Slow-burn AI: When augmentation, not automation, is the real threat
Automation eats your job, but augmentation eats your career
A common meme on the impact of Gen AI on jobs goes something like this:
“AI won’t take your job but someone using AI might.”
You’d be quick to conclude then, that, those who start using AI will win and those who don’t use AI will lose.
But what if those who use AI also lose?
Well… Not all of them, but enough of them to make you sit up and notice.
Automation bad, augmentation good?
The popular narrative around the impact of AI on jobs goes something like this:
AI can lead to automation or augmentation.
Automation eats your job.
Augmentation makes you better at your job.
Hence, avoid automation and embrace augmentation.
This argument is as flawed as it is simplistic.
The argument possibly worked well in a pre-internet world where robotic automation repeatedly substituted factory jobs.
However, with the rise of online platforms, and the networked markets of labor that emerge as a result, automation is only a short term threat to jobs.
A far greater threat now emerges: the long-term commoditization of jobs through augmentation.
Drawing on my extensive work with the ILO’s Future of Work Commission, I lay out the rationale for AI-driven commoditization of work and how that gets accelerated in a world of digital platforms.
This issue right here is Part 1 of a three-part analysis.
Let’s dive right in!
But first… Subscribe here to receive these posts every week.
Unbundling and rebundling of work
Every job is a bundle of tasks.
Some of these tasks require specialization, some don’t, but they still remain part of the bundle as the cost of unbundling and delegating those tasks may be high.
Every new technology wave (including the ongoing rise of Gen AI) attacks this bundle.
New technology may substitute a specific task - for instance, intelligent scheduling tools may fully substitute a task previously performed by a human.
New technology may also complement a specific task - for instance, improvements in AI may provide diagnostic support to doctors and radiologists, enhancing their ability to perform the task.
Automation is technology in substitute mode.
Augmentation is a result of technology working as a complement.
First order thinking suggests that technology-as-substitute (automation) is bad, and that technology-as-complement (augmentation) is good.
However, this reasoning ignores the effects of feedback loops and their compounding effects over time.
But, let’s start at the very beginning…
The chainsaw massacre of jobs
Up until the start of the twentieth century, axe-wielding was a high-skilled job. In order to be any good at logging, you needed to perfect the right angle of the axe swing as well as the grip on the axe, through years of practice, that developed both muscle and muscle memory.
The invention of the chainsaw changed all of that. Low-skilled loggers, who lacked the knowledge or the muscle memory to perform well, could now perform at a much higher level of effectiveness.
One way to think about this is that this chainsaw augmentation helped loggers level-up and get better at their job.
But that’s only half the story. The chainsaw helped low-skill loggers level-up but didn’t benefit high-skilled loggers quite as much.
The real long term effect of the chainsaw was the commoditization of logging as a job.
Since anyone could now be a logger, the job’s ability to command a skill-premium eroded. As the market of potential loggers exploded, wages stabilised and the high-skilled loggers lost the ability to charge a premium. Yes, increase in logging also drove increased demand downstream in the short term but those effects soon stabilized and logging became a commoditized job.
When technology augments skilled work and enables historically low-skilled workers to perform at par with historically high-skilled workers, such augmentation makes workers more substitutable and eventually commoditizes the job.
Generative chainsaw massacre
But, surely, this doesn’t apply to knowledge work… or does it?
A recent much-cited HBS study on the effects of Generative AI augmentation on BCG consultants tested performance across a sample of BCG consultants before and after AI augmentation (impact of AI augmentation on performance).
The study reveals two interesting findings:
Consultants who tested in the lower half of the group when not using AI increased the quality of their outputs by 43% went using AI. Consultants who tested in the top half of the group when not using AI increased the quality of their outputs by only 17% with AI.
Without AI, the score gap between the top and bottom performers was 20%+ (4.5 vs 5.2). With AI augmentation, the score gap had collapsed to 4%.
This is an early study, possibly with many caveats, but it demonstrates one important point.
AI augmentation makes high-skilled knowledge workers relatively more substitutable by lowering the skill barrier to achieve the same performance.
Augmentation improves productivity and output but does so to a greater extent for those with lower skills than for those with higher skills. When this plays out, the worker becomes more substitutable as the skill becomes more commoditized.
Similar studies have been carried out with customer service agents, law students, and writers. All these studies demonstrate similar results. What happened with the chainsaw is increasingly possible across a lot of knowledge work in the age of Gen AI.
AI + platforms = Accelerated substitutability
AI augmentation commoditizes knowledge workers making them more substitutable.
