Latest / Deep Dive / Why the Most Frustrating Parts of Your Job Might Be the Reason AI Can’t Replace You
Transcript
- 0:00What if? What if the messy, frustrating
- 0:03friction you feel at work everyday?
- 0:05You know those parts of your job that make you want to literally
- 0:08pull your hair out? Oh yeah, we all have those.
- 0:10Right. What if that friction is
- 0:12actually the exact thing saving you from being replaced by
- 0:16artificial intelligence? It's a crazy thought, but I
- 0:19mean, it's incredibly relevant right now.
- 0:21It really is because if you spend like any time online
- 0:25lately, you are just drowning in these apocalyptic hot takes
- 0:29about generative of AI. Absolutely.
- 0:31The noise is deafening. Yeah, we're constantly told the
- 0:33creative industries are collapsing, that machines are
- 0:36going to write all our scripts, design our buildings, compose
- 0:38our music. Trying to figure out the actual
- 0:41truth of human work right now feels completely impossible.
- 0:45Well, the fundamental flaw in almost all of those panic driven
- 0:47narratives is that they treat creativity like it's like it's a
- 0:52singular magical spark. Like a light switch basically.
- 0:55Exactly. A switch that's either flipped
- 0:57by a human or flipped by a machine.
- 0:59But when you move AST the headlines and actually look at
- 1:01the granular mechanics of work, the ecosystem of human labor is
- 1:05infinitely more complex than just, you know, simle
- 1:08substitution. Which brings us to the mission.
- 1:11For this deep dive, we're throwing out all that
- 1:13speculative panic. Thank goodness.
- 1:15Right, We're focusing entirely on hard, granular data.
- 1:18Today. We've got this incredibly dense,
- 1:20groundbreaking academic paper from our MIT university.
- 1:23By Emmanuel Wolkowiak. Yes, exactly.
- 1:26And what makes this research so compelling is the foundation it
- 1:29uses the December 2023 Australian Skills
- 1:32Classification. So this is not just some
- 1:34theoretical thought experiment. No, not at all.
- 1:37They took 126 different occupations across the cultural
- 1:40and creative industries and they dissected them.
- 1:43Down to 593 highly specific specialist tasks, we're looking
- 1:47at the actual daily mechanisms of what creative workers do to
- 1:51earn a living. And by breaking work down to
- 1:53that molecular level, the data asks a much more sophisticated
- 1:57question instead of that tired Will AI replace designers?
- 2:02Debate. We're looking at how humans and
- 2:05machines navigate specific rules together.
- 2:08And the study operates through this lens of evolutionary
- 2:11economics. Yes, let's ground that concept
- 2:13real quick before we dive into the actual numbers because
- 2:15evolutionary economics sounds a bit, you know, academic.
- 2:19How is that different from how we normally think about tech at
- 2:22work? Well, traditional economics
- 2:24usually views tech as this static disruptive force.
- 2:27Like a more efficient machine shows up, it substitutes the
- 2:30human worker and boom, the equation is solved.
- 2:32Like the automated loom replacing the hand Weaver.
- 2:35Exactly. But evolutionary economics
- 2:37argues that economies are complex living systems, humans,
- 2:40institutions, technologies. They don't just replace each
- 2:43other, they constantly Co evolve.
- 2:45Co evolve. OK.
- 2:47Do we have a good historical parallel for that?
- 2:48Yeah, a great one is the introduction of spreadsheet
- 2:51software in the 1980s. People genuinely assumed it
- 2:54would completely eradicate the accounting profession because
- 2:57obviously a computer could calculate instantly.
- 2:59Right. Why do you need a human if the
- 3:01math is automated? Exactly.
- 3:03But instead, the tech just absorbed the mechanical
- 3:06calculations and the role of the accountant evolved into
- 3:11strategic financial advising. Oh wow, so the technology
- 3:14reshaped the environment and the human adapted to a higher value
- 3:18function? Precisely.
- 3:19OK, so we're looking at how the rules of the work itself change
- 3:22as Gen. AI enters the chat.
