nicholasdecker.substack.com/p/ai-alignment-is-the-default
1 correction found
There is no apparent anticipation of machine learning
This is contradicted by LessWrong posts from 2008 that explicitly discuss neural networks, training data, learning rules, and other machine-learning approaches to AI.
Full reasoning
LessWrong did in fact discuss machine-learning-style AI in the 2000s.
Two concrete counterexamples from 2008:
-
In “Logical or Connectionist AI?” (Nov. 17, 2008), Eliezer Yudkowsky explicitly contrasts symbolic AI with connectionism, describes the Perceptron, explains that it learns from training data, and discusses backpropagation and gradient-based learning. That is direct anticipation of machine-learning-based AI, not just a hand-written symbolic agent.
-
In “Neural Categories” (Feb. 10, 2008), Yudkowsky explicitly proposes creating an artificial neural network, says that “A neural network needs a learning rule,” and discusses training rules such as Hebb’s Rule. Again, this is plainly machine-learning framing.
A third 2008 LessWrong post, “Artificial Mysterious Intelligence,” also shows that neural-network-based AI was already part of the discourse: it opens with a dialogue about whether AI should be built using neural networks and later lists proposals to use “the most advanced available neural network algorithms.”
So the blanket historical claim that there was “no apparent anticipation of machine learning” on LessWrong-era writing is not accurate. Those writings may differ from modern deep-learning practice in important ways, but they unmistakably anticipated machine-learning approaches as part of the AI discussion.
3 sources
- Logical or Connectionist AI? - LessWrong
17th Nov 2008 ... “And then along came the amazing revolution, the new AI, namely connectionism.” ... “To learn from the training data, you checked the current classification on an input, and if it was wrong, you dropped a delta on all the weights to nudge the classification in the right direction.”
- Neural Categories - LessWrong
10th Feb 2008 ... “Suppose I want to create an Artificial Neural Network (ANN) ...” ... “A neural network needs a learning rule.”
- Artificial Mysterious Intelligence - LessWrong
7th Dec 2008 ... “Him: Oh, you're working on AI! Are you using neural networks?” ... “Use it to run the most advanced available neural network algorithms.”