Dear Friends,
I hope to live to 100 with a clear mind to engage with my writing from the previous 80 years.1 I sure wish that Vernor Vinge made it to 100. But alas, he passed away last week at 79 — the average American life expectancy. Back in 1993, while I was pubescently slow-dancing to Boyz II Men at my junior high formal dance, Vinge presented his paper “The Coming Technological Singularity: How to Survive in the Post-Human Era.” A decade later, after I graduated from college, I first encountered blog posts about the singularity and transhumanism — the belief that we’ll transcend our biological and mental limitations through technology. At the time, I dismissed it as sci-fi nerdery. Twenty years later, as I ease into middle age, it seems not just likely but practically inevitable and already underway. (I hope I’m wrong. I suspect that most AI applications — similar to social media, junk food, and plastic surgery — won't enhance our well-being, despite their irresistible allure.2)
Vinge was remarkably specific in his prediction for when computers would achieve human-level intelligence: “I'll be surprised if this event occurs before 2005 or after 2030.” Wouldn’t it have been nice if Vinge had lived to at least 85 so that he could have seen the state of artificial intelligence by his 2030 deadline?
AI predictions: Why was this time different?
In his 1993 paper, Vinge cites Charles Platt’s 1980s observation that artificial general intelligence was always “30 years in the future.” So what changed? Vinge thought it would be advancements in hardware. But a new approach to software proved equally important:
Steve Levy narrates the fascinating inside story of how “8 Google Employees Invented Modern AI” in 2017.3 It is a reminder of how often we undervalue new ideas: Google executives were slow to recognize the potential of the transformer AI model, and OpenAI pounced while Google hesitated. Levy also notes that six of the eight researchers were born outside of the United States, and the other two were children of immigrants. All eight have since left Google.
I expected AI to advance quickly in my predictions for 2023, but not this quickly. I thought it would take several years to get to the world-building possibilities of OpenAI’s text-based video generator:
Nor did I expect AI to become embedded so quickly in robots so that they can learn 1) by observing human behaviors and movements, and 2) learn from infinite optimization experiments in virtual worlds.
We seem to be racing very quickly toward the kind of “digital people” described by Holden Karnofsky in 2021.
In an interview last week, Sam Altman said he expects that most employees at OpenAI will soon spend more time considering the risks of new AI products than developing those products.4 They will wait at least a year — and notably until after the US election — before they release Sora, their powerful text-to-video generator. And Altman refuses to provide a timeline for ChatGPT 5’s release, which he says will show a bigger performance leap from version 4 than 4 did from 3. 🤯
I can’t think of another consumer technology product that has been held back not because it’s not ready, but because it’s too powerful.
The Most Special Generation
I’m sure every generation feels special.5 Maybe most generations even feel theirs is the most special. If so, then I am taking part in the time-honored tradition of justifying why my particular century of existence is the anomaly. First, ours is the special generation to witness a massive expansion of our species, which is likely to decline as quickly as we boomed:
Second, as we approach peak population over the next 50 years — with growth driven primarily by sub-Saharan Africa — we will develop robots and digital people that are more capable than humans.
Vernor Vinge dedicated his entire life to imagining the wild scenarios for the world after artificial general intelligence. But he didn’t live long enough to see what would — and wouldn’t — come to pass.
I have too many interests to focus entirely on AI. But I’m grateful that I have this sabbatical year to spend hours reading papers, fiddling around with new models, and listening to podcasts — from Tristan Harris’ alarmism to Marc Andreessen’s optimism — to consider what awaits us in the next half-century. And to write to future me. Hopefully, I’ll like it. Hopefully, my humanoid companions taking care of me will like it too.
What a time to be alive!
I am spending a good portion of every day writing to future me. Some of that writing is here in the Time Capsule meant for a public readership in the 2040s but most of it is in my journal — free from the self-censorship that comes with writing in public. I’m okay with my body deteriorating in the 2070s so long as my mind stays sharp.
Here’s the full article:
“How do you provide that layer that protects the model from doing crazy, dangerous things? I think there will come a point where that's mostly what the whole company thinks about.”
Imagine belonging to the generation that defeats Fascism, or that overthrows the monarchy, or “discovers” the Western Hemisphere, or builds a nuclear bomb.