- 3:25And to figure out if AI can actually execute a creative job,
- 3:29we first have to decode the how. Right, the underlying rule book
- 3:32of creativity. Yeah, so the study categorizes
- 3:35work by breaking it down into cognitive rules, which govern
- 3:38how you mentally process info, and behavioral rules, which
- 3:41govern how you physically act. And it puts every task into 3
- 3:45distinct buckets. Right, so the first bucket is
- 3:47categorized as defined rules in the language of this study.
- 3:51This is the retention phase of knowledge.
- 3:52Retention phase meaning what exactly?
- 3:55These are explicit, highly documented, step by step
- 3:58procedures. Think of it as the fully
- 4:00codified parts of a job. The parameters are locked in
- 4:03place and the goal is to just successfully retain and
- 4:07replicate a known process. And the numbers on this first
- 4:10bucket are striking. The study reveals that when a
- 4:12task relies heavily on these defined rules, creativity
- 4:15actually plummets. Yeah, negatively correlates with
- 4:18creativity. Dropping the score for
- 4:20Intentional Creativity by up to 59.3 points.
- 4:23And if a task is fully codified, generative AI absolutely
- 4:27dominates. Completely.
- 4:28The machine is practically designed to retain and execute
- 4:31defined rules with an efficiency we just can't match.
- 4:34But work rarely stops at just retention, which brings us to
- 4:37the second bucket, which is tacit rules.
- 4:40The adoption phase. Right and tacit rules are
- 4:42unpodified. They represent experiential
- 4:44knowledge, gut feelings, intuitive judgment.
- 4:47It's that deep expertise you develop over years of physical
- 4:50and social practice. Yeah, the stuff you just cannot
- 4:52write down in an instruction manual.
- 4:54And the data points to tacit rules as the golden ticket here.
- 4:58They are the strongest predictor of creativity in the entire
- 5:00study. Boosting adaptive creativity by
- 5:03a massive 71.8 points. Yeah, 71.8 points.
- 5:07And here is the crucial detail. When the study's tasks are
- 5:10categorized as fully tacit, they are 100% matched to human only
- 5:15workers. Zero machine involvement.
- 5:18Not at all. OK, let's unpack this with a
- 5:20metaphor. It's like being in a kitchen.
- 5:23Yeah, I. Like that?
- 5:24Defined rules are the strict baking recipe.
- 5:27You weigh exactly 200 grams of flour.
- 5:30You set the oven to 350°. Bake for 20 minutes.
- 5:33Right. And a machine can follow that
- 5:35Florida State. Exactly.
- 5:36But tacit rules represent the Baker knowing exactly how the
- 5:39dough should feel in their hands.
- 5:41They intuitively add a splash more water because they can
- 5:44literally feel the humidity in the room even though the recipe
- 5:47never mentions the weather. What's fascinating here is that
- 5:51physical awareness is central to tacit knowledge.
- 5:53It's fully embodied. Which brings us to the third
- 5:56bucket, completing the spectrum. Novel rules.
- 5:58The origination phase, yes. This occurs when there is no
- 6:01existing framework at all. You have to create an entirely
- 6:04new rule book from scratch. So back to the kitchen.
- 6:08This would be the Baker inventing an entirely new
- 6:10category of pastry that no one has ever tasted before.
- 6:14Pure. Unadulterated creativity favors
- 6:18humans heavily. I have to pause here though, and
- 6:21push back a little on this idea that tacit knowledge is an
- 6:24impenetrable shield for us. OK, go for it.
- 6:27Because if tacit knowledge is fundamentally just unwritten
- 6:30roles like the Baker feeling the dough, why can't an AI simply
- 6:34observe a human Baker for say, 10,000 hours?
- 6:37I see where you're going. Right, Like with enough visual
- 6:40data and sensor feedback, couldn't a machine eventually
- 6:43learn the feel of the dough? Isn't tacit knowledge really
- 6:46just data we haven't figured out how to measure yet?
- 6:49That is the core argument of technological determinists,
- 6:52actually, but it misunderstands the nature of what walkway at
- 6:55calls rule ambiguity. Rule ambiguity, yeah.
- 6:58Generative AI requires low ambiguity to function in these
- 7:01clear parameters, even if those parameters are highly complex.
- 7:05The tacit knowledge involves sensing shifting, unquantifiable
- 7:08and often contradictory variables simultaneously.
