GenerativeAI and the Great Jobs Unbundling

One of the best things to come out of the AI Revolution might be that we realise how much more work we do than is in our job descriptions, and start recognising each other for it. 

‍

In recent weeks I have grown frustrated at all the uncritical discussion of which jobs will be automated out of existence first: lawyers, software engineers and admin support staff alternately appear at or near the top of the lists. This is based on the ability of AI systems to automate an ever increasing array of outputs at near, or beyond, human levels of performance. Seems reasonable that it will spell the end of our jobs, right? Wrong. 

‍

Once again we are falling into a trap of valuing only that which we can measure and optimising systems to game the metrics. A fallacy so well established that Harvard Business Review devotes pages to it, even if the supposed originator Paul Drucker appears not to have said it. Absolutely, GPT4 can scan and summarise a long document more quickly than a barrister; AI EdTech platforms can give personalised feedback to each student and Copilot can write code, but that doesn’t mean they are doing everything a lawyer, teacher or software engineer does. 

‍

To prove it, just ask yourself how much you enjoyed your last conversation with an automated phone system? 

‍

Amazon recently found this out to their detriment at the fully automated Amazon Fresh stores. Removing the need for cashiers was supposed to save money on staffing. However, they discovered that staff on the shop floor were also doubling up as security, reducing the incidence of shoplifting. Without them, Amazon had to employ people to reduce theft - a service other shops get implicitly for free. The story is the same with helping direct shoppers around the store and providing help and a friendly face for more vulnerable customers. The smart shop technology was excellent for a range of functions, but missed out on the other implicit roles that shop assistants play. 

‍

We can ask the same question of a range of other roles.

‍

• Lawyers - In addition to interpretation of large amounts of text (automatable) provide guidance to clients at a time of need, empathetic discussions of the risks involved in various courses of action and, perhaps most significantly, implicit insurance in the form of professional liability for incorrect advice. 
• Teachers - In addition to content provision and tuition of students (perhaps, automatable) are responsible for classroom control, spotting early warning signs of children in need, childcare whilst parents are at work, being role models, nurturing children through their social development and much, much more
• Software Engineers - Write code (probably with the help of AI tools) but also negotiate interfaces to other teams, patiently explain code to product managers, push back on requirements where the person writing them hasn’t understood the full implications and provide watercooler chat, xkcd comics and personal support to their coworkers. 

‍

For all of these roles the employees who excel are doing so much more than the tasks that are being automated - and the functions are most valuable when combined as a bundle. 

‍

A thought experiment on Twitter from Nate Silver recently asked if a robot could (or ever would be able to) play poker at a table of humans. They would need to: 

• Handle poker chips
• Lift up its cards to read them without revealing to other players
• Visually recognize action without verbal cues (e.g. Player X bet $200)

‍

The bigger question is whether I would ever want to invite one to a poker night with my friends. Fulfilling the functions of the game is a key criteria, but so is the ability to take part in table chat, pause for a minute to refill drinks and to explain what’s going on to someone playing for the first time. We must not confuse the core function of a job, or the one on the job title, with the totality of the work. 

‍

Clearly, there will be tasks within a job that we will all spend less time doing going forward. Some of these tasks even currently take up a large amount of professional time. But let's not confuse a tool which makes certain functions more efficient, or effective, with someone’s total contribution to a job and workplace. If we do, we risk making mistakes that have already been made, and improving what we can measure at the cost of much that we can’t. 

‍