The Future and AI, Part 2
I published the first part of this blog a while back, and wrote most of this post around the same time. I just found some time to clean this up and publish it. I was surprised I still agreed with most of my thinking nearly a year later.
AI Will Cause Rapid Job Loss
The most concerning impact Artificial Intelligence will have on society is job displacement. Yes, every technological revolution in the past has threatened job security. Yet, there have always been more than enough jobs. Especially in the current age, we can’t find enough workers to do the jobs that we need done as a society.
Although history is against me, I do believe that AI will displace jobs in a way that is truly different than previous technology revolutions.
Think about these categories:
It’s hard to imagine the number of jobs that will be displaced by some of these technologies even if they’re deployed only partially.
For instance, take AI self-driving cars. If they take off in a big way (which doesn’t require new cars on the road), there won’t be a need for insurance companies, car dealerships, or mechanics. Demand for owning a car will decrease if you can "rent" a robotaxi by the hour, which will then reduce the need for servicing these vehicles. The cost of real estate will fundamentally change as you can have a low-cost taxi drive you to work (I wouldn’t mind working in the back of car, but I’m allergic to the idea of a long commute). Parking lots will become obsolete, freeing up critical real estate in cities across the country. The cost of delivery will drop, and delivery drivers won’t be needed anymore. With sufficiently reduced delivery costs, goods from other countries will only become cheaper and cheaper, and supply will change fundamentally (this could also go the other way; the cost of domestically-created goods could decrease due to automation and completely change the US dependence of foreign producers).
And this is just thinking about self-driving cars, never mind a humanoid robot paired with a self-driving vehicle.
Here’s another way to think about it: after the Industrial Revolution, humans could move into roles to leverage our intelligence over our muscles/dexterity. When AI beats us at intelligence (and dexterity via humanoid robots), there’s no clear characteristic where AI or existing machines do not have an advantage. Cars did not create new jobs for horses. At some point, you have nothing to offer that the mechanical minds and muscles can’t do for cheaper.
Globally high interest rates, combined with labor regulations (such as higher hourly wages), will accelerate the adoption of these technologies. The relative buildout cost will decrease as the cost of human labor continue to increase, especially on the low end of the labor market. Companies are (
motivated by profit and have a fiduciary responsibility to their shareholders, so they will deploy these technologies quickly.
What Can AI Do that Humans Can’t?
Humans can only manage a certain number of variables in their heads. We are not skilled at extracting trends and conclusions from a massive data set.
Increased compute and model parameter count will only increase AI’s ability to be fed massive amounts of structured data and extract trends. This allows for answers to questions that humans could not answer before. Yes, this will definitely create privacy concerns and generally cause chaos, but there’s also a massive ability for a positive impact on humanity.
This will have a significant impact on critical fields such as medical diagnosis and pharmaceutical research. We’ll be able to uncover trends and find new ways of detecting and eliminating disease that would have never been possible before. Especially for those who have been abandoned by the medical system (common with chronic long-tail diseases), AI will unlock solutions to their chronic problems that would have never before have been discovered.
An interesting question here is, what is the difference between analysis and creativity? Can AI truly create and invent new ideas and mental models? Or is it limited to analyzing existing data and patterns? What type of thinking truly makes humans unique? What is truly human about humans?
I asked my daughter and she said "only humans can pray". I think she’s right. Only humans have a soul and therefore the ability to form authentically human relationships with others and with God.
Countries Will Compete Intensely to Build SOTA Models
Many think AI needs strong regulation and control, like nuclear weapons, and requires detailed regulation and government intervention. We’ve even seen some initial regulation attempts.
I fundamentally disagree with the AI nuclear analogy that many make. Nuclear weapons require expensive, hard-to-refine materials that can only be sourced from a couple of specific places on the Earth.
AI can run on a consumer MacBook. Training costs will continue to drop as new chips come online and more efficient computing architectures are developed. Basic AI access is already free: it’s amazing to me how many people just use the free ChatGPT model and have no idea that there is anything better. It’s already Good Enough for them and there’s no interest is investigating alternatives.
Additionally, AI will benefit everyone. Nuclear bombs mostly (only?) benefit state actors. Companies and individuals have a massive profit incentive to understand and implement AI as quickly as possible. Companies do not have a universal incentive to use nuclear bombs. My wife uses AI to get second opinions on medical questions for our kids today.
Attempting to regulate AI or creating treaties across nations won’t work. If just one country, company, or powerful enough actor doesn’t follow along, they will have a structural advantage against everyone else. Additionally, open source is a bit of a Pandora’s box: once open-source software (like llama) is published, it will continue to evolve on its own, and it’s very hard to shut down.
Crypto is a good analogy here: it’s possible to regulate crypto in specific countries, but you can’t "shut down crypto" through regulation, and any government that makes it hard to leverage crypto will miss out on the potential gains the new technology could bring to its economy.
The same applies to AI. It’s been interesting to watch the best open source models come out of China. Since they aren’t winning on hardware, or model performance, they chosen another dimension to attempt to take the field on. This shows the insane incentive that countries and companies have to win, or at least keep users engaged, while they continue to compete. Users on a model provide a virtuous cycle of data to improve performance, and if a country misses out on this, it could put them second in model performance, which will have long-reaching impacts.
Scaled Transparency Will Change Organizational Management Systems
One specific function that AI has made possible is cheap, fast, and incredibly accurate audio transcription. This doesn’t require special hardware, you can run this on your consumer MacBook: just check out this awesome macOS tool for accurate text-to-speech on your mac.
