machine apprenticeship

art by Max Hudson - maxhudson.io

The most rewarding part of learning something hard lies in the spaces between formal instruction: observation, toil, and improvement. This is why close apprenticeship under a good mentor feels like a revelation, some inarticulable knowledge is better observed rather than rigidly following a manual or a guidebook.

But the shape of apprenticeship is shifting. AI applications across industries are becoming capable enough to act as intelligent third-party interlocutors between system and human. Experts often can't explain how they make complex decisions or balance nuanced trade-offs built over years of honing their craft. But through humans collaborating and talking to an AI system of record, for example, the system can organize, orchestrate and synchronize best practices for the new generation of workers.

Of course some problems will always resist codification, ones that require human’s last mile judgment. Others, however, become legible with data. Patterns emerge, and decisions that once seemed opaque can be systematized.

It is for this second class of problems that AI apprenticeship will be valuable, as interaction patterns themselves become a proprietary dataset. When AI products get embedded deeply in and automate chunks of daily workflows, users co-create outputs by adding comments and feedback. And if multiple users engage with the same workflow, the system becomes a collaborative interface, capturing the flow of conversation between multiple parties.

The result is an aggregation layer that all participants standardize on. Over time, the system itself becomes an arbiter of quality.

As such, new workers can enter specialized industries with a baseline of knowledge readily available. They can learn from AI systems, leveraging the 'collective memory' of best practices and can utilize an archive of dynamic assets rather than building from scratch.

This can be extremely impactful. For the many industries facing a “grey tsunami” of retiring workers (such as manufacturing, construction) or critical understaffing (healthcare), the golden promise of AI automation, then, is both efficiency and training. Amplify the productivity of a shrinking workforce by acting as a great leveler, raising the floor of performance across the board; Transfer knowledge to a new generation of workers more effectively.

Every AI application might eventually feel like this: a bespoke, purpose-built system that users willingly spend time in, delight in, and co-evolve with.

Form fits function like a glove fits a hand. The best products feel like extensions of the self, and one would hope that learning is a natural by-product of use.

This evolution may mark the age of the agentic generalist as barriers to entry in specialized fields are dismantled. And as competence becomes too cheap to meter, this accelerated learning curve allows workers to climb higher up the pyramid, tackling work that feels more meaningful and impactful.

I think it’s really exciting that there’s multiple new configurations of possible machine apprenticeship as it regards learning and training. Still: mastery remains mastery, craft remains craft. We’ll turn to machine mentors for problems requiring speed, repetition, and precision. We’ll turn to human mentors for more illegible problems — those requiring relationship building, discerning ambiguous human preferences, and the cultivation of taste.

---

Thanks for joining me today on technopoetic. My posts are intended to be short, living drafts. Look forward to more writing in 2025.

---

Software, society, and soul

Technologies are not mere exterior aids, but interior transformations of consciousness… the use of a technology can enrich the human psyche, enlarge the human spirit, intensify its interior life

— Walter Ong, Orality and Literacy

Through this blog, I’m excited to explore the socio-cultural artifacts of AI: how modern systems reshape work, relationships, and meaning-making. I will also directly share investing themes and some learnings from the venture industry.

Here are a few curiosities I am exploring:

• Multiplayer Interface of AI Applications — in a fragmented world, technologies that maintain collaborative consistency are vital for proliferation and consistency of shared idea and thought. One might think of how Figma used different types of CRDT (Conflict-free Replicated Data Type) to create a new landscape for design collaboration. What does the future collaboration interface look like when the system of record becomes an interstitial layer between different parties? What network effects can be built?

• Agency and Recommendation Systems — Beyond the prescient warnings in works like M.T. Anderson's Feed, AI has already cut into certain parts of what it means to be human. I’m interested in how we maintain autonomy and sense-making when our choice architecture is increasingly shaped by recommendation systems. Can we license out trust? How much of our taste is already outsourced? How do we define our preferences and seek them out effectively?

• Pattern Languages in Software — the perseverance of recurring patterns dictate our experience of a place, a product. From architectural theorist Christopher Alexander: beautiful architectural patterns give places a distinct sense of aliveness. Similarly, thoughtful product design can give software a sense of intuitive elegance. How do we create design principles that remain meaningful from individually delightful features to deep product ecosystems? AKA: How do we make software feel fun and alive?

Bio

My name is Nicole. I’m working at a venture firm based in San Francisco. My current focus area is healthcare + other vertical software.

Here are some select books that affected my internal model of the world. Some recent, some less so. If you share a few of these favorites, perhaps we’d get along. As someone once told me, the world of elected affinities is truly small:

• John Williams (Stoner, Butcher’s Crossing)

• Christopher Alexander (The Timeless Way of Being, A Pattern Language)

• Lawrence Weschler (Seeing Is Forgetting The Name Of The Thing One Sees, 30 years of interviews with Bob Irwin)

• Jorge Luis Borges (Labyrinths)

• Walter Ong (Orality and Literacy)

• Stephen Zweig (Beware of Pity, World of Yesterday)

• Shirley Hazzard (The Transit of Venus)

• Ted Chiang (Stories of Your Life and Others)

• Virginia Woolf (To The Lighthouse, Mrs Dalloway)

• Jeanette Winterson (The Passion)

• Greg Egan (Axiomatic)

• Benjamin Labatut (When We Cease to Understand the World)

• Hugh Eakin (Picasso’s War: How Art Came to America)

• Elaine Scarry (On Beauty and Being Just)

• Robert Hass (Apple Trees at Olema - Poems)

• Cixin Liu (Three Body Problem, Dark Forest)

• Brian Eno (A Year with Swollen Appendices)

• Simone Weil (Gravity and Grace)

• Hayao Miyazaki (Starting Point, Turning Point)

• Magda Szabo (Abigail)

Other interests include architecture (Le Corbusier, Olmsted, Alexander), painting (watercolor), and fitness (I have tried a majority of boutique fitness studios in SF). I am trying to watch more films this year! I also write an alter-ego blog on craft, literature, and what we can mean to one another in this brief lifetime.

“There's an ancient saying in Japan, that life is like walking from one side of infinite darkness to another, on a bridge of dreams. They say that we're all crossing the bridge of dreams together. That there's nothing more than that. Just us, on the bridge of dreams.”

— Feed, M.T Anderson