“A proud man is always looking down on things … as long as you are looking down, you cannot see something that is above you.” — C.S. Lewis
Humans are prideful.
We think that human intelligence is the pinnacle, and thus mock artificial intelligence, even the face of its incredible intellect and creativity.
AI can code in any language, make amazing art (we’ll get into that), and drive better than us with Tesla’s Full Self-Driving. These are all advancements of the last year.
To think that AI won’t eventually do your job better than you is sheer pride, or the ignorance of exponential growth.
AI generators — whether for text, art, code, or music — used to be quirky, fun little tools, producing what you see below on the left (a fever dream monstrosity with eyes and dog-faces everywhere). Now, you can tell AI to draw an idea (e.g. “an armchair in the shape of an avocado”), and it’ll instantly draw creative, functional art.
Copyist.AI, one of many recent products I found built on-top of OpenAI’s GPT-3, uses the technology to generate marketing copy, including ads, growth ideas, article outlines, and SEO optimization. The art above, on the right, was made by DALL-E, which isn’t open for public access yet (outside of the OpenAI team.) What is public, however, is text-generation (things like article titles, advertisement copy, marketing ideas, and so on.) I asked GPT-3 for article title ideas about “AI job loss” and here’s what I got, in an instant:
- “The Rise of Robots and the Decline of Jobs: How AI Will Impact the Economy”
- “The Machines Are Coming for Your Job”
- “What Happens to the Workforce When Robots Replace Humans?”
Whether you’re a coder, illustrator, copywriter, Uber-driver, factory worker…or, well, anything, it doesn’t take much imagination to see AI taking your job at some point.
Besides being prideful, humans are notoriously bad at understanding exponential growth. This is why COVID-19 took most of the world by surprise, even though experts saw it coming clear as day. 1 case turns into 2, into 4, into 8, and before you know it, nearly 100 million people (as of writing) have or have had COVID-19.
AI is another example of exponential growth, where it went from small and insignificant, to currently on par with or exceeding humans in many ways. Soon, AI will be billions of times smarter than us.
With exponential growth functions, it can be really hard to imagine what the long-term future will look like — but it’s inevitable, so we should at least consider it.
Many people think in the short-term or intermediate-term, in which AI job creation may actually occur, but this is truly insignificant when considering a larger time horizon (not just AI for factory automation — but some of what we explored above, such as in design, marketing, and coding).
Already, factory automation led to the loss of 400,000 jobs in the US from 1990 to 2007. Two million more manufacturing jobs could be automated away by 2025.
But the US isn’t a manufacturing economy, and it hasn’t been for a long time. Less than 13 million Americans are employed in manufacturing, while around 60 million Americans are knowledge workers — in fields like design, marketing, coding, and so on.
Globally, automation is expected to displace up to 800 million jobs by 2030, but even these estimates largely don’t consider recent advancements like GPT-3, DALL-E, and the inevitability of many-orders-of-magnitude improvements over those models.
It’s fascinating to consider the many objections people have to the idea of AI superiority and AI-induced job loss. I’ve mentioned that pride and ignorance lead many to criticize AI, but let’s take a closer look.
In one Medium article titled “Will the Latest AI Kill Coding?,” one of the top comments says this:
“A programmer will need to provide an example.”
This is an example of the inability to understand exponential growth, as AI naturally will improve to the point that it doesn’t need examples to create code. Indeed, it has already gotten there since that 6-month old comment. The GPT-3 “instruct series” enables generating text, including code, without giving examples.
Another comment says this:
“You know what you call a description of the behavior you want specific enough to generate a working software? Code.”
It’s just not true. This is all I told GPT-3 (no code examples given) to make a working website that functions as described:
Create a working HTML website that says “this website was made without a programmer, bro” above a button that says “X to doubt,” that, when clicked, creates a pop-up that says “it really was.”
Here’s the code GPT-3 generated immediately upon hitting “Submit” with the above request:
<html><head><title>this website was made without a programmer, bro</title></head><body><h1>this website was made without a programmer, bro</h1><button onclick="alert('it really was')">X to doubt</button></body></html>
And yes, the site works. It’s a fun example, but this can easily be productionized into something much greater.
Here’s another popular comment on the article:
“GPT-3 is the biggest lookup-table ever created.”
Recall the Avocado-chair from earlier — that does not exist on the Internet, so it cannot be retrieved via lookup. It was generated anew. Similiarly, the exact article title I generated earlier, “The Rise of Robots and the Decline of Jobs: How AI Will Impact the Economy,” returns 0 results on Google. The website I just made above did not exist, except for in my head.
Another comment makes a strawman argument by attacking Tesla:
“This is why Tesla vehicles are still crashing into box trucks.”
Real-world statistics show that Tesla is already far safer than humans.
We could go on all day with the criticisms against GPT-3, and even AI more broadly. In the Boston Dynamics video of robots dancing to the tune of “Do You Love Me?,” you’ll see comments saying things like “this is CGI.”
Humans truly are incredibly prideful. I’m sorry, but AI will take your job.