- 7:11The Baker isn't just measuring humidity.
- 7:13They're adjusting based on the specific crop of wheat that
- 7:16month. Exactly.
- 7:17And the temperature of their own hands.
- 7:19The ambient airflow in the room. As a task becomes more tacit,
- 7:23the ambiguity skyrockets. AI fails in high ambiguity
- 7:28environments because it lacks physical embodiment.
- 7:31It can't instinctively shift context that has no lived
- 7:34experience to draw on. Right, but obviously every
- 7:37single creative job involves some amount of defined rules
- 7:40too. Which is why people assume mass
- 7:42unemployment is coming, because if AI is that dominant at
- 7:45defined rules, the logical outcome feels like we're all out
- 7:48of a job. But the data says the exact
- 7:50opposite, yeah. So let's look at the mezzo
- 7:52level, the actual job level, to understand why this math isn't
- 7:55math. Well, let's slow down and look
- 7:57at the granular task level statistics first.
- 7:59The study found that 86.3% of tasks show human AI
- 8:04complementarity. Meaning the human and the
- 8:06machine are actively working together.
- 8:08And 11% of tasks are completely immune to AI, while a
- 8:12microscopic 2.7% of tasks face pure AI substitution.
- 8:16Wait, take a second to explain that 11% of immune tasks.
- 8:21What does an immune task actually look like mechanically?
- 8:24Why does the machine hit a brick?
- 8:25Wall, an immune task is one that is completely saturated with
- 8:29tacit and novel rules. There is essentially 0
- 8:33codifiable retention phase. So nothing to memorize or
- 8:36repeat. Exactly.
- 8:37Imagine a live theater director giving notes to an actor.
- 8:41The director is reading the subtle emotional temperature of
- 8:44the room. They're observing microscopic
- 8:46tension in the actors shoulders. And adjusting feedback in real
- 8:50time based on the fragile egos of the cast.
- 8:52Right, there's no defined rulebook for that.
- 8:55The ambiguity is absolute. A machine cannot execute the
- 8:57task because there is no retained data to extrapolate
- 9:00from. The data is being originated in
- 9:02the moment, through human empathy.
- 9:05OK, so that's the micro level, but when we zoom out to the
- 9:08broader job level, the numbers get even more wild.
- 9:10They really do. The study notes that 57.9% of
- 9:13jobs are actively transforming and AI has the capability to
- 9:17complete at least some tasks and 65.1% of jobs.
- 9:21But Despite that, massive penetration into daily tasks.
- 9:24Zero jobs in this study are entirely AI dominated.
- 9:27Wait, hold on, if AI can do components of 65% of jobs but
- 9:32zero jobs are taken over, how does that math actually work?
- 9:36Because a job is not just a loose pile of isolated tasks,
- 9:39Wakuiak defines jobs as mezzo level institutional
- 9:43arrangements. Meaning they're complex
- 9:45coordination mechanism. Exactly.
- 9:46You can't simply slice off all the defined rules, hand them to
- 9:49a large language model and walk away.
- 9:51The output requires integration. Right, like who checks the AI's
- 9:55work? Who decides if the AI's rigid
- 9:57output actually fits the uncodified tacit context of what
- 10:00a specific client wants? So wait, the study calls this
- 10:03the triple triple configuration, right?
- 10:05It sounds like a massive balancing act.
- 10:07Are you saying the human is basically acting as a traffic
- 10:09cop between the rigid AI output and their own gut instincts?
- 10:13That is a highly accurate way to visualize it.
- 10:15The data reveals that 91.3% of jobs in these industries fall
- 10:20into this triple triple configuration. 91.3% Wow.
- 10:23Yeah, it means the human worker must simultaneously juggle
- 10:28defined, tacit and novel rules across both their cognitive
- 10:32reasoning and their physical actions.
- 10:34You're just constantly switching contexts.
- 10:36Right. You follow strict protocols one
- 10:38minute, rely on pure physical intuition the next, and then
- 10:41generate unprecedented ideas right after that.
- 10:43Yeah, and AI fractures when forced to rapidly switch across
- 10:47those cognitive paradigms. But the human acts as the
- 10:49ultimate project manager, stitching all those isolated
- 10:52outputs together with overarching judgment.
