I posted a very long post in early February about the current factors labor should consider at this link, and with how quickly things have moved since then I want to provide a more targeted follow-up.
The TL;DR is basically:
- As insane as it sounds, expect full video generation (with cohesive narrative) by 2025.
- Expect robots to begin working closely as “assistants” to human workers in nearly all manual-labor professions that are currently undergoing labor shortages.
- Fully expect AI “assistants” to start to supplant knowledge and creative workers in high-wage, low-labor availability industries.
- Expect anti-labor and anti-democratic propaganda to get better than you can imagine.
Part 1: Synthetic Media
How did we get here?
Everybody basically thought art would be the last thing to go. There has been a prevailing belief for thousands of years that one of the major features of the human brain that sets us apart from animals is our ability to be creative, in all of its forms. So, it was assumed that since AI would naturally follow the same path animals did to form human intelligence, this would hold true. We were wrong.
As it turns out, art is, at its core, simply the result of taking pieces of information and creating the proper composition of components to evoke feeling in humans. I apologize to any artists reading this, I genuinely believe what you do is endlessly important, and it is a rare gift for a person to possess advanced artistry. But apparently it's not hard for machines.
Where are we now?
As I am sure many of you are aware, AI art has gone from niche to mainstream in mere months. For the uninitiated, here are some of the most state of the art tools to create it:
https://openai.com/dall-e-2/ – This tool was created by a for-profit entity called OpenAI, funded by the likes of Elon Musk, Microsoft, and many other Silicon Valley organizations. At 13 cents per “prompt” (with 4 images to choose from), it still provides the best quality of anything available to the public.
https://huggingface.co/spaces/stabilityai/stable-diffusion – This tool was created by a group called Stability AI, who purports to create “AI by the people, for the people”. They were basically founded to be a truly public version of something like OpenAI.
https://www.craiyon.com/ – Similar to stable-diffusion, this tool was created to make what OpenAI and others had done more widely accessible, for free.
Where are we headed?
It's important to understand those three AIs have only really come into existence in the last 6 months, with Dalle 2 becoming available to the public in July, and Stable Diffusion in August.
In May, this paper was published: https://plai.cs.ubc.ca/2022/05/20/flexible-diffusion-modeling-of-long-videos/ (here is a link to the actual paper: https://arxiv.org/abs/2205.11495). It went under the radar, as these things often do, but it is one of the first examples of generating not just images, not just videos, but long videos.
If you consider the average piece of media content you consume, you will quickly realize it falls within the following constraints:
- Somewhere less than 4 hours long.
- Made up of many scenes, each of which has narrative connection back to a central plot.
- Has a cast of characters, each with their own distinct and self-consistent personalities.
I do not have access to some special knowledge, all I know is that it seems clear longer videos are already being produced, albeit without a cast or plot. But it seems incredibly reasonable to assume that if short videos can be generated (scenes), and they can be strung together by humans into a plot, then having an AI string them together based on a script is hardly one leap away. The only real issue is consistent characters, which you will see in section 3 can already be generated, just as text for now.
The reason this is relevant to labor is that it is genuinely possible most creative industries will see truly massive labor disruption as a result of this. If this is you, don't fret. Learn how to use these tools. Do it now. Be the person that can say they have 5 years of experience making AI art in such and such field in 2027. Humans are not going away entirely. There will be demand, just maybe not for what you do now. It's even possible the creative fields could grow because of this, but it will be disruptive for a while.
Part 2: Robotic Assistants – Manual Labor
This is the one that always feels like a false promise. People have literally been making labor saving devices since humans started using rocks and sticks as tools, yet the promise of the “robot” has always fallen flat. That is no longer true.
I could go on about this for pages, so I'll just provide an anecdote, and a few examples, then move on.
Anecdote: Earlier this year, I was taking a tour of a factory somewhere in the Rust Belt of the United States. This factory belongs to an industry that has not yet heavily embraced automation. I was asking the general manager dozens of questions, and one of them was “Given the technology that you currently perceive to exist right now in automation, what is the biggest thing holding you back from embracing it?”. He simply held up his hand, stretched out his palm, opened and closed it, then twisted his wrist 90 degrees in both directions and said “We just don't quite have something that can do this. If we did, 30% of our workforce could be eliminated.”
He, unfortunately, does not even have a good perception of the current state of robotics. Meet Google's Everyday Robots project:
https://ai.googleblog.com/2022/08/towards-helpful-robots-grounding.html – In this blog post from August, Google basically lays out how they made a robot that can just sort of… do what you ask it. “Go get me a can of soda”, so it goes to the fridge, opens it, grabs the soda, and brings it to you. That's a bit of an under-explanation, but that's what all the technobabble in that basically article means.
Here's the kicker: In February, they actually published the first blog post on the same topic, https://ai.googleblog.com/2022/02/can-robots-follow-instructions-for-new.html and it is literally astounding how fast they made progress. They went from “put a red grape in a bowl” to “get me a soda from the fridge” in 6 months. They are not slowing down.
