AI and the future of work - repost from Feb 2024

In a world of AI, we’ll be paid for what we know vs. what we do.

It’s not news AI is reshaping how some people work today with more changes coming quickly. There’s been a lot written about it, but I find even the thoughtful analyses (here) are not very insightful. The authors don’t know enough about the technology to appreciate how it will evolve, and therefore how it will change how we work. In this post I want to use my understanding to speculate about the coming changes to work.

GenAI levels up

Today GenAI is largely limited to “word work,” retrieving information from textual documents and creating new texts. However, it’s quickly adding capabilities:

·??????? Math work. LLMs are famously bad at math, but conversational overlays to data systems are development priorities. In the near future GenAI will also affect work with numbers, a bigger slice of the professional workforce.

·??????? Taking actions. Today GenAI has limited ability to take actions, but that will change soon. Demos are showing how GenAI can write its own program to create an Outlook appointment or take actions on 3rd party websites. These agents currently require a human to approve the AI’s plan and action before it’s final. Look for GenAI to trigger more actions soon.

·??????? Vision. Today, when most people think of AI and vision, they think of creating novel images (eg: Dall-E) and video deep fakes. I will talk about deepfakes next week in a post on what I’m scared of in the coming US election. Beyond these, AI is getting much better at making sense of images and video. Vision is important for AI to function in the real world.

·??????? The real world. As GenAI gets more reliable I expect we’ll start to see it reaching out to the computers in the off-line world, think robots, autonomous vehicles, drones and scariest of all advanced weapons. These are the scenarios that warrant all of us having a better understanding of what’s going on so we can be smart about risk and opportunity.

Winners and losers are upside down.

GenAI is unlike prior bursts of automation; it disproportionately affects workers at the top of the pecking order. For professionals and other knowledge workers, GenAI is automating out rote tasks, letting them spend their time on what’s interesting and meaningful to them. Studies report higher productivity and greater job satisfaction. Yes, GenAI means employers will need fewer people to get the same amount accomplished, but I’m not too worried because it will mostly be displacing well-educated people, whose skills can be redeployed to solving new problems. Who wouldn’t prefer a world with fewer accountants and more teachers? Fewer grant managers and more aid workers? Fewer lawyers and more…anything else?

The Apprentice’s Problem.

Young professionals, digital natives who can’t remember a time before the internet, seem best positioned to bring GenAI into work. I know millennials who revel in figuring out how to have ChatGPT do the uninteresting parts of their job. However, the generation coming after them may not be so fortunate. In some fields GenAI already has the ability to create a credible rough draft of a client deliverable. I’m pretty sure those firms will continue to need a senior lawyer, ad exec, management consultant, etc., to add the contextual insights which take it from a draft to a winning final product. GenAI has automated out the work that pays for the apprentices while they learn the craft of their field. I wonder how we’re going to grow the next crops of senior folks when we don’t need so many junior ones, but that is a problem we can solve.

GenAI is changing word work first.

Today GenAI is primarily affecting people who work with words. GenAI is becoming a standard part of doing work for professionals who create, edit and otherwise manage textual information. GenAI is becoming the go-to tool to draft written content in marketing, consulting, and HR. It is supplanting some journalism and screenwriters feel threatened.

A good example of the disruption is the practice of law. Pre-trial discovery has changed from laborious copying, sorting, and cataloging documents into an IT/AI exercise. E-discovery software is a $5B business, growing 13% annually. It uses AI to cull through mountains of documents in minutes. Managing e-discovery is a new high-status role in law firms. No one misses late nights crawling through boxes of documents except the managers who used it to cover the salaries of the junior associates. Now the big online legal research providers are all-in on AI-assisted legal research. (WestLaw, LexisNexis) Given law firm billing rates, the savings are likely to be huge. Everyone has their favorite lawyer joke, so they don’t get much sympathy. The fact so many people with legal training have gone on to lead functions and organizations beyond the law suggests their critical thinking skills will be useful in new roles and they won’t suffer too much.

GenAI will handle data next.

It’s counterintuitive that GenAI would be better at reasoning over text than numbers and other structured data. If you ask ChatGPT to read an annual report it will do better at understanding the commentary than the tabular data and may choke on the charts. However, smart people are working hard on enabling conversational interfaces with data. Soon, a VP of sales will ask her sales tracking system to draft a summary of the quarter with charts and text including trends and anomalies. Today the process of pulling the different information together and formatting it can take days, but soon will take minutes. The important point is the machine will only create a draft. Our VP and her staff will need to think and ask questions to dig into the results. The rote work will be done so she and her staff will spend their time building insight vs. building charts.

Going one further, soon – with permission - AI will listen in on a doctor-patient conversations and extract the information relevant to the patient’s record. It will draft prescriptions and orders. The physician will need to approve the AI’s suggested changes, but instead of spending the precious 9-12 minutes looking at a computer screen, she will have been engaging with her patient. The uber prize is to reduce physician burnout by letting doctors focus on patients and having AI handle the mountain of paperwork that comes with every patient.

Vision ins and outs.

AI’s ability to make sense of inputs from cameras is improving fast. Smartphone apps for the blind have been a boon to AI development since real world use gives the models workouts they need to get better. (See post 1 for why AI models need use to improve.) The ability to take in signal from the real world is key for AI development, but most of us are most concerned with the visual images AI is putting out, specifically deepfakes.

The real world is harder for AI:

AI is progressing more slowly for people who do their work with tangible objects vs. the digital realm.

·??????? Self-driving cars have gotten huge investment, but the launch dates keep sliding out.

·??????? Robots are making big gains in the controlled environments of factories, but have a harder time dealing with all the unexpected happenings of the real world.

·??????? Farmers get guidance from AI on what, where, and when to plant. Drones monitor crops and even target pesticides, but as far as I know the laborious process of moving up and down rows in a field can’t yet be outsourced to AI.

AI is coming into the real world, but slowly. Today an astonishing range of mechanical devices are connected to the internet, called the Internet of Things. Most connections are sensors sending information back to a central controller, but some can take action. Theoretically they are all within reach of control by GenAI. This is the arena in which we need to go slow and be thoughtful. Governments in the US and Europe are taking first steps. A great deal is being written about the risks of AI, but I fear most readers - and many writers – don’t know enough to form meaningful opinions. My goal in writing this newsletter is to fill that gap. Please let me know where I’m succeeding and where I’m falling short.