The headlines still talk about AI job Armageddon; widespread job loss through AI and automation because AI is faster and better at everything humans can do. We don’t think this is correct and now, thankfully, amongst researchers and AI early adopters, the rhetoric is shifting: jobs need to be designed to allow people and machines to both learn and co-work.
We were early in this view: our 2016 Machine Employee research showed that most work would be more valuable if humans were augmented rather than replaced. Our analysis of O*net (the US labor department’s database) data suggested that replacement would be around 10% of jobs, while augmentation would apply in around 40% of jobs. The most important new AI applications would be in the fields of computer vision, natural language and emotional AI. Basically, if we evolved to do it, it would be the next thing for computers to be able to do.
The academic research is beginning to confirm this. MIT guru Erik Brynjolfsson, author of Machine, Platform, Crowd, has done more work on the impact of machine learning on jobs. In a paper describing the research, the authors come up with a measure called Suitability for Machine Learning, SML, which they apply across the same O*net database. They came up with similar findings as us: most occupations in most industries have at least some tasks that are SML while few (if any) have all tasks that are fully substitutable. Interestingly, if machine learning is “unleashed” it will require that there is significant redesign of the task content of jobs as SML and non-SML tasks within jobs are unbundled and rebundled. Machine learning is pervasive but unequal and highly variable.
Suboptimal building of jobs can block potential productivity gains from AI. Theoretically, productivity gains could be locked up if SML and non-SML tasks are bundled, because this could prevent specialization. This is because workers will always choose to do something that AI cannot. If the capital cost of AI is zero AND firms give workers jobs where tasks that are SML and non-SML are preset, there’s an opportunity cost of foregone potential non-SML labor.
That is, if there is no incentive or opportunity for humans to be creative and figure out what the AI can’t do as well, then humans can’t use the AI to its best economic advantage.
This means the focus for entrepreneurs needs to shift from automation to job resign or they risk stalling on the uptake of AI because of inflexible task bundling.
At Sonder Scheme, we help companies make better AI, faster. We offer offer services, training and tools to help you lead in the age of AI using the best human-centered design techniques combined with the unique requirements of AI.