Here are some real-world robots already being used for various tasks I have read about recently:
https://sf.eater.com/2022/8/17/23308389/mezli-robot-restaurant-open-menu-san-francisco
This is relevant to labor because, as that factory manager told me without knowing it, about 30% of manufacturing jobs, service jobs, any human labor jobs will be rife for automation in the next few years. Just like with media, this does not have to be cause for panic. Join a union as soon as you can. Include automation provisions in your collective bargaining agreements. If you find yourself out of a job, reach out to non-profits and allies in your area. Get together with friends and start your own business, but try to make it fully automated. This can be a time for creation, not despair.
Part 3: Robotic Assistants – Knowledge Work
You know how in movies there's always the crazy scientist who has figured something out, but no one quite believes them until it's too late? That's how I've felt the last few years. First watching the virus spread through China like a wildfire starting in December 2019 and even the virologists I know ignoring it and calling me alarmist until literally the week the first shutdown orders occurred in March (lesson: virologists are not epidemiologists), and now unfortunately with AI. So let me just say this loud and clear:
IN MAY 2022, HUMANS CREATED THE FIRST HUMAN-LEVEL ARTIFICIAL INTELLIGENCE
I'm sure this seems ridiculous, you surely would have heard of this big of a leap, right? Right????
Probably not. Because, in classic “scientists are bad at talking to the public” fashion, it was buried deep in a single research paper put out by Google. But here it is:
In this blog post Google laid out their new AI: https://ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html
This is the actual paper: https://arxiv.org/pdf/2204.02311.pdf and that image is from the top of page 15.
What does it mean?
Basically, Google is saying “with this AI, as long as you give it 1-5 examples of what you are trying to do, it is now better than the average human at it”. So if I say “5 + 5 = 10” and then I say “10 + 10 =” it will respond with “20”.
Now, consider all the average tasks the average person performs each day. Literally, if a task can be described to you, a person can do it, and now so can AI. Oh and also? It's getting cheaper to make these. Read my comment here from the day PaLM came out for the economics involved. The basic gist is that make one of these that exceeds human maximum intelligence would cost a few hundred million dollars. That's it. That's a lot, but that's also it. And these things are already getting cheaper and cheaper every few months, I would guess that number is now more like $50 million for smarter than human AI. This isn't a joke, or a movie, or anything fake, this has already happened. If you've read the whole post, you'll notice PaLM from May is what Google is using to power the robots from August. This is its brain, those are its bodies.
Who does this actually affect?
Not who you would expect. Most immediately, the effects will be on teachers, healthcare professionals, and anyone that performs repetitive computer tasks. Very likely call centers and such as well to a lesser extent. AI powered teaching assistants and medical assistants are now possible, and they will be a $100 billion industry in the next few decades. Teacher shortage getting you down? Just have an AI tutor supplement the teacher's limited individual attention. Nurses leaving your state because of bad legislation? Just have an AI ask all the initial questions to the patient, labor problem solved! This path of advancement will be both a major boon and a major bust to the labor movement, and it must be considered as a factor sooner rather than later.
Part 4: Propaganda
Okay, so you've gotten this far, congrats! How might this all be used to dissuade the global labor movement? Well, unfortunately it's all the obvious stuff if you've watched any sci-fi movie in the last century.
Better media synthesis? Great, now a government can make a full-length feature-film to dissuade every individual citizen from their exact individual beliefs and make watching it fun and simple for all.
Better robots? Great, better weapons of war and police oppression worldwide.
Better AI? Great, now you can make an imaginary best-friend that follows every citizen everywhere they go on their phone, and reports any dissident activity immediately to the proper authorities.
So what can you actually do to combat this? I say, fight fire with fire.
Better media synthesis? Great, now messages of hope and resistance can be spread further and wider.
Better robots? Great, now community gardens can flourish, the poor can be fed, and the neighborhood watch can go toe to toe with any rogue state launching an invasion (*cough* Ukraine's use of cheap drones demonstrates that a conventional military actually making long-term territorial advances might be almost impossible for the first time in 12,000 years but no one has told the public *cough*)
Better AI? Great, now every human on the planet can be personally tutored by Einstein, Socrates, and Da Vinci whenever they want, basically for free, medical diagnosis of rare conditions can flourish, and people can live longer, happier, more stable lives.
SO, to sum up. These things are happenings. They are big. They are transformational. Stay educated. Stay smart. Use AI to make you smarter as much as you can: https://beta.openai.com/playground (When you talk to this AI, it helps a ton to use 'Q -' followed by your question, and end your prompt with “A -” so it knows you're trying to perform question answer. Just like PaLM, the more you've been talking to it all at once, the smarter it is). I only use this account for this kind of post, but feel free to message me any anxieties or questions about any of this and I'll try my best to get back to you. By the way, even with some of my alarmism, I am actually optimistic. I think the next 20-50 years are going to be super weird, and then the next 1000 super cool. That might just be the price we pay for progress this vast!