Why AI Won’t Take Your Job
Reports of your death are greatly exaggerated.
Background
Lee Kai Fu, author|computer scientist|investor|businessman, recently made headlines in an interview where he stated that 50% of jobs will be displaced within the next 3 years. Lee is an ex-Apple, ex-Microsoft, ex-Google tech giant who did his PhD in AI and has had a string of success in the AI world. His interview was brought up as a discussion point at a recent panel discussion with MBA and postgraduate students at a Singapore university where I’m an adjunct faculty. As the “data guy” on the panel, I was asked to opine on the claim. My answer was that I thought it was preposterous.
AI has obviously been around for some time, with implemented solutions over the past couple of decades. Much of AI has been in the background, taking on the work of automating workflow and processes. (I wrote a piece recently about background vs foreground AI.) AI has been used to filter out unwanted emails; it has been used to improve customer targeting. But until the arrival of Generative AI on the scene, no one was talking about “mass extinction” of the human race or the massive displacement of jobs.
And so I dedicate my 51st weekly article to unpacking 3 prescient points as a counter-argument to why I don’t subscribe to the claim of massive job displacement by AI in the near future.
(I write a weekly series of articles where I call out bad thinking and bad practices in data analytics / data science which you can find here.)
Workflow Re-engineering
First, it’s important to note that AI, or any technology for that matter, won’t directly change your work. Rather, it is the opportunity to re-engineer workflows or process afforded by the new technology that disrupts or displaces work. To that end, I have to assume that Lee Kai Fu wasn’t referring to ALL jobs but rather 50% of knowledge-work jobs that would get displaced. The current state of AI (both traditional and generative) is unlikely to disrupt, let alone displace, manual work as it would require very significant enhancements in AI-powered robotics to do that, and we are still decades from that.
Current estimates have it that knowledge work represents somewhere between 20–30% of all work globally; near 50% in developed economies. So that’s the realm we are talking about here. This narrative of en masse displacement in the knowledge economy plays into the hands of AI vendors and management consultants; however, it does not stand up to scrutiny. How fast can work processes be re-engineered? Typically, it requires that the technology matures to a state where the unit cost declines considerably before we see any meaningful scaled adoption. And then there’s a whole change management and adoption socialisation that takes years to embrace. Gen AI is still at least 2 years out from maturity. We may see some interesting workflow re-engineering in the next 10 years, but 3 years is simply too ambitious.
Problem-Defining vs Problem-Solving
I read a fabulous article on Medium recently, titled “A theory of intelligence that denies teleological purpose” by a writer using the handle “From Narrow to General AI”. The article makes a compelling case for why we are not on track to achieving the “holy grail” of Artificial General Intelligence. One of the points raised resonated with me. The author hits the nail on the head when they say that humans don’t just engage in problem-solving, but a big chunk of our cognitive processes is dedicated to problem-defining. We go about our daily lives finding problems. AI is not engineered for that; AI is great at problem-solving. Problems need to be well-defined so that AI can go about working out an optimal set of possible solutions, subject to constraints of course.
The realm of knowledge work involves problem-defining rather than problem-solving. When confronted with situational ambiguity or equivocality, (some) humans possess the ability to construct good hypotheses to move the discussion forward. Problem-defining is closely associated with having a strong competency in sensemaking. I wrote a piece about data sensemaking last year that makes a case for the criticality of this competency in a knowledge economy, while also acknowledging that it is critically lacking in the knowledge workforce. Nonetheless, having poor sensemaking (some humans) is still better than having none at all (AI).
Value Creation vs Value Preservation
Is your job about value-creation or does your job exist to reduce friction, and thus value-preserving? There are many engaged in knowledge work whose job is mostly to “lubricate” those friction points in the workflow. For example, if your job is to summarise the daily news into bite size consumption for senior management or clients, then your job is about value preservation. AI will most definitely displace your job at some point. But if your job is about synthesising seemingly separate and independent strands of news into a unique comprehensive informational whole, then you are creating new value, and it’s very hard for AI to displace you.
While we have seen some graphic designing and copywriting jobs displaced by Gen AI, it usually impacts the bottom 25%, and particularly for those whose work is to churn out standard non-unique fare based on a set of creative templates. These kinds of works are obviously value-preserving rather than value-creating. The initial fear that Gen AI will displace writers and creatives has been very much overstated. Globally, we see that readers shun AI-generated books, although they would be happy to pay less for an AI-generated textbook (i.e. value-preserving). A good human writer brings a unique perspective to any content, creating connections that are non-obvious and yet meaningful. Even though it can be about the same topic, we discover something new with each author.
Conclusion
While AI will definitely displace some jobs, it won’t replace most jobs. Jobs will progressively get disrupted through process re-engineering as organisations find cost effective ways to incorporate useful AI capabilities; this will happen over an extended period of time and not overnight as doomsayers would like to believe. As the knowledge economy matures, competition would push the actors to be better adapt at problem-defining and data sensemaking, which leads naturally to more resilience against AI displacement. Futurists are going to be sorely disappointed to find that the world of work will still look pretty similar in the next 3 years.
