job is a fair weather friend. God allows Satan to take everything from Job: his wealth, his family, and ultimately, his health. Three friends come to comfort him, but instead default to the traditional interpretation of suffering: that it must be divine punishment. They insist Job must’ve done something wrong to merit such treatment. Job denies it; they argue in a long poem that forms the bulk of the book. A fourth friend suggests another take: perhaps the suffering is a means for instruction. Eventually, God addresses Job from a whirlwind. His message: as a mere man, Job can’t grok the mysteries of divine will. God is all-powerful; his ways unknowable. So there is no point in asking why. Job takes back his complaining and questioning in one of the book’s most famous passages: Then Job answered the LORD and said: After this follows an awkward prose epilogue where God restores Job’s riches and even gives him a new family. I say ‘awkward’ because it feels different in tone and intent from the rest of the book. (Isn’t this a regression to the ‘divine punishment’ view?) In the Biography, I learned that Job might be the work of several authors. Whether that’s true or not, the prose frames certainly feel different to the rest of the book. Audiovisual Music: Three Haydn symphonies: 45 (Farewell), 94 (Surprise), and 104 (London). I’d heard 94 and 104 before, but 45 was new to me. Of all the major classical composers, Haydn is one I’ve not paid much attention to; these listenings were an invitation to dig deeper. Art: Gioia suggested looking at ancient Roman art and architecture. During my architectural studies, I spent two semesters in Rome and spent time living among these works, so I didn’t dwell much on this subject now. That said, I watched this episode of Rick Steve’s Europe, which is a short and engaging overview: Cinema: I asked Perplexity for movies inspired by the Book of Job. First on its list was Terence Mallick’s THE TREE OF LIFE. I’d never seen a Mallick movie, and this was already on my to-watch list. But what pushed me over the top was seeing it mentioned in The Book of Job: A Biography as a movie that explored these themes. THE TREE OF LIFE is a slow, poetic, and impressionistic meditation on suffering via a portrait of a small-town American family in the 1950s. The movie is relatively recent, so I won’t spoil it. Suffice it to say, family members experience joy and suffering – and like Job, wonder about the meaning of the latter. The intimate documentary style and gorgeous cinematography make for a highly emotional experience – especially for parents. I felt strong echoes of one of my favorite movies, 2001: A SPACE ODYSSEY, which is also a slow, poetic, and impressionistic meditation on philosophical themes. There are scenes of astronomical events in THE TREE OF LIFE that could’ve easily been in 2001. Still, I was surprised to learn that Douglas Trumbull (SFX lead for 2001) also worked on THE TREE OF LIFE. Both films also feature transcendent classical music soundtracks. Reflections This week’s works focused on one of the central questions of the human experience: how do we deal with suffering? This isn’t the first time we grapple with this question; suffering is central in Buddhism, and we’ve already read the Dhammapada. The Stoics and Job provide answers from a Western perspective. In many ways, Stoic teachings parallel Buddhism: Both aspire to a clear understanding of reality by accepting what is happening without judgment. Both aim to alleviate suffering by transcending it. Both identify suffering with desire: yearning for things to be different than they are. Both emphasize the dynamic nature of reality – we can’t hold on to things, since everything is in flux. Both advocate dealing with events (either positive or adverse) with equanimity, composure, and dignity. One major difference: Stoicism places more emphasis on becoming the best possible version of oneself, whereas Buddhism emphasizes reducing suffering for all beings. The Book of Job also explores these ideas. But while Stoicism and Buddhism advocate developing tolerance for suffering, Job emphasizes its unknowability. To put it crudely: shit happens; the Stoics and Buddhists prescribe personal development whereas Job prescribes surrender to divine will – i.e., transcending the ego. Mallick’s film is saturated with the latter. The film opens with a quote from Job: Where were you when I laid the earth’s foundation… while the morning stars sang together and all the sons of God shouted for joy? Its two central characters – the parents – have different approaches: the father is driven by nature/will and the mother is driven by grace. Left unchecked, both cause suffering; balance seems the best path. But neither provides definitive answers to the question “Why do we suffer?” Rather, the answer seems to be: suffering and joy come with the package. As Job puts it when he learns of the death of his children, Naked I came from my mother’s womb, and naked shall I return. The LORD  gave, and the LORD has taken away; blessed be the name of the LORD. It’s pointless to ask why. A better question is: how do we deal with suffering? These works provide important answers. Notes on Note-taking As with previous weeks, I took notes in Obsidian. Since this was my third reading of Job and Meditations, I expanded already-existing notes for both books. I also started a separate note for Stoicism itself, since this philosophy has ideas that relate to many other works. In all cases, I used GPT-4o (both via the chat interface and through the Text Generator plugin) to expand on my understanding. I asked the LLM for feedback on my summaries, comparisons between these works and each other, and between the ideas in these works and the Buddhist tradition. The results were often pithy and clarifying. Here’s an example: While both Stoicism and Buddhism offer paths to transcend suffering through acceptance and detachment, their ultimate goals differ: Stoicism emphasizes personal virtue and rationality, whereas Buddhism focuses on universal compassion and enlightenment. Up Next Gioia recommended Suetonius’s The Twelve Caesars for week 12; standard Ebooks has a beautiful free edition. Check out Gioia’s post for the full syllabus. Again, there’s a YouTube playlist for the videos I’m sharing here. I’m also sharing these posts via Substack if you’d like to subscribe and comment. See you next week!

