“Partnership between human labor and AI” - Dall-E 2 & Jacob Brown
Labor partnership is the term I’ve started using to describe a human and an artificial thing doing some kind of work together as a team.
Right now, everyone is excited about creative AI partnerships like those that are enabled by Stable Diffusion and Dall-E 2 image generators. But that’s not the complete concept. It might be that a machine helps someone in a factory traverse a line efficiently, or a robot works in tandem with a human to enable the duo to do both fine-detail work and heavy lifting. Automating that kind of work may seem old, but it’s as cutting edge as anything.
The key idea is that a machine or—increasingly, a machine intelligence—will be able to share the load and contribute to a human’s productive work: their labor.
This is a pretty important thing to get right. Amazon is introducing robots to their warehouses that make picking and packing more efficient. It doesn’t take much to imagine that these robots will be more and more capable of replacing the humans entirely. It’s certainly obvious to everyone who does the picking and packing currently.
Jasper is building out an application that is increasingly capable of writing marketing copy to assist marketers and copy writers everywhere fill the blank page. It’s obvious that soon Jasper will be capable of replacing the humans entirely — first for some kinds of writing, and then for all kinds.
But is that the future we want to build?
This is the question inherent in labor partnership. Instead of looking at labor that a human does today and asking: “How can we build technology to replace the human?” Maybe we should be asking: “How can we build technology to make this human’s labor the most rewarding it can be?”
I suspect that which question we ask will determine the course of our civilizations. We must make the choice to build technology that will partner with us — on everything from making art to legal decisions to war — or we will build a dystopia that will kill the human spirit.
Why is it the goal to make work rewarding for a human? Why not just ask what increases productivity? We now understand that making sure people are as rewarded as possible is what makes people productive.
The Industrial Revolution was an exercise in labor extraction. There was never an effort to understand what makes people productive, maybe in part because labor at the time was mostly mechanical in nature. How much do you really need to understand about what makes wrench-wielding rewarding? Today we know a lot more now than we did a hundred years ago about what makes people’s work rewarding. Dan Pink has taken a hard look at the research and summarized what makes work fulfilling: autonomy, mastery, and purpose. Fulfilled people are productive people.
With better technology and a better understanding of what helps a populace thrive, we now have an opportunity to do better for ourselves and get a better outcome from it.
A few weeks ago I was at Web Summit in Portugal, and on stage Cristina Fonseca from Indico Capital Partners gave a great example of how most people are looking at the future of AI with ambivalence. Cristina works with companies that offer support through call centers, and these companies have tried introducing AI tools that make filling out all the data entry forms much easier and quicker.
But the employees can see that if an AI can already do part of what they’re getting paid to do by the hour currently, then it’s just a short jump to their taking over the next part of their job, and the next. So rather than use the bots to make their jobs noticeably easier by reducing the annoying data-entry, they don’t like to use the tools and prefer to just hammer on the keyboard themselves.
In short, yes, AI really is coming for everyone’s white collar job, and surprise, people don’t like that idea.
It’s tempting to think that there are only two paths forward. Either we have to reject AI in order to keep our jobs (through new laws, for example), or we have to embrace the automated future and go find something new to do with ourselves because the computers will do all the stuff we do for work today.
But there is a third option: partnership.
Most people feel there is something innately human about succeeding at what they do, whether for work or pleasure. The idea that an AI could succeed like a human is met with skepticism. I certainly feel that way about early stage venture investing!
Let’s imagine that we’re right. That there is something about our labor that a machine isn’t going to be able to grasp. Unlike asking ourselves what are the most rewarding parts of our jobs, this is a more practical and answerable question that can get us started. What might it look like to partner with an AI so that it does the stuff we don’t want to do, and leaves us to do the stuff that only we can do?
In the book Thinking Fast and Slow, the Nobel Prize winning psychologist Daniel Khaneman introduces us to two different parts of our brain. He calls them “system one” and “system two”.
System one is the part of our brain that thinks intuitively — we approximate math answers or regurgitate ones we know “by heart.” We form impressions of people, we read facial expressions, and we throw baseballs with accuracy. It also happens without the high calorie energy of studied thought. In other words, it’s easy for us.
System two thinking is effortful by comparison. When you want to recite the alphabet backwards, or work through a logical tumble — these are tasks that require you to activate system two. Which sucks. In fact, it’s so costly that our bodies often react when we engage system two: our stomach turns, we get sweaty or cold, we need to use the bathroom, and so on.
Computers, by contrast, are great at system two thinking. They’re literally built for it.
What parts of your job require system two? What parts require system one? Can you really build a company around the idea of taking the system two bits out of a job? I certainly think so.
The Information recently ran a profile on George Sivulka, the founder of Hebbia, a “search” engine that looks a lot more like Wolfram Alpha. The product is designed to take drudgery and system 2 out of the picture. Here is how George describes it:
I remember my friend in investment banking would spend all day reading this one kind of [Securities and Exchange Commission] document. It’s this 400-page document and you’re just really looking for a few different questions. [Hebbia] saved her like three or four hours…
Hebbia recommends the right answer for you and shows you exactly what you need for work, whether it’s in a really large filing or some academic research paper. I don’t want humans to be replaced. I want them to be empowered.
This reminds me of the times I’ve had to manually go through large legal documents (venue contracts, funding documents) looking for or checking for minor details that are easily missed if I’m not paying attention but can have critical ramifications if missed. For me, that’s not background work — and having an AI do it would make for a huge time and effort save.
What else might be possible?
Imagine going to a doctor’s office and getting a consult with both a doctor and an AI working together. The doctor might ask you questions about symptoms or worries, but ask the AI about your family history. The doctor might then suggest to the computer a couple of different drug ideas, but the AI might suggest that a large number of people with similar genetic backgrounds have a reaction to those drugs in combination, so a different prescription is in order.
Imagine a designer working on a new logo for a client. She starts describing different evocative ideas to her AI copilot, which renders stylistic versions of her descriptions in sequence for the designer to evaluate and react to. Eventually, she finds a thread. She takes the AI output, traces some of the lines, and builds on it. Then she hands her design back to the AI to produce the file types and formats required by the client, freeing her to work in the realm of the creative and not worry about the tools.
It’s trickier to imagine how a job like picking at an Amazon warehouse or working the deep fryer at McDonalds could be enhanced by AI partnership, but I think in the short term they will be. In the long term, these jobs will probably be entirely replaced by intelligent machines, because they’re a type of work that not many people actually want to do. (I recognize that most people want to do their jobs, even if it’s only because they get paid by doing them. So there is nuance to this point that belongs in another essay sometime in the future).
This idea of partnership is not likely to be a perfect solution for the laborer, especially not in the short term. Some jobs will inevitably be replaced or reduced. But choosing to build technology that enhances what people do instead of building technology to replace them is going to be how we stumble towards a future that can ultimately be very rewarding.
Intention and creation have to stay with a human. Labor that requires neither will eventually be handed to machines. And for everything else the technology — even the really advanced generative AI stuff — can become a partner in service of greater human capacity.