- 10:54The human is the little glue holding the process together.
- 10:58I mean, an AI might generate A flawless architectural rendering
- 11:01based on strict math the defined rules, but the human architect
- 11:06intuitively knows the client is going to hate that layout
- 11:09because of a subtle hesitation they showed during a site visit
- 11:123 weeks ago. The AI is entirely blind to that
- 11:14tacit reality. Yeah, if you remove the human
- 11:17coordinator, the defined rules generated by the AI become
- 11:20useless. They aren't anchored to the real
- 11:22world. The job itself is the
- 11:23integration point. Human and machine must Co
- 11:27evolve. In fact, hybrid agents, where
- 11:30the human and AI collaborate, now dominate roughly 74% of the
- 11:34tasks studied. OK, if hybrid dominance is the
- 11:37reality for 74% of tasks, what does an ideal collaboration
- 11:41actually look like? Because there is a stat in this
- 11:44section of the paper that makes absolutely no intuitive sense at
- 11:47first glance. Oh, the structured novelty
- 11:48effect. Yes, we usually assume AI needs
- 11:51boring routine environments to function or else it hallucinates
- 11:55and breaks down. But then they introduce the
- 11:57structured novelty effect. Here's where it gets really
- 12:00interesting. It completely upends our
- 12:02assumptions about machine creativity.
- 12:04The data proves that AI autonomy surprisingly increases by .114
- 12:09points when you combine high levels of strictly defined
- 12:11cognitive rules with the human creation of entirely novel
- 12:15rules. That seems like a total paradox.
- 12:17How does introducing pure novelty make a machine perform
- 12:21better? Because of how the environment
- 12:22is structured, when a human worker originates purely novel
- 12:26rules, they are stepping into the unknown.
- 12:28If they also have to simultaneously manage the
- 12:31tedious defined boundaries of that unknown space, their
- 12:35creative capacity is drained. But if the AI handles the rigid
- 12:40cognitive defined rules, it creates a stable foundation.
- 12:43Oh, I see. The machine's creativity
- 12:45actually thrives when it's navigating a highly structured
- 12:48environment, which then frees the human to originate pure
- 12:51novelty without worrying about the foundational math.
- 12:54It's like playing in a sandbox. OK, tell me more.
- 12:57Like if you just dump sand on a flat driveway, it washes away.
- 13:01You can't build the complex structure because the
- 13:03environment is too unstable. True, but in this collaboration
- 13:06the AI acts as the rigid, highly structured wooden borders of the
- 13:10sandbox. It provides the absolute defined
- 13:13parameters and because those boundaries are so strong and
- 13:16computationally reliable. It gives the human a safe
- 13:19environment. Exactly.
- 13:20It gives you the perfect safe space to get inside and build a
- 13:23completely unprecedented wild novel.
- 13:26That sandbox framework perfectly illustrates the data.
- 13:30The existential challenge for creative industries isn't
- 13:32choosing between machine or human creativity.
- 13:35It's a false binary. The real challenge is
- 13:38intentionally building these structured environments.
- 13:40Delegate the heavy lifting of defined rules to the AI so the
- 13:44human is freed up to manage high ambiguity and originate novelty.
- 13:48It's surprisingly optimistic, honestly.
- 13:51The tech isn't flattening our potential, it's providing the
- 13:54structural support for us to be more creative than ever.
- 13:57But I guess how this applies to you depends heavily on your
- 13:59specific profession. To make this entirely concrete,
- 14:02the study measures something called rule alignment.
- 14:05Rule alignment. Yes.
- 14:08It analyzes how closely a worker's internal cognitive
- 14:11thinking matches their external behavioral actions on the job.
- 14:14Essentially, does the way you think smoothly aligned with the
- 14:17way you are required to act? Precisely.
- 14:20And the data segments creative jobs into two very different
- 14:23categories. First, you have highly lined
- 14:25jobs. Think communication
- 14:27professionals, classical musicians and journalists.
