On the War Nerd Podcast
A few months ago, one of the people that read this blog introduced me to John Dolan and Mark Ames, who run the War Nerd Podcast.
I think I was able to touch upon a few of the issues impacting our perceptions of nuclear proliferation.
I think the interview went well.
Please let me know what you think.
posted by maverick @ 4:29 AM
15 Comments:
FWIW - they got the spelling of my name wrong and they got my credentials mixed up - but overall a good interview.
Great iview. Accessible to the layman. Any chance your posts here are available in an ebook or similar?
Dear dg - thanks.
I was trying to stay out of all the details of nuclear weapons design (I think there is too much knowledge out there already on that topic).
As most of the people in the world now have never seen or heard of a nuclear bomb - I wanted to explain to people why these weapons came to exist in the first place. I also wanted to sensitize the next generation of decision makers to the fact that we are "over and done with" when it comes to these weapons.
Unfortunately when these issues come into the mainstream media, the discourse is dominated by the two extreme factions - the Non Proliferation Ayatollahs of the Beltway and the Ha patishim proponents. This is not a sensible manner in which to conduct a discourse on proliferation issues. We desperately need a middle ground in which to address proliferation threats.
I have not released an ebook. I should just do it. Let me see if I can pull it together in the coming month or so.
And almost on cue...
https://www.washingtonpost.com/world/national-security/a-secret-group-bought-the-ingredients-for-a-dirty-bomb--here-in-the-us/2016/08/03/46901c6e-58ae-11e6-9767-f6c947fd0cb8_story.html
"Here’s how they did it: In Dallas, they incorporated a shell company they never intended to run and rented office space in a nondescript industrial park, merely to create an address for the license application. In a spot on the form where they were supposed to identify their safety officer, they made up a name and attached a fake résumé. They claimed to need the material to power an industrial gauge used in oil and gas exploration.
Last year, their application was sent not to Washington but to Texas regulators, who had been deputized by the NRC to grant licenses without federal review. When the state’s inspector visited the fake office, he saw it was empty and had no security precautions. But members of the group assured him that once they had a license, they would be able to make the security and safety improvements.
So the inspector, who always carried licenses with him, handed them one on the spot."
And this is why I keep saying - we need to stay on top of the weapons issue.
I really liked the interview. That's how I got to know this great blog, I've been following The War Nerd for a long time. I think they made intelligent questions that you answered greatly, so all of us listeners ended up knowing more about nuclear weapons.
Well done!
Dear Miguel,
Thanks
http://www.newyorker.com/magazine/2008/12/15/atomic-john
Fascinating
This comment has been removed by the author.
Great interview, found your choice to avoid the specifics of bomb technology and emphasize the history of the motivation to create the technology in the first place very helpful. Since the interview I've run into, more than once, what seems an odd comparison between nuclear weapons and weaponized AI, with the premise being that AI will exist in a context of dispersed, globalized knowledge - in strict contrast, they say, to protected, centralized, non-distributed knowledge of nuclear weapons! Would be curious about your thoughts on this
Dear Willis,
Thank you for the compliments.
I don't like discussing the specifics of weapons design is never a good idea to have in the open. Also the exact details of how a physics idea is actually implemented in an actual weapon (for example how "dial-a-yield' works) is extremely secret and releasing that kind of information to the public carries very serious legal penalties in addition terrible moral responsibilities.
I am actually not sure what a weaponized AI is. AFAIK no AI is "free" in the same sense that "Northern Lights" or "Samaritan" is in the fictional world.
AFAIK as things stand most AIs are "blended models' which use a variety of signal filters (which can be AIs in themselves) to do a peculiar form of unsupervised learning. A lot of what is published in open source publications on AIs is pretty useless when it comes to an actual working AI in the field. This is because each step layer you add to a blended model comes with resource costs - which usually cannot be accommodated in high "velocity" data environments.
There is another hidden cost (which the movies typically gloss over). Even if you come up with the best AI idea in the world, you still have to wait several years before reliable codes can be built to run the AI on a data volume of size. You can write something small on your home computer, but to get a piece of code that works reliably on say the Amazon customer data cloud - that is a different story.
That being said - I think the general trend is to have each AI layer produce accuracy data periodically and that is reviewed by a neutral panel of human judges. If either the false positive rate or the false negative rate becomes unacceptable then the panel recommends that corrective action be taken.
Again - it is difficult to comment on this with any level of certainty as there aren't exactly commonly held protocols for evaluating AIs.
The exact knowledge of the blending used to make a viable AI at scale is sequestered just as nuclear know-how. This is due to commercial reasons (primarily).
Thanks for the detailed reply, that's very interesting. There's plenty of hype circulating about AI, and in its shallower and simplistic forms that hype seems to lend itself to a periodization of our age as somehow "post-nuclear" or moving in that direction, so it's good to hear your down-to-earth points.
Dear Willis,
My own view on AI is informed largely by my personal experiences and I absolutely can't claim be an expert on AI.
There are people who have dedicated their lives to it, I have done no such thing, I have used AI to solve some problems I was interested in.
The main problem with building AIs is you have to create a large enough training data set. This can be a really resource intensive affair and people try to get around this problem in various ways.
I was quite impressed by Microsoft's Tay chatbot, which I felt was a brilliant attempt at unsupervised learning. As expected trolls infiltrated tay's training data and the experiment had to be stopped, but I think it would have been interesting to see how long Tay would have to operate before she "unlearned".
Hi Sanil,
I'm a regular War Nerd podcast listener, and I really enjoyed this interview. I liked your honesty and straight forward speaking style. Easy for the layman. Hopefully you'll be back on the war nerd podcast soon.
Hi Matt,
Thank you for your kind words.
Best Regards
Sunil
https://m.facebook.com/story.php?story_fbid=1677469852576583&id=100009407541064
On Sweden's secret nuke program (from the war nerd FB group)
Post a Comment
<< Home