Right now, the standard in many high-performance organizations is to record and distribute detailed meeting notes for every meeting. This is incredibly tedious, even when employees are disciplined enough to execute it. Not to mention the human bias that permeates the notes and any follow-up.
With AI, all meetings can be recorded, transcribed, and indexed. You’ll be able to quickly ask, "What did we conclude six months ago at our strategy meeting about topic X?" and retrieve a summary and word-for-word transcription of the conversation.
Email transparency and recording every meeting have been attempted in the past, but it’s hard to extract the right information at the right time from this mound of data. AI will change this and make it easy to extract the right information from your company’s archive in real-time.
That’s going to change how organizations make decisions and hold themselves accountable.
Independent Media Will Become More Widely Distributed
Audio transcription, coupled with lip dubbing, will make it easy to edit audio and video content cheaply and then translate it into multiple languages.
Imagine if the content (text, audio, and video!) of every speaker could be translated into every language on earth.
What if inspirational stories made by indie filmmakers could be translated and perfectly lip-dubbed for less than $10k? What if sweeping B-roll for these videos could be generated for $100s instead of $10-100k and weeks of shooting? This is all going to be possible very soon.
And then imagine if this could be done in less than a day. How much broader could the impact of any of these efforts be? The cost of this sort of transcription and video editing will drop very quickly over the next months and years.
Indie Movie Producers Will Beat Hollywood
Hollywood has completely lost the script.
I am no movie buff, I have very average tastes. I watch a movie or two a month and enjoy a good series once or twice a year.
There are very few films that I’ve really enjoyed over the last decade: Dune, Vikings, House of David, and Turn are the only things I can think of. I’m not sure exactly how this has happened, but it’s at least partly due to loss of a true anthropology in the US. A technocratic society is by definition less interested in the human, and simmering in a disenchanted world that has been stripped of all mystery, makes for a really boring story.
One of the things I worked on in the last year around this these was a platform to distribute an indie movie a friend made (think substack for indie movie producers: self-hosted streaming and a kickstarter-like experience for fundraising for events at particular locations).
This friend raised a small amount of money (< $1M, including distribution costs) and before Veo was released made a great movie. It’s a "simple" movie in the sense that he designed the production to be low-cost and built more around the story and dialog. However, it worked. He produced the film, distributed it on his own.
It would be even easier to do this now. Give Veo a couple more iterations and that $1M film can be done for $300k. That’s going to make the economics of these niche films make sense for investors and they’ll be an explosion of interesting stories built by all sorts of folks across the world.
AGI is Far Away and Isn’t Worth Worrying About
We are a long way from true AGI. Matrix-like sentient systems that can operate, think, evolve on their own without input from the human creators of the systems. Right now, I’m struggling to get LLMs to reason about computer code longer than a couple hundred lines—and that has not seen a step function improvement in the last year (although the incremental improvements have been impressive)
If it does come, which it may (and probably sooner than we think, if it ever comes!), it will change everything. I’m not smart enough to understand all of the follow-on effects, so I largely ignore the fact that this is a possibility. If it does come, there’ll be lots of change to work through.
Megatrend Convergence
Multiple massive technology revolutions are happening simultaneously. It’s hard to reason about how all of these megatrends will interact. Here are a couple that I’ve been thinking about that are worth keeping in mind.
Crypto and Monitoring
Peter Thiel said "Crypto Is Libertarian, AI Is Communist". If AI hands additional power to the state, crypto hands power back to the people. The government can close the "exit doors" of the monetary aspect of crypto via regulation, but it is not possible to regulate crypto out of existence. Many crypto technologies are not related to money transfer (Diode, IPFS, ENS) and enable a "new internet" that is not dependent on any one entity and cannot be monitored like the existing internet substrate.
This will be a counterweight to the centralizing and surveillance possibilities that AI introduces.
Crypto and "Proof of Humanity"
The crypto community has done a lot of work creating systems to validate real-world data points in a secure, decentralized way. These are called "oracles" in crypto speak—a bridge from the decentralized virtual world to reality.
One of these inventions is the ability to validate various components of your life without actually transmitting the details. For instance:
- Answering the question "Is this person real?" without revealing who they are
- Generating tax documents without transmitting your name, social, etc.
- Allowing computation to run on data sets without exposing the data sets themselves (homomorphic encryption)
This will be really important in a world where it’s difficult to determine what is AI-generated and what is not. A SSN number and SMS verification won’t be enough—they’ll need to be a new way to sign anything online and verify that you are who you say you are. It’s hard to imagine whatever the solution to this problem is not being built on crypto rails in some way.
Crypto and Content Signatures
Relatedly, AI models make it easy to generate content that can be impossible to discern from reality. Lawn Care Planner is an interesting example. Most of the content and images on the site are AI-generated. The entire site was basically built by one person as a side project.
Crypto enables signed content to verify that it was created by a real person at a specific time, at a specific place. These two forces will compete with each other.
In other words, it’s not obvious that other technologies will not be quickly developed to combat the disruption of our traditional models of understanding what’s real and what’s not.
Virtual Reality and AI
Apple launched the best VR headset (If you haven’t tried the Vision Pro, book a demo in an Apple Store and experience it). It’s still in the early days, but it’s easy to imagine AI video models combined with AR experiences to generate infinite, compelling worlds.
For many, virtual reality may be more interesting, comfortable, and compelling than their actual life. I think this has an extreme ability to disrupt social norms and cause incredible harm to humanity. All vice is, in some way, divorcing yourself from reality: focusing on yourself and your needs instead of the wider reality around you.