- 14:30The paper notes journalists have a near perfect alignment score
- 14:33of .02. So for a journalist, the
- 14:36cognitive rules of investigating, structuring a
- 14:38narrative, they perfectly match the behavioral rules of typing
- 14:42the article and publishing. Their mental processing and
- 14:45physical outputs flow in the exact same direction without
- 14:48resistance. The alignment is incredibly
- 14:50synchronized. But then on the other end of the
- 14:52spectrum. You have highly misaligned jobs.
- 14:56This includes web designers, interior designers, and most
- 14:59notably fashion designers. Fashion designers have a huge
- 15:02misalignment gap, right? Like .26.
- 15:04Yes, which is a massive jump from the journalist .02.
- 15:08It sounds like driving a high performance sports car
- 15:10representing those wild, uncodified ideas straight into a
- 15:14dense, gridlocked city. This is a.
- 15:15Great visual. Right.
- 15:17The sports car is the cognitive freedom of the fashion designer.
- 15:20They're pulling tacit inspiration from a mood or an
- 15:23abstract feeling, but the gridlock city represents the
- 15:26unyielding behavioral constraints they have to
- 15:28navigate, the strict physics of how fabric drapes the tensile
- 15:32strength of thread manufacturing supply chains.
- 15:35That analogy captures the daily reality perfectly.
- 15:38You have a free flowing ambiguous mental space colliding
- 15:42violently with a rigid physical reality Sounds.
- 15:45Stressful. It is, and this raises an
- 15:47important question. Is that misalignment of flaw in
- 15:51how the job is designed? I would assume yes.
- 15:54The study argues that it is absolutely not a flaw.
- 15:56It is in fact their greatest advantage.
- 15:58That friction between how they think and how they must act
- 16:02creates what Walkaway calls productive coordination tension.
- 16:06Productive coordination tension. Wow, so the friction is actually
- 16:09the engine of the work. The tension is the very source
- 16:12of their creative problem solving.
- 16:14They're forced to take a highly novel, entirely tacit idea and
- 16:18brutally force it into a defined physical reality.
- 16:21And a machine does not feel tension.
- 16:23No, AI doesn't get frustrated trying to make a stiff fabric
- 16:26express a soft melancholic emotion.
- 16:28It simply executes code and stops when it hits an error.
- 16:32But the human ability to navigate that severe
- 16:34misalignment, to live inside the frustration of the gridlock,
- 16:37manage the clutch and the brake, and force a creative solution
- 16:41into existence that is entirely irreplaceable.
- 16:44Which brings us back to the friction we talked about the
- 16:46very beginning. By automating the routine
- 16:48aligned portions of creativity, AI is going to force human
- 16:52workers to become necessarily more human.
- 16:54We'll have to rely heavily on our physical intuition.
- 16:57And our ability to read context, our lived experience navigating
- 17:01tension because those are the only domains the machine cannot
- 17:05touch. Gen.
- 17:06AI is really a Co evolutionary partner that will absorb the
- 17:09defined rules, leaving us to specialize in the tacit and the
- 17:12novel. So look at your own day-to-day
- 17:15work. Do you have a journalist job?
- 17:17A role where your internal thinking and external outputs
- 17:20are perfectly aligned. If so, you're likely looking at
- 17:23a future where AI can cleanly assist with or even automate
- 17:27large portions of that flow. But if you have a fashion
- 17:30designer job. Exactly.
- 17:31If your daily professional life is defined by the messy friction
- 17:34between strict rules and wild, uncodified ideas, that very
- 17:38tension is what secures your value.
- 17:40That is the triple triple configuration in action.
- 17:43It's all about that tacit knowledge.
- 17:45Which leaves us with one vital question for you to Mull over.
- 17:50If tacit knowledge, those deep experiential gut feelings you've
- 17:53developed over years of practice, is the ultimate
- 17:56boundary keeping humans relevant, what is the uncodified
- 18:00intuition that you rely on every single day?
- 18:03It's a great question to ask yourself.
- 18:04Think about the complex, ambiguous problem solving you do
- 18:07at work that you couldn't possibly explain to a new
- 18:09trainee if you tried. And more importantly, how do you
- 18:12make sure you never, ever write it down into a defined rule that
- 18:16a machine could learn? Thank you for joining us on this
- 18:19deep dive. Keep protecting your tacit
- 18:21knowledge and we'll catch you next time.