Skill commoditization is the first piece in the puzzle here.
But these effects are particularly amplified in a world of online platforms. Platforms create markets over which skills can be traded.
Centralized market-making accelerates worker substitutability when augmented workers have largely undifferentiated skills.
How exactly does this play out?
London cabbies may have a point of view…
The Uber for X is back… and it wants your job
Cabbies and taxi drivers have traditionally had a significant knowledge advantage in their work - knowledge of the city’s maps and navigation, a mental model they constantly update with the latest information as they go about their job.
This was a significant knowledge-based barrier to entry.
That is, until Google Maps came along and commodified this knowledge making it readily accessible to anyone. And until Uber came along and created a ‘wrapper’ over Google Maps, making this knowledge component freely accessible to anyone.
There were four effects that played out:
1. Commoditization of skill
First, London cabbies lost their competitive advantage. Any maps-augmented amateur could compete with their city navigation skills even if they couldn’t immediately compete with their driving skills.
2. Supply expansion erodes skill premium
Second, as ride-hailing apps integrated maps with the ability to charge for a ride, the market of potential drivers expanded, leading to greater competition and an overall ‘price war’, which eroded the ability of high-skilled drivers to charge a premium.
3. Centralized market-making drives faster commoditization
Third, and most important, ride-hailing marketplaces like Uber centralized this growing market and absorbed this ‘price war’ into their algorithm, effectively driving down (and standardizing) the cost of the ride and the payout to the driver. Hailing a cab off the road - the last remaining information advantage where a cabbie could charge a premium on account of being at the right place at the right time - was lost.
4. Centralized ‘job discovery’ reduces negotiating power
Finally, the advent of Uber changed the mechanics of ‘job discovery’, or in this case the mechanics of finding your next ride. Cabbies could either be assigned jobs by the algorithm and compete with all other 5-star-rated, maps-augmented amateur drivers or they could stay out of the system and miss out on the demand coming in through Uber. Cabbies lost their negotiation power leading to a variety of effects - they had to accept rides without knowing the destination, they had to adhere to acceptance and cancellation rate metrics. They had lost not only the ability to set the price but also the agency to accept or reject work opportunities based on whether it made commercial sense.
These four effects, together, drove commoditization of the cabbie’s ‘job’, effectively leading to value migration away from the drivers to market-makers like Uber.
Uber, for its own part, hasn’t been short on efforts to move the market from commoditization to substitution, investing heavily in self-driving cars. While there are multiple factors holding back a future of self-driving cars, there is no denying the role that today’s drivers play in potentially training self-driving technologies every time they set out on a maps-augmented drive.
Skill absorption compounds substitutability
With skill commoditization and centralized market-making out of the way, we now look at the third key factor - skill absorption.
Chainsaws made loggers more substitutable and logging more commoditized.
But this was largely a one-time effect. The wage market for logging largely stabilized after the effect had played out.
Things play out a little differently with AI.
AI compounds commoditization by continuous absorption of skills.
The more successful the AI is at augmentation, the more effective it gets at future augmentation.
Let’s take the example of highly specialized work performed by doctors and radiologists - clinical diagnosis.
As I explain in The Lego Blocks of Healthcare, two simultaneous shifts - increasing data interoperability and improvements in AI - are simultaneously commoditizing this highly specialized skill set.
Advances in AI and ML have commoditized prediction - the ability to anticipate future outcomes based on available data. Improvements in computational speed (using superior GPUs and TPUs) as well as data processing and analysis techniques over the past decade have worked together to bring down the cost of predictions based on machine learning, thereby commoditizing them.
Greater investment and regulatory push towards data interoperability increases the availability and accessibility of data, making predictions more accurate as well as more applicable across a wider scope of diseases.
This drives greater commoditization of a highly specialized skill-set:
The commoditization of predictions reduces the cost of medical diagnosis, which can now be increasingly performed by machines. This, in turn, makes it feasible to perform diagnosis more frequently.
All of this sounds great. Doctors can now spend less time in diagnosis, allowing them to serve many more cases more effectively.
As AI augmentation plays out, doctors are constantly training the model driving more of their skill into the model.
In China, PingAn Good Doctor started out as a tele-health platform with AI-augmented doctors treating patients over the platform. Several years of model training later, the AI model now beats human doctors at many tasks.
The AI has moved on from augmentation to substitution.
This brings in the third piece of the puzzle: Skill absorption.
AI commodifies and absorbs proprietary knowledge and learning advantages that workers (e.g. doctors) may have developed in a specific industry.
