3 ms·
This is exactly why I have been advocating for a public Manhattan Project for General AI for a long time. I made the case while I was part of the Defense Intell
by AndrewKemendo 10y ago
This is exactly why I have been advocating for a public Manhattan Project for General AI for a long time. I made the case while I was part of the Defense Intelligence Agency in 2014 but basically nobody cared at that point.
I was in the Defense Innovation Board small group session with Mattis, Schmidt etc... about a month ago and it was clear that since then the message was received by some but not all (which is partly why I was invited as a reserve officer).
As a result the Deputy Secretary of Defense issued a memo [1] recently which took some of the first steps to get the ball rolling, but frankly it may be too little too late.
OpenAI is generally on the right track, but even then it's primarily dominated by the big players and they don't have incentives to push their best research, data, models, resources to that - of course why would they?
[1]https://www.scribd.com/document/346681336/Establishment-of-the-AWCFT-Project-Maven#from_embed https://www.scribd.com/document/346681336/Establishment-of-t...
- jorblumesea 10y agoThe entire idea of the Manhatten Project was that it's such a game changer you would want it classified. It also included luminaries from the public and private sectors. For all we know, the Government is collaborating with tech partners to create such a thing as we speak but will not disclose because of national security. Remember the internet, gps, modern telecom etc all came from government research, or some combination of public and private money but were generally declassified 10-15 years after its invention and use (some exceptions obviously). It stands to reason the next generation of AI is already being built in a similar fashion and we'll look back in time and see the patterns in hindsight. For example, how does the NSA sift through its giant treasure trove of information? Isn't that what AI/ML/Deep Learning is designed to do? In my opinion, the government is very up to date with current trends, maybe more so than people realize.
- jackpirate 10y ago>Remember the internet, gps, modern telecom all came from government research, or some combination of public and private money but was only declassified 10-15 years after its invention and use. It's true that a lot of the funding was from ARPA, but I believe that the vast majority of this research was published in standard academic venues. I know of no results in these fields that were classified at the time of discovery and released only 10-15 years later as you claim (but I'd love to see references if you have them!).
- AndrewKemendo 10y agoIt stands to reason the next generation of AI is already being built in a similar fashion and we'll look back in time and see the patterns in hindsight. Except I can tell you that it's not at least not at the same pace as AFGAM.
- vowelless 10y agoI completely agree. While DARPA et al have been involved with robotics (self driving cars, drones, etc), there should be a public AI project as well. Why do you think the message was not received?
- AndrewKemendo 10y agoBecause it doesn't seem like an imminent existential threat yet. The USG generally only acts at scale when there is a huge big scary thing pounding at the door. Unfortunately at that point it's probably too late. The USG is so big with so many different "rice bowls" that it can't efficiently coordinate between all of them - especially not for future facing projects. It's true that there are pockets of research and innovation in ML across the USG, specifically the DoD, however it's nowhere near at Google/Facebook scale.
- robertk 10y agoAdvocating a Manhattan Project for General AI without corresponding propaganda for AI Safety Research is unfair and reckless, even if you disagree with the premises of AI safety research. Edit: see below. Especially if you disagree with the premise.
- AndrewKemendo 10y agoif you disagree with the premises of AI safety research Which I do, so I disagree with your premise.
- robertk 10y agoYou don't get what I'm saying. If you're advocating for X and another group advocates for not X, you should ensure both sides have representation. Otherwise your action is equivalent in sanity to Trump cutting global warming funding because he disagrees with the premise. Put another way, if reputable sources contradict you on matters with great impact, do not neglect to fund their voice on the off chance they are correct. This is the same reason why I give the biotech threat guys a voice even if I disagree with their premise. Better?
- AndrewKemendo 10y agoI understand your point but I disagree with the concept of making AGI existentially safe in the long term - I argue it's a functional impossibility and a waste of time and money to research. Biological humanity will ideally be obsolete in the long term. That said, ceding control of critical control systems (SCADA for power, water etc...) needs to be done in the right way to actually ensure that the implementation is better, but that's just good engineering.
- flycaliguy 10y agoCan you point an outsider towards some reading that will help me make sense of the statement "Biological humanity will ideally be obsolete in the long term."?