"To Know" in the AI Age
- 9 minutes read - 1728 wordsOne of the most charming aspects of English and many other Old-Norse-derived/-cognate languages are kennings. Kennings are words created by bonding multiple words, usually to poetic effect. The Beowulf poet speaks of “the whale road” to mean “the ocean.” Elsewhere, a “sword-storm” describes a battle. The kenning succeeds where its constituent words fall short on the same grounds as Impressionist art:
an artist's work as framework
+ the perceiver's personal history and experience
= living, breathing artistry
The infinitive to know is in need of rescuing.
Over these last ten years, I’ve been party to multiple conversations where misunderstood expectations around to know have hampered effective communication. While “to know” might have been wheezing under the just-in-time epistemology of the internet age, the arrival of AI in workplace, school and home is a type of respiratory failure for this overloaded infinitive.
An example from my own profession: misaligned expectations around “to know” or “knowledge” have proven onerous and lead to frustration on behalf of job role candidate and manager.
But using kennings to examine our usage of this verb might allow us to speak more accurately about our knowledge.
Making Contact with Multiple “To Knows”
For many native English speakers, the poverty of English’s “to know” first becomes evident when they encounter Romance languages. In French, there’s the difference between knowing a fact (savoir) and knowing a person (connaître). This is also repeated in Spanish and Italian. As one explores other languages, one may come across other languages’ conceptions of different types of knowledge that we obscurely render in English as “to know.”
OK, so we can know facts, or we can know people. That’s the tip of the iceberg. But we can know as-in savoir different types of facts based on their provenance.
Daniel Everett’s work with the Pirahã (which I encountered in Don’t Sleep, There Are Snakes) relates that within this tribe speakers are required to append a signal that specifies the source of knowledge that they are relating. They can specify a fact as coming from:
- Hearsay
- Direct observation
- Deduction from evidence
As the action of the book has it, one can readily understand why it would have made evangelizing to this group uniquely challenging. As I recall: “Dan did you see Jesus? No. Did your father? No. His father? No. Oh, so this is nonsense. Why are you telling us nonsense?” It’s an interesting book.
In any case, we can readily imagine sub-types of hearsay:
- Things our tribe upholds (e.g. religious values, taboos)
- Things I heard from a living kinsman (appeal to patriarchal, trusted authority)
- Things (either of the above) that happened within some time period that we recognize as “recent” (appeal to newsworthiness, recency bias, or irrelevance due to age e.g. “Our tribe knew we were the only people on Earth (…until the Europeans came)”)
Subtypes of observation:
- I saw with my eyes
- I experienced with some other sense
- I perceived with a trusted/untrusted tool (e.g. Galileo showing the actual state of the planets versus what the Church and Aristotle held to be their dogmatic status)
Subtypes of deducted fact i.e. inference:
- I perceived with a (observation via tool) tool which affirms a deduction (e.g. “I heard a thump-thump via the stethoscope and I have studied the human heart and thus I conclude you have a measurable blood pressure.”)
- This polygon has three connecting sides (observation via eyes) and therefore its interior angles sum to 180° (deduction)
Riffing a bit, we can readily imagine a special word for knowing something as if gifted by the gods (“adspirate meis priumumque ab origine mundi”): inspiration.
Or perhaps there ought be a type of knowledge that is visceral/intuitive grasp
Or perhaps there is a thing you know, but only at a subconscious, kinaesthetic or proprioceptive level like riding a bicycle.
When you think about it for any length of time, to know falls over like a Potemkin storefront: its a cipher and a mystery whose meaning is largely dependent on the speaker, context, and culture in which the conversation happens. But this is no mere “Isn’t language interesting” philo-linguistic music. Given that I have the professional responsibility of educating software developers, specifying objectives and outcomes that lean on this leaky term can be vexing. Let me dig a bit into that.
The Epistemic Misalignment Problem
When we speak of training software developers, or when we speak of designing curriculum, stakeholders and educators will make liberal appeals to “knowing” and “to know” — but we now recognize doing so as grossly imprecise.
Over the last ten years, I’ve taught a lot of people coding and engineering. In this era, it was not rare for me to walk up, look at a piece of code and probe the author’s understanding. Oftentimes I would nod, assent, and agree that they understood. They would agree that they understood. But if I highlighted the code and deleted it, they would look at me in horror and strike “Undo” as quickly as they could – and it was rare that I’d deleted more than a dozen lines. Yet in that instant, in some very real way, it was very clear that they did not know for certain understandings of that word. And I’m not an empathetic monster, I myself have often fooled myself by reading a bit of code or a book and thinking “Oh, I know this.” But then when faced with a new code buffer, when facing “that big black empty screen feeling,” I didn’t know where to start and what had previously felt I knew so well was an illusion.