With large language models (LLMs), the scope of knowledge that can be commodified and absorbed into a model increases further, increasing the potential for similar commoditization to play out across a larger range of jobs.
Much like Google Maps ‘absorbs’ a cabbie’s navigation skills and productizes it to enable anyone to navigate effectively, LLMs ‘absorb’ knowledge, potentially enabling ‘lower-skilled’ or ‘less-informed’ actors to perform tasks that were previously handled by ‘highly-skilled’ actors.
There are three effects that play out with skill absorption:
1. The cycle of continuous commoditization
Skill absorption into AI constantly drives greater commoditization. Unlike the chainsaw example where skill commoditization was largely a one-time event, an ever-improving AI model, trained by a larger base of users, drives continuous commoditization of the skill.
Successful augmentation expands the training data set not just through driving greater usage among workers but by fundamentally expanding the overall base of workers training the models. A feedback loop is set in motion here where greater productization and democratization of the knowledge starts accelerating commoditization.
As we note here, improved LLM augmentation lowers barriers to performance and expands the base of workers who can perform a particular task when augmented. This, in turn, generates larger scale and scope of data, helping fine-tune the model further and improve augmentation specific to that task.
This flywheel demonstrates the inevitability of commoditization of knowledge work through AI augmentation when:
LLM capabilities constantly lower barriers to performance and expand the base of workers that can perform a specific task (thereby increasing competition among them), and
Model training constantly absorbs specialized skills required to perform the task.
The impact of Google Maps on London Cabbies now translates into the impact of LLMs on a much broader scope of knowledge work.
Such continuous commoditization may constantly reduce barriers to entry and performance as doctors with ever-lower knowledge and experience may take on roles previously accessible to a knowledgable few.
Continuous skill absorption drives continuous commoditization.
2. Centralized market-making erodes the judgment advantage
A common argument in favour of augmentation is the following. When diagnosis gets commoditized, judgment (about the patient condition and the appropriate intervention to be followed) becomes more valuable.
According to this logic, doctors can now focus less on the rote problem of diagnosis and more on performing judgment.
But centralized market making again plays spoilsport here.
If doctors and patients interact over an online platform (as in the case of PingAn Good Doctor), the more data the platform can capture across the doctor-patient interaction, the better the model becomes at absorbing not just the diagnosis but also the judgment (based on an ever-growing training set of real-life doctor-patient interactions).
This doesn’t apply to tele health alone. Patient interactions at a doctor’s office could also be digitized. Essentially, greater digitization of the end-to-end doctor-patient interaction can further drive doctor commoditization by eventually absorbing the judgment as well. PingAn’s AskBob (also mentioned further up) is one example of how this plays out.
This is an extension of the London Cabbie meets Uber argument. Uber isn’t merely a wrapper on top of Google Maps, it also extends to cover the whole cab journey, tracking start and end points as well as route taken, essentially absorbing all price-setting judgment away from the cabbie and into the platform.
Greater capture of the end-to-end interaction absorbs not just the immediate skill but also its complements.
3. Skill absorption reduces demand, eroding pricing power further
Skill absorption into machines has a larger effect as well.
It can fundamentally reduce demand across the market.
With GPS, a traveler travelling to a new city may now choose to self-drive at their destination (instead of getting a cab from the airport).
Even in healthcare, greater skill absorption into AI may enable continuous assessment and diagnosis, performed independent of traditional diagnostic labs, reducing the occurrence of episodes, which may further reduce demand for diagnoses at labs and open up participation from more informal caregivers, now augmented by AI. These factors would together reduce the ability of traditional diagnostic labs to differentiate and capture value.
Over a longer period, greater skill absorption in the machine lowers overall demand for the human skill.
Early days yet…
Commoditization is inevitable when augmentation is at play.
There are two exceptions though.
First, high-skilled workers may figure out entirely new ways to use AI and move into entirely new skills when augmented by AI.
Second, AI itself may evolve in ways that specifically benefit high-skilled workers.
These are early days with Gen AI. We don’t yet know if one of these factors will play out. But if the history of technology-based augmentation is anything to go by, commoditization and increasing substitutability is not just inevitable, it will also accelerate faster when model training meets centralized market-making.
This was part 1 of a three part commentary on the future of work in the age of Gen AI.
Part 2 - Gen AI in the era of Uber drivers - looks at the impact of Gen AI on the vast majority of connected knowledge workers.
Part 3 - Gen AI in the era of Taylor Swift - looks at the impact of Gen AI on a small minority of creators who stand to benefit most from it.
If you liked this post, throw in a comment or a like.
And stay tuned for more!