Thoughts on AI
Tech, productivity, and preparing for an artificially intelligent future
Everyone’s talking about AI. Time to throw my hat in the ring! (And no, ChatGPT did not write this post, though I did ask it to help me create the cover image with Midjourney.)
OpenAI officially released ChatGPT to the public in November of last year. By January, it had already amassed over 100 million monthly active users, making it the fastest growing consumer application of all time (nearly 5x faster than TikTok, the previous record holder, and 10x faster than Instagram).
Companies have been using AI & machine learning for years to automate processes and personalize customer experiences, but working with AI/ML has always required serious technical skills… until now. Large language models (LLMs) like OpenAI’s GPT and Google’s LaMDA changed the game overnight.1 Now, humans can interact with AI using intuitive, natural language, and the results have been mind-blowing.
People and businesses are rightfully excited about “generative AI” tools built on LLMs. We can now use natural language instructions to create excellent copy, beautiful designs, full architectural plans, working code, and so much more. The possibilities are pretty freaking cool, not to mention more efficient than alternatives. And there are countless ways to use LLMs and AI that nobody’s even thought of yet.
It’s clear that AI will change how we work (and play), but to what extent? And what will that mean for the economy, workforce, and society?
From everything I’ve read over the past few months, the mainstream narrative seems to be that AI is amazing and will transform many jobs, especially boring/highly manual ones. By doing so, AI will unlock a new age of productivity growth around the world. Even if AI eliminates some jobs, the productivity gains and new opportunities that emerge will offset any pain caused by worker displacement.
This narrative is largely based on the results of past technological revolutions. Historically, most tech revolutions have followed a similar pattern. After an initial discovery is made, decades pass before applications become commercially viable on a large scale. For example, Faraday invented the electric dynamo in 1831, but it was 50 years before Edison fired up the first electric power station in New York. Thirty-something years later, most factories in the US were running on electric power.
Once new solutions are commercially viable, new applications (e.g., new automated manufacturing techniques) radically transform industries, displacing many workers in the process. However, the new technologies ultimately boost quality of life, increase economic productivity, and create new opportunities (even industries) in the process, thereby increasing the size of the proverbial pie and softening the blow for most displaced workers.
Many experts believe this will happen with AI, too. Economist Paul Krugman, for example, argues (NYT, $)2 that AI won’t have a meaningful effect on productivity gains for a while because it took a while for earlier tech advancements like electric power and IT to boost productivity (seems like a pretty weak argument for a Nobel Prize winning economist if you ask me…).3
New research from the smart folks at Goldman Sachs supports the “AI is similar to electricity and the internet” argument as well. Their research argues that AI will affect most jobs but eliminate very few, and should offset displacement with the creation of many more jobs (especially in AI creation/management). They suggest that AI could ultimately boost productivity by 7% or more.
It always feels safe to use historical data to predict the future, but what if we’re vastly underestimating AI’s impact because we’re too focused on the past?
AI is not like electricity, or even the internet. AI is now accessible to billions of people (anyone with an internet connection)—no expensive factory retrofits to switch from steam to electric power, no waiting for utility infrastructure to be put into place. Businesses are ready to adopt AI today, too. They see the benefits and can invest with little hesitation because AI is generally as good as humans at generating content, it’s extremely cheap compared to earlier breakthrough technologies, and, for the most part, companies can embed AI with minimal effort.
Additionally, our economy is completely different than it was 100 years ago when electric power was transforming manufacturing. Our modern economy is driven by the all-powerful “services” sector, which includes pretty much everything except farming and manufacturing, two industries where physical materials are turned into physical products.4 The main inputs in the productivity equation today are human knowledge and software, not natural resources or physical goods. These inputs are far more replaceable than steam engines, for example.
Today, AI is mostly helping people do their jobs better and faster, which should lead to labor productivity growth in the short term. It’s also democratizing knowledge, helping people accomplish tasks that previously took significant time to learn. We’re on the cusp of a transition, though. AI is meant to do far more than automate relatively simple, manual tasks that don’t require highly specialized knowledge or creativity. The big, lasting benefits of artificial intelligence will be realized when AI systems can “think” for themselves and complete complex tasks with minimal or no human intervention. At that point, many (or most) knowledge-focused jobs will be at risk. For example, imagine a future in which an entire movie or game is created by one creative person with some generative AI models. That doesn’t even sound that farfetched to me, but a movie that grosses $50M with a budget of only $500K in software/hosting costs would radically change the entertainment industry, right?
So, what happens if AI does eliminate more jobs than we think or if it has more of a transformational effect on more industries than expected?
Today, our economy works because of a delicate balance between production and consumption (and monetary policy when the Fed does its job). Individuals are a critical piece of both sides of the supply-demand equation. Average people use their time, skills, and effort to produce stuff for companies in exchange for money. They use that money to consume stuff, making production and economic growth possible on an endless loop, ceteris paribus.
As AI evolves, it will become more powerful. It should match the average human’s intelligence and abilities. When this happens, we could see a rapid commoditization of human labor, whereby the “buyers” of labor (companies) suddenly realize that they’re paying way too much for their labor inputs when AI is just as good (or better) and can work 24/7 because it doesn’t get tired, demotivated, hungry, or need pay/benefits. What will stop companies from replacing their human workers with AI? Some outdated moral obligation to society? Not without some regulatory guardrails…
Where would all those displaced workers go?
