> If this was 2025, this would be called machine learning because that's really what it was.

It would be called "machine learning" because that's the buzzword du jour.

> She was teaching the network to learn how to respond to nodes dropping out.

That's just called "writing software" not "teaching the network."

> Machine learning was definitely nonexistent at that point.

Are you sure about that?

> And yet, if you look at this 1964 paper, it's kind of unquestionably what it is.

The document: https://www.rand.org/pubs/research_memoranda/RM3103.html

The claim: highly questionable.

The paper is interesting in it's own right, but, to hype it up in this way is gross.

> > Machine learning was definitely nonexistent at that point.

> Are you sure about that?

Incredible statement to make, not only did machine learning exist, but neural networks existed!

The first perceptrons were built in the 50s: https://en.wikipedia.org/wiki/Perceptron

If you take a machine learning class, what is the most basic network you will probably build/learn about as an introduction? The MLP - multi-layer perceptron.

It's not even remotely obscure to know ML existed in the 50s and 60s.

> That's just called "writing software" not "teaching the network."

I would have expected better from Scientific American. The transcript read as very repetitive.

It's interesting reading how TTL evolved from the 'handover' field (p14-16).

Also if you read Wikipedia it looks like the main contribution was a simulator.