Recently, I wrote about using AI to solve taxonomy drift — the all-too-common problem of lists of terms (tags, categories) falling out of sync with the content they describe. A response to that post raised an important distinction worth clarifying: the difference between creating taxonomies and applying them. First, a bit of context. I’m talking specifically about taxonomies for organizing web content. CMSs like Drupal and WordPress allow authors to tag content items, helping users find information later. But both content and taxonomies evolve over time, and tagging consistently is a challenge — especially for small, resource-strapped teams. In my earlier post, I suggested AI can help with this challenge. But what, exactly, should AI be doing? Here’s where confusion arises. The approach I’m working with uses AI to tag content with terms from a predefined taxonomy. This is different from having AI generate new taxonomy terms. Put simply: I’m using AI to apply taxonomies, not create them. These are different challenges. AIs are better than humans at spotting patterns in text at scale, which makes them useful for applying terms consistently. But humans are still better at the skills needed to define taxonomies — deciding what terms should be included and how they relate to each other. A taxonomy isn’t just a set of words and phrases; it’s also a model that reflects how users understand the domain, aligns with an organization’s strategic goals, and fits within broader cultural norms. Defining this model requires, among other things, judgment, contextual awareness, and an understanding of strategic priorities and organizational politics. Doing these things well is beyond the grasp of current AI systems. Which isn’t to say AI can’t help – of course it can. LLMs excel at processing large volumes of content, finding patterns, summarizing, and outlining. In this capacity, they can be an invaluable aid to taxonomists and information architects defining or refining taxonomies. But there’s another way AI can help – one that’s both more profound and more exciting. Taxonomies aren’t static or defined in the abstract. Often, gaps become apparent only during application. You might realize you’re missing important terms when you start tagging content at scale. Or you might find that part of the model isn’t quite right and needs adjustment. The problem is that tagging content at scale takes time, so you can’t easily see the effects of your changes until much later. This isn’t how other creative disciplines work. When painters daub oil on a canvas, they can see how colors spread and combine in real time. They alter the mix and strokes to achieve particular effects. They can do this because they get immediate feedback on the effects of their actions. Taxonomists have lacked this level of feedback at scale — until now. AI makes it possible to see how distinctions apply across large sets of content much faster. It helps reveal gaps, inconsistencies, and opportunities for refinement at a different scale and speed, pointing to a new way of working. So, even though defining and applying taxonomies are different tasks, AI will likely blur the line between them. And that might be a good thing — so long as humans are still wielding the brushes and palettes.

AI is everywhere, but most websites are still managed manually by humans using content management systems like WordPress and Drupal. These systems provide means for tagging and categorizing content. But over time, these structures degrade. Without vigilance and maintenance, taxonomies become less useful and relevant over time. Users struggle to find stuff. Ambiguity creeps in. Search results become incomplete and unreliable. And as terms proliferate, the team struggles to maintain the site, making things worse. The site stops working as well as it could. Sales, engagement, and trust suffer. And the problem only gets worse over time. Eventually, the team embarks on a redesign. But hitting the reset button only fixes things for a while. Entropy is the nature of things. Systems tend toward disorder unless we invest in keeping them organized. But it’s hard: small teams have other priorities. They’re under pressure to publish quickly. Turnover is high. Not ideal conditions for consistent tagging. Many content teams don’t have governance processes for taxonomies. Folks create new terms on the fly, often without checking whether similar ones exist. But even when teams have the structures and processes needed to do it right, content and taxonomies themselves change over time as the org’s needs and contexts evolve. The result is taxonomy drift, the gradual misalignment of the system’s structures and content. It’s a classic “boiled frog” situation: since it happens slowly, teams don’t usually recognize it until symptoms emerge. By then, the problem is harder and more expensive to fix. Avoiding taxonomy drift calls for constant attention and manual tweaking, which can be overwhelming for resource-strapped teams. But there’s good news on the horizon: this is exactly the kind of gradual, large-scale, boring challenge where AIs can shine. I’ve worked on IA redesigns for content-heavy websites and have seen the effects of taxonomy drift firsthand. Often, one person is responsible for keeping the website organized, and they’re overwhelmed. After a redesign, they face three challenges: Implementing the new taxonomy on the older corpus. Learning to use the new taxonomy in their workflows. Adapting and evolving the taxonomy so it remains useful and consistent over time. AI is well-suited to tackling these challenges. LLMs excel at pattern matching and categorizing existing text at scale. Unlike humans, AIs don’t get overwhelmed or bored when categorizing thousands of items over and over again. And with predefined taxonomies, they’re not as prone to hallucinations. I’ve been experimenting with using AI to solve taxonomy drift, and the results are promising. I’m building a product to tackle this issue, and looking implement the approach in real-world scenarios. If you or someone you know is struggling to keep a content-heavy website organized, please get in touch.

Episode 5 of the Traction Heroes podcast featured a conversation about strategic priorities. What do I mean by that? I’m referring to prioritizing at two levels: When you’re thinking about what to focus on now – what you should and shouldn’t do next. When you’re thinking about the higher-level objectives those granular actions serve. What use is it to be productive or efficient if it’s moving you in the wrong direction? Or as designers put it: there’s a difference between designing the right thing and designing the thing right. Of course you want to do things right. But first, you want to ensure the thing you’re doing is worthwhile to begin with. Both levels of priorities matter. As you may expect, this is straight up Harry’s alley, as it’s central to Managing Priorities. I was glad to have this conversation with him – especially since it was prompted by a passage from a book he found through me! I hope you find this conversation valuable. As always, I’d love to know what you think. (And if you’re enjoying the show, please leave us a rating – it helps other folks find us.)