What if one could indicate a type of “to know” where they mean “With documentation and a few examples at hand.” Or what if one could say “I know I had practical skill with this once, and I have work to prove capability, but at the moment I cannot” – that’s me with knitting.
Or what if a supervisor understood that a given task was to be completed on quirky hardware and the expectation would be to understand the mechanisms of some technology from their absolute first principles. Someone who knows that technology on a later hardware profile might not meet expectation.
Many managers taking on early-career new-hires are frustrated when the new joiner knows something, but they don’t know it the right way or the right “depth.”
These misalignments lead to flawed recruiting, flawed candidate/role matching, flawed mentorship, flawed estimation of timelines – all because of the outsize obfuscating power of to know. And here’s where, perhaps, the kenning shows the way: what if instead of the flat nigh-vacuous “to know” we had kennings with it that meet our current epistemological moment?
A New Vocabulary for AI-Assisted Knowing
Taking our cue from our Jutish/Anglo-Saxon forebears, perhaps we can create word-pictures that articulate more precisely the type of knowledge we possess:
know-(claude|chatgpt|gemini|et al) (verb)
To understand a concept well enough to recognize, evaluate, and apply AI-generated solutions, but not to generate them independently.
“I know-claude React hooks - let me get back to my desk and I can have you a groomed edit before you’re back from coffee.”
know-google (verb)
To possess sufficient conceptual familiarity to construct effective searches and recognize correct solutions among results.
“I know-google how to configure an .xinitrc or an .xsession but I can
never remember which applies only to startx. Let me get back to my desk and
set you up with that window manager I was just talking about.”
know-prompt (verb)
To understand a domain well enough to craft effective prompts that are likely to generate close-to-valid output.
“I know-prompt the evolution of the Apple product line.”
know-memory (verb)
To possess deeply internalized knowledge that can be accessed and applied without external aids; traditional “knowing by heart.”
“I know-memory my multiplication tables, but I only know-claude machine learning algorithms.”
know-muscle (verb)
To have developed physical or procedural memory for a task through repetition; embodied knowledge that flows without conscious effort.
“I know-muscle touch typing and parallel parking, but I only know-google the different levels of database normalization.”
know-tribe (verb)
To possess knowledge through belonging to a community of practice; understanding gained through shared experience and cultural transmission.
“As a successful venture capitalist, you can trust I know-tribe how to actually get to Atherton from Mountain View.”
know-say (verb)
To possess secondhand knowledge without direct experience; understanding based on reports, testimonials, or social proof; short for “know-hearsay.”
“I know-say that TypeScript improves code quality - everyone says so, but I’ve never actually used it on a large project.”
know-bones (verb)
To possess intuitive, visceral understanding that transcends logical explanation; knowledge felt at a cellular level.
“I know-bones that that smiling couple is minutes from divorce.”
There are doubtless more, English speakers will readily recognize all these types of epistemic assessment are flattened into “to know.”
Why This Matters
As we enter a world where AI assistance will be ever more ubiquitous, the distinctions will matter practically in order to prevent communication breakdown. I expect pilots in the air to know-memory how to pilot an airplane and I’d like them to know-tribe the subtleties of landing at JFK and I’d hope their simulator time affords them know-muscle memory on the level of Captain Sullenberger how to act when emergency measures are required.
Or, to revisit an earlier example, when hiring developers, hiring managers may expect a know-say familiarity with editors, but they should definitely have know-muscle fluency with their primary coding tool. On the other hand, know-prompt skill is expected with data science algorithms, but (perhaps) know-memory capability is required on all core searches and sorts.
On the flip side, trainees might reasonably complain that they’re being short-changed when they expect to learn to know-memory something but the business only requires know-tribe or — perhaps even more jarringly in the present environment — know-prompt capability. In an imperfect world with vastly more things to know than there is sufficient time or attention, being able to pinpoint level of knowledge is fair and forthright and, if we’re honest, required.
Use of to know-kennings might require of the speaker the additional introspection that our current use of to know wallpapers over. It might also allow good candidates to prepare more fruitfully and allow them to enter the hiring negotiation from a less-adversarial place. As admonished from ancient Greece, let us not speak carelessly of the most-significant things.