Today’s AI is as crazy as the President. Cloud Czar Mark Zuckerberg has decided it’s profitable, and that the market battle to come demands dissenters be turned aside.
The leader of those dissenters is Yann LeCun. LeCun, a French immigrant, is one of the founders of modern AI, a seminal figure. Throwing LeCun aside cost Zuckerberg’s company $20 billion the day it was announced.
LeCun believes, quite rightly, that AI must have a humanist perspective to be useful. LLMs need to be grounded, their processes based on true facts. But assembling and orchestrating such a library is a tedious and even controversial job. What facts are we talking about? Whose facts?
I suspect this is where much of the pushback against LeCun originates. The Cloud Czars, and their allies in the Administration, want to be the people who decide what facts are real and what facts are “fake news.” That’s what Musk’s Grokipedia is about. The Reich will tell you what is true and what must not be thought. (Musk is making himself into both a Cloud Czar and an AI Czar.) If an AI is tied to Musk’s encyclopedia, if the Cloud Czars are allowed to define truth, then there will be no escape from the crazy.
This is what LeCun’s revolt is all about.
Building a world model doesn’t just require true facts, but programming that builds answers around them. Existing world models are small, often defined within corporate databases, dramatically limiting what honest AI can do. LeCun was working up his vision for a decade at Meta, before being tossed overboard. He always said it would take time, and would be frightfully expensive, that there are no shortcuts.
Not what a Cloud Czar wants to hear.
LeCun He Do It?
LeCun claims to have few complaints about the Wall Street Journal article announcing his rebellion. Right now, money to fund his vision is abundant. Will it still be abundant when people learn how much his vision will cost, and how far off it is?
This is the difference between science and engineering, something my son has experienced moving from a job at Emory University to a Ph.D program at Georgia Tech. Science is willing to wait for truth.
You must do that if you’re doing it right. If a project takes 20 years and comes up with no worthwhile drugs it was still science and proved many theories wrong.
Engineers don’t work that way. Once you’re connected to the financial blowtorch, you’re on deadline to come up with a profitable product. Agile programming, where you release a “barely good enough” prototype, then build on it, comes from an engineering mindset.
LeCun is approaching the financial blowtorch for a journey that will take years to reach its destination and may not be useful in the meantime. That’s what Meta found out. That’s why LeCun was shunted aside.
Where is he going to get his money?
I am certain he will get some big money. It just may not be enough money to fulfill the vision. I am also uncertain whether LeCun’s vision can work if it’s just built by a small number of wizards. Just getting the facts straight is a Wikipedia-sized task. Will he license Wikipedia?
Making AI Human Takes Humans
Making AI human, in other words, is going to take a lot of work from a lot of human beings. They can’t all be paid like wizards. Most will have to take “Creative Class” salaries. It will be middle class people who save us, if anyone can.
All this makes LeCun’s rebellion the most important AI story of the year. His revolution is as important in its way as the 1776 rebellion Ken Burns is documenting this week. It won’t all be fought in Silicon Valley, but in 10,000 places around the world.
The result will define the world your grandchildren live in.
