AI, Jobs, and Red Herrings
“The future cannot be predicted, but futures can be invented.” —Dennis Gabor
Ex-Atlantic journalist and Abundance co-author Derek Thompson recently wrote a compelling account of the rollercoaster that is AI’s impact on entry level jobs. His conclusion that, plausibly, yes, AI is reducing work for young Americans, is based on a widely covered study by Eric Brynjolfsson et al, showing correlation between declining hiring rates on one hand, and "substitutability" by AI of certain jobs and tasks on the other.
The question was settled, for me at least, until three days later, when a countering piece by Noah Smith, called into question why AI driven job losses would fall on entry level workers only, vs mid-career or older workers, who are experiencing a significant increase in hiring rates in those same jobs.
This debate encapsulates perfectly the back and forth on the (seemingly) critical question of AI's impact on jobs. Presented through an endless stream of research and opinion from The Financial Times, The Economist, various academic journals, Substack, Medium, banks, frontier AI companies, NGOs, think tanks, consulting firms, research labs, and every Tom, Dick and Harry who has an opinion, researchers and writers seek to articulate, mostly in good faith, what the adoption of Generative AI means for incumbent workers and for workers entering the labor force, today and into the future.
These are consequential goals, especially for actuaries or policy makers trying to calculate the number of people who may claim unemployment in a post-labor future. But for the vast majority of us, and especially those of us working in education, workforce development or social impact broadly, these analyses are a distraction, a red herring that keeps us from the important work of preparing, because they abstract and extrapolate beyond reasonable assumptions, and more importantly, beyond our ability to take action.
Take software engineering jobs as an example. Much has been written on the decline of entry level software engineering jobs, but then hiring bounced back. But did it? And for whom? What will it look like next year, or the year after, and will past trends predict the future? Zooming out, if fewer software engineers are getting hired, is this because of AI, or because of macroeconomic uncertainties? And looking forward, will the ability to make software more easily mean engineering jobs are more commonplace but less well paid, or rarer and more highly paid?
Even if we could describe or predict with a high level of confidence what is happening to software engineering jobs at the aggregate level, the level of heterogeneity at the firm level (differences between companies), the occupational level (differences between jobs within a single occupation), and the industry level (changes to general processes and patterns of consumer demand) mean that the most significant question of all - SO WHAT - in a predictive sense, is impossible to answer, especially as someone navigating their career, or trying to help someone else who is doing so.
Many parents, educators, and career advisors are experiencing this challenge first hand, as they try to advise people of all ages how to plan for an uncertain job market. But if we can’t predict the future, or even describe the present with fidelity, what should we be doing instead?
Rather than spending our individual and collective energy predicting, I propose we spend more time preparing, or as Dennis Gabor put it, inventing.
We recently assembled a roundtable of some of the most thoughtful people I know to discuss the world we should be trying to invent, irrespective of AI’s impact on jobs today or into the future. Contributors straddled the worlds of workforce and education, data and research, and public policy and narrative shift. Blending their perspectives with a number of other conversations I've had lately, three major components emerge that need to be invented for a future that doesn't depend on our ability to predict it:
First, we need better intelligence, not on job losses, but on the specific ways in which businesses are adopting AI, and the skill changes that these are necessitating for employees. Today, this information is fragmented, at best, across hundreds of different sources, and in most cases, totally imperceptible due to corporate secrecy, competitiveness, and intellectual property concerns. Platforms like O*NET, and even modern technology products that ingest job postings, are woefully out of date, and lack the contextualized insight required for educators or learning designers to use effectively. To avoid the mistakes of the last few decades, we need not just data, but a means of turning information into powerful insights for educators, and educational standards, as close to real time as possible.
Second, we need tools and practices, informed by better intelligence, that can help prepare people for new careers at scale. Technologies that help people to figure out their path and purpose, build skills and work experience, and grow their networks and social skills will be vital in responding to any possible future of work. And, while there are certainly promising solutions emerging, from my own time running research and development efforts in this space for the last five years, it is remarkable how small the number and variety of effective solutions is, and how far we are from serving learners in the millions, or billions, that will need support. In my experience, much of this can be attributed to a lack of demand from educational systems, rather than the failures of entrepreneurs.
This leads to the last and most ambitious prescription of all - the need for a groundswell of advocacy to change the paradigm of education away from rote knowledge accumulation, and towards the invigoration of young people with skills, knowledge, and purpose that are resilient to any given future. We need parents, educators, business leaders, and elected officials, to push for a fundamental reimagining of the role and goals of education, from the current industrialized model, towards transcendent skills and aptitudes like curiosity, independence, continuous learning, logic, and discernment; how to think, not what to think.
Ironically, AI may play a role in meeting some or all of these needs. But worrying about AI's impact on jobs and industries, and basing our investment and policy decisions on the latest research, is not only a moving target, it is fundamentally unhelpful until we have the tools to do something about it. To that end, I propose we spend less time obsessing over our future with AI, and more time inventing it.