New industries and jobs will likely emerge around this new AI ecosystem, but that may not be enough, especially considering that the displaced workers probably won’t be capable of working with AI (at least not without some hardcore up-skilling).
Productivity gains are worthless if people can’t find work & can’t afford the stuff being produced. Think of the implications for our financial system if people couldn’t afford housing, goods, and services. Without consumers, the production-consumption feedback loop would grind to a halt.
This is a dramatic example, and probably unlikely, but without a plan for maintaining economic equilibrium, the whole system will go out of whack, with unknown ramifications for the function of other systems.
My hypothesis is that we’re closer to reaching that point with AI than most people think. If AI gets good enough to replace humans completely for many jobs (I think it will), how will we be able to feed the production-consumption feedback loop when fewer people can find work to make a living? What’s the next frontier that will get people into new fields and jobs?
Advancements in AI may require us to rethink our economic and governmental systems. It may require us to spend significant time & money on upskilling and training to get displaced workers into critical industries like energy, infrastructure, and climate tech. It may require a shift in government policy around the world to look more like Denmark and less like the US.
In summary, the future of AI is uncertain, but it could have a bigger effect than previous innovations like mechanized agriculture, electricity, and the internet for several reasons:
Our modern economy is built around work that AI can theoretically replace.
AI is already good as humans at generating content (or approaching the baseline).5
AI is widely accessible & cheap to use.6
Businesses are ready to adopt AI (and have been for years).7
Trying to predict the future is always a losing game, especially the far future. Complex systems adapt and change over time. Their behavior is unpredictable. Adding AI into the mix will likely lead to amazing emergent behavior that can’t be predicted, much less controlled.
We need to prepare for our artificially intelligent future with a systems perspective. We can’t think about AI in narrow terms — Will it take jobs, or won’t it? How much productivity will we gain from AI adoption? How do we make AI ethical?
These are good questions, but they narrow our field of vision too much to see the systemic impacts AI could have. The economy is tightly interconnected to so many other complex systems. An abrupt shift in how the economy works could have dramatic and catastrophic ripple effects on society.
We can’t predict the future of artificial intelligence, but we can deepen our understanding of how it might affect the labor market, economy, and society. We need to think about what our preferable AI future would look like and how to build that future responsibly. If we don’t, we will be forced to react. When we react, we tend to make bad decisions…
We can proactively shape the future we want and prepare to adapt if things don’t work out as hoped. I won’t claim to know what our future should look like, but I know I don’t want to end up in an AI dystopia like in William Gibson’s Neuromancer.
Only time will tell how far AI can go. For now, I’ll be trying to learn how to use AI tools for myself. In the short term, AI probably won’t take our jobs, but a person using AI might.
Thanks for reading! I’d love to hear your thoughts on AI and its possible impacts on the economy & society. Send me a note or comment here if you want to share.
To learn more about LLMs and what they do, I suggest reading Stephen Wolfram’s article “What is ChatGPT Doing…and Why Does It Work?”
Let me know if you need to get around the NYT paywall. I’ll send you an open link.
Productivity is a measure of economic output per labor hour. Technologies increase productivity by reducing the inputs needed to produce a certain amount of output (or increase the output from a certain amount of inputs — kind of the same thing). Electricity boosted manufacturing productivity in many ways. The simplest of these was that factories could stay open later and work around the clock, thereby increasing daily production.
Productivity gains are often associated with rising quality of life. For example, electrification reduced many burdens of daily life and allowed people & companies to do more with less time. This unlocks innovation by freeing up time and energy to focus on what’s new, not just what’s necessary for survival.
Per Statista:
In 2021, the agriculture sector contributed around 0.96 percent to the Gross Domestic Product (GDP) of the United States. In that same year, 17.88 percent came from industry, and the service sector contributed the most to the GDP, at 77.6 percent.
Soon it will be as good or better at many other things too, leading to commoditization of knowledge labor. This hasn’t happened with robotics yet because nobody has figured out how to make robots cheap enough, but AI is already getting there. If companies can increase productivity dramatically by reducing input costs (labor), what incentive do they have to keep humans on board?
LLMs and APIs make AI widely accessible and very cheap to use. Generative AI tools are mostly plug-and-play — little technical knowledge required — and available to anyone with an internet connection. Electrifying a factory, on the other hand, took electrical engineering expertise, which few people had at the turn of the 20th century.
The cost today is essentially pay-as-you-go server costs, like with cloud platforms. Unless things change dramatically, AI will only get cheaper to use as computers, chips, and cloud providers get more efficient and cheaper.
Per Deloitte research, payroll is one of the biggest expenses for most companies. Their research revealed that Fortune 100 companies alone spend over $1 billion per year on payroll (equivalent to 50% to 60% of total spending). LLMs are already dramatically cheaper than human labor. It’s reasonable to assume that highly advanced AI models built in the next few years will be much cheaper and more powerful.
Can the productivity gains of using AI justify keeping the same amount of staff? Or will it be possible to replace many employees with AI for productivity gains and much, much lower costs?
There are some notable examples of AI in business already. For example, Duolingo has used AI for years to optimize learning paths. Amazon generates a large portion of its revenue by using AI to recommend products to customers. Hell, I even use AI to help me generate content for work (don’t tell anybody).