Don't Worry – It Can't Happen (1940) | Hacker News Reader

Don't Worry – It Can't Happen (1940)

dropbox.com

108 pointsapsec11228 comments

Of course it can't happen. As the Germans discussed at Farm Hall, you'd need giant factories, maybe the biggest buildings in the world, more than a mile long. They'd need giant power plants, like big hydroelectric dams, to power them. You'd need entire cities of scientists just to work out the theory. Other new cities for the plants. It would take hundreds or thousands of test explosions to get the chemical explosive components right. You'd need some way to watch an explosion as it happened, maybe even with an X-ray machine. The project would cost billions of dollars and take years. Even then, you might only make a few bombs.

Just one of those facilities, the Clinton Engineer Works, employed nearly a hundred thousand people. The vast majority of those did not realize what they were a part of. For example, this quote underneath one of the article photos:

"Gladys Owens, the woman seated in the foreground, did not know what she had been involved with until seeing this photo in a public tour of the facility fifty years later."

For those who want to know the scale of it all: https://en.wikipedia.org/wiki/Manhattan_Project#Project_site...

This is not as silly an article as it may appear to be in hindsight. When it appeared (in May 1940: https://www.scientificamerican.com/article/dont-worry-it-can...), most scientific authorities, including such non-idiots as Niels Bohr, believed that, while an atomic chain reaction was theoretically possible, the only way an it could be achieved was by bringing together a quantity of nuclear fuel so large as to make the effort impractical.

The first proof that this was not in fact the case came in the Frisch-Peierls memorandum (https://en.wikipedia.org/wiki/Frisch%E2%80%93Peierls_memoran...), which had been written two months before Harrington's article appeared, but which was highly classified and thus out of reach of pop-science journalists. Frisch and Peierls' figures indicated that a critical mass could be achieved with just one kilogram of uranium-235; this turned out to be an underestimate, but not so much as to invalidate their fundamental point, which was that criticality could be achieved with a mass of fuel small enough to be delivered by a 1940s-era airplane.

This came as news to the brightest lights in the world of physics, up to and including Albert Einstein, so it's probably unfair to expect a contemporary writer for Scientific American to be further ahead in the science than they were.

We talked a bit about the state of physics in 1940 in my 'Modern Physics' (aka Physics 401 aka nuclear/quantum). Looking backwards through time its easy to imagine they were planting disinformation or trying to throw off people from the potential there, but that view is mostly bolstered by knowing what is possible now. Had it come out in 1942 after 'pile 1' was demonstrating fission then yes, it would be subterfuge.

But perhaps most amazing are these results from physics experiments in 1940 and then the changes in the world by 1945. Such profound changes are also likely when someone figures out how to tap the energy from fusion as well.

There was some experimental evidence (and open publication) by 1939, to boot. The threat outlined by the Einstein-Szilard letter is that the Germans must be aware of the idea and might be developing a weapon so planting disinformation was probably not high on anyone's list of priorities.

Clark's First Law:

"When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong."

I wonder whether the author was naive or intended to misdirect. The french experiment he cited answered the question "With the samples as is could it explode?" and the answer was "It can't happen.".

The relevant question however was "Can the setup be engineered that it could be made to happen" and the answer to that with all that was known and elaborated in the article was: "Maybe if a larger percentage could be made slower or absorbed by other means".

Answering the wrong question is a common way to misdirect.

OP claims that the same other wrote a "worry about it" piece a year before.

https://mobile.twitter.com/wellerstein/status/84169090329099...

So basically, clickbait through the ages.

Too late to edit, but s/same other/same author/

Note that the Einstein letter which kicked off the Manhattan Project wasn't sent until August 1940[1].

[1] https://en.wikipedia.org/wiki/Einstein%E2%80%93Szil%C3%A1rd_...

The letter is dated August 2nd, 1939.

Apologies, you are entirely correct.

  Thus the reaction poohs out as the temperature rises.

'poohs out' ? I've looked up the definition of 'pooh', and there's nothing to indicate what this means in this context, can someone elaborate?

It's slang, meaning to lose momentum and stop. Saying "died out" or "fizzled out" is more common these days (eg: A spark jumped from the fire but quickly fizzled out on the snowy ground).

Euphemism for "poops out".

I.e shits out, shits the bed.

Of course this assumed you didn't enrich the uranium.

Right. The author shows no understanding of the difference in fissionability between the U-235 and U-238 isotopes. Naive, and yet, there's so much life and natural curiosity to the author's style, it's a delight to read regardless. Why isn't today's SA like that?

So what's the equivalent of enriched uranium for AI? Some form of recurrent neural net?

And what's critical mass? Some form of feedback system (a 'strange loop')?

All analogies are leaky, and I think this one is particularly misguided. Why would you want to draw parallels between uranium and AI? You can go a long way speculating about...

What's the equivalent of enriched uranium for healthcare? What's the equivalent of enriched uranium for calculus? What's the equivalent of enriched uranium for sweatshirts?

It's actually a much better analogy than you think. Machine learning (which is at least a significant part of AI) requires a certain 'critical mass' of computing power and learning data before it becomes powerful. We had neuron models in the 90s that were roughly equivalent to what we have today and they were deemed largely useless, just like the uranium in the article. All that changed was the data and compute power available.

As for 'critical mass', a defining characteristic of consciousness is that it's self-referential, so that seems like a fair guess.

but growth of AI usefulness hasn't been faster than Moore's Law, has it? so it doesn't seem to have a feedback loop

Well yeah, the feedback loop is assumed to happen once the AI system becomes powerful enough to understand and optimize its own operation better than humans can. Basically the premise is that AI development speed is bounded by the capability of the mind working on the AI, so max(human, AI). AI intelligence grows faster than human intelligence (which has reached the limits of easy improvements and is thus arguably capped by population size, which has also stagnated), so the inflection point is assumed to happen once the second term outstrips the first.

It can replace all those pythons, javas with super efficient platform tailor-optimised machine code, gaining remarkable computational benefits. I mean programmers will also loose jobs one day, AI will replace them too only couple of years later than taxi drivers. But actually I'm looking forward to it, as programmers are extremely cocky these days.

Funny how they said: "With typical French - and scientific - caution ..."

More readable: https://www.dropbox.com/s/hzhilstmj94jvlx/1940-sciam-harring...

Thank you! The Twitter version in the OP is useless on a mobile device.

Reply on news.ycombinator.com