AI and the Future of Work
AI automation yields productivity, but also disruption & displacement
Introduction
As AI has become more powerful - leveraging more data, improving how it learns, becoming more accurate, solving more problems - its has transitioned from a research curiosity into the most powerful technology ever conceived. With great power, comes great fears around AI. Many fear that AI will harm society in various ways.
One primary concern is that AI will displace jobs and replace work. AI will become so much better than us, we will be useless. This fear is not unfounded; AI has similar potential to automate cognition as the Industrial Revolution automated uses of energy and manual labor. Industrialization created the society and comfort that we enjoy today, but along the way obsoleted existing craft industries and disrupted workers’ livelihoods.
AI is no less potent a disruptive force. What should we look out for in AI? How can it help us do our jobs?
Principles of Automating Cognitive Tasks
What AI does is automate cognitive tasks. Indeed, instead of ‘artificial intelligence’ we should call it ‘automated intelligence’ and its role in serving us is ‘augmented cognition.’
AI displaces human cognitive tasks, not necessarily jobs. Waves of prior business automations and IT implementations have automated many tasks that were once manual or managed by humans: You can pump gas, visit an ATM and withdraw money, automatically check out groceries, and buy airline tickets online, all without interacting with a human service worker.
What can be automated by AI technologies that wasn’t automated before? Today’s generative AI can tackle ever more complex cognitive tasks, if given sufficient direction. Any jobs where the job tasks and interactions are well-specified and bounded can be subject to some sort of automation, and those job tasks that are impacted by AI include a component of social interaction or use of language.
More about the tasks we can expect to see automated by AI:
The tasks most amenable to AI automation would be small, well-defined, specific tasks that can be clearly specified.
Both decision-making and generative AI quality, depth, and reliability is good enough for a wide variety and expanding set of tasks.
AI doesn’t have to be good at all things, just the task(s) it’s asked to do. Rarely will any job or task-based AI require AGI.
Task-specific AIs will be developed and trained, fit for purpose: An AI receptionist, an AI medical or legal assistant.
If a job is simply a role involving a set of tasks, then if most of those tasks are automated, that job can be displaced by automation.
For example, consider the legal field. As it interacts with language, yet has heretofore been too complex a field to be ripe for automation, the latest LLMs could be disruptive. We will not see AI replace lawyers anytime soon. However, there are a number of tasks they can replace:
Research: Going beyond search engines, AI can search, summarize and develop textual interpretations of citations, precedents, and background understanding of legal issues.
Reviews: AI can review legal language in contracts or other briefs to call out issues, corrections, and concerns.
Drafting of contracts, terms and legal briefs; using personalized and customized language, adapting ‘boilerplate’ to specific items.
The role of AI in law may be similar to its role in healthcare, software, marketing and other cognitive-heavy fields. AI will not be capable enough to make end-to-end creative decisions by itself, but it can be a productivity-boosting assistant, tutor, sidekick and co-pilot in these efforts.
Fields at Risk
The ‘well-specified and bounded’ jobs that were automated in the last generation or so were factory assembly jobs, where automated tools and robot welders replaced humans doing the same task. With AI that can interact with people via natural language, automation comes to impact roles and jobs where language is paramount.
Retail: Checkout is becoming automated, and orders are taken at a kiosk; we might soon show up at a fast-food location and order our meal in natural language to an AI avatar.
Customer Service: Many menial questions can be answered by tailored AI tools, and an AI never loses accuracy or politeness in response. Specific questions that cannot be answered by AI can be identified and forwarded to human team members. As AI advances, it will automate more and more of this process.
Law: As noted above, with perfect access to databases of laws and law cases, AI can quickly draw on precedent cases, review legal language, and help research, review and draft tightly worded contracts and other legal documents.
Education: While a good teacher’s ability to bond with their students and adjust curriculums as necessary is not easy to replace, AI could step in as tutors. Natural language can answer simple questions, and plugins cover up weaknesses in some subjects. If current AI scores on standardized tests are anything to go by, tutoring could be completely automated in the next few years.
Media and journalism: From screenwriting to writing news stories, we have the challenge that AI can spin up words quickly. But are they the right words? Human writers with a co-pilot may be more productive and higher quality, but the low-cost of AI creates a temptation for displacement of humans even if quality suffers.
The ‘well-specified and bounded’ can apply to tasks and not whole jobs. In the previous examples, the whole task of a lawyer may not be easily replaced by an AI, but several smaller tasks such as researching precedent cases could be automated.
Displaced, not Replaced
The significant importance of the human touch in most fields will go away in the near future. Rather automation of cognitive tasks may replace the 20% to 40% most repetitive and menial part of a professional worker’s job, leaving people with more productive and fulfilling jobs.
Doctors and healthcare: The role of a human doctor cannot be entirely replaced. Patient care, surgeries, physical therapy, checkups, and any other number of tasks require a human eye and physical interaction that AI cannot yet handle. AI can, however, assist with diagnoses, record keeping, prescriptions, and anything that can generally be handled as data or language. More on that here.
Teachers and education: As we discussed in the previous example, teachers cannot be fully replaced. However, AI could prove a useful tool alongside teachers, performing data analysis on student growth and helping to prepare new curriculum based on those circumstances. AI can also be a customized and personalized tutor.
Benefits of AI
Most fields and job types impacted by AI will find greatly increased productivity and accuracy thanks to AI replacing menial tasks and giving human cognitive workers more time to focus on higher level tasks. AI can also work quickly and can operate 24/7.
AI has the potential to replace jobs where AI automates most of the job. But AI doing more of the ‘grunt work’ has the potential to alleviate stress on societal roles like doctors and teachers, and transform professions by making them more creative and fulfilling.
The advance in productivity from AI yields more demand that results in positive demand for jobs on balance, greater than the loss of jobs from AI job displacement.
The productivity benefits of AI are clear. Even the use of ChatGPT alone can be shown to improve writing productivity. In a study of college-educated professionals called “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence”, study subjects were given a “treatment” to sign up for ChatGPT.
The following figure shows the results of the study. Evaluator grading of the group’s performance in incentivized writing tasks show that, with the aid of an LLM to make suggestions and revisions, students became faster and better at writing in many fields.
Conclusion
For now, the rise of LLMs is putting the disruption onus on those industries and jobs that use language, such as customer service, copywriters, marketing, etc. When advanced AI intersects with robotic AI, manual jobs of many types, such as janitors and electrical workers, will be impacted.
The scope of tasks AI can take on will increase as AI advances, and since AI is advancing rapidly, what can be automated in years to come will be greatly expanded on what can be automated today. Job disruption will be an ongoing process, and no job will be left untouched. All will be impacted, some job categories displaced and disrupted.
In the end, AI will be doing most of the cognitive work in our economy, and automating job tasks with AI will be beneficial, acting as a force multiplier of human creativity and thought.
What must we do to benefit from AI and not suffer disruption?
There is one job role that may become common thread to all professional workers: AI manager. Knowing how to get AI to do the various parts of your job effectively may end up becoming the AI-job-shifted version of every profession.
Our future is directly influenced by how we approach and use AI being made today. It’s up to us as a society to determine where to place newly freed labor, and how to value it.