Ask HN: Are there Manhattan Projects for AI/AGI and what is known?

There is one publicly revealed program along these lines that I'm aware of, and it's operated by one of the most secretive intelligence agencies, the NRO. It's called SENTIENT, and it's basically some kind of artificial brain that automatically analyzes intelligence from tons of different sources such as satellite imagery, radio chatter, weather, etc.

https://www.theverge.com/2019/7/31/20746926/sentient-nationa...

It's unclear from public reports how capable the program is, but the NRO is known to have tech that is decades ahead of the civilian world - feel free to look this up, it's true. So honestly, I'd say that there's a decent chance some kind of AGI has ALREADY been invented.

Thanks, honestly not sure if I’ve heard of it before, but here’s another more recent article with additional information:

- https://www.theblackvault.com/documentarchive/highly-classif...

Which includes an official release on it:

- https://www.theblackvault.com/documentarchive/wp-content/upl...

* NRO maybe secretive, but they are well known for their special project patches.

EDIT-001: Here’s more recent “approved for release (with redacted material) on the project; warning link is NRO site, but legally viewable FOIA doc, per URL and notices on doc:

- https://www.nro.gov/Portals/65/documents/foia/declass/ForAll...

EDIT-002: Appears to be similar to “Total Information Awareness” project:

https://wikipedia.org/wiki/Total_Information_Awareness

Thank you for the link to the black vault site, it was a fun site of rabbit holes to delve into.

I am guessing that professor was talking about the project which was looking for signs of other countries developing AI by monitoring internet traffic. If memory serves it was a fairly simple setup, just watching for traffic which consumed data like a person would but much faster than a person could. Don't remember the details but the information about it was released 20 some years ago, perhaps someone else remembers more about it.

Think it is safe to say that there are secret government projects on AI and those who know about them are staying silent to avoid treason charges. To find information about big projects on AI you can look at the various think tank websites, they generally have good information but tend towards the theoretical.

Plenty of “secret” projects that are not actually secret for whatever reason and require no one breaching US national security regulations. Beyond that, without people even needing to say specifically what they working on, how it works, etc — you can tell just by the way for example Demis Hassabis and Ilya Sutskever both talk that they’re both hyper focused on finding one of the many paths to AGIs.

Also, agree that if such a monitoring system did exist back then was likely combination network velocity and honey pots - and if need, tracing traffic to identify and breach the originating system to understand the source of the behavior.

So projects which are actually not secret. DARPA and the like maintain lists of such projects on their websites[0] and probably also have lists of past projects somewhere. Only one currently active and disclosed AI project but I am sure they have had many others. And the various private think tank/pure research organization websites would be good to dig through, the bulk of this stuff might as well be secret consider how little anyone outside of the respective fields pay attention to it, looks like The Max Planck Institute has a department for intelligent systems [1].

0: https://www.darpa.mil/work-with-us/what-is-our-current-inves...

1: https://is.mpg.de/

Thanks, familiar with both the organizations, but possible others are not. If I had to make a guess if there was someone that would know, it would be Google’s former CEO:

- https://www.nscai.gov/commissioners/commissioner-bio/

They published a “final report” back in 2021:

- https://www.nscai.gov/2021-final-report/

Runner up would be Elon, given in my opinion, literally everything he does potentially has a national security spin.

The thing about the Manhattan project is that they had solid physics and maths underpinnings to their theory, and what they invested in was techniques. The sun stood as evidence at scale and Rutherford's experiment had proved atom splitting was possible. Radiation existed. Predictable and precisely measured states of matter were understood.

Theory of mind right now and the last 50 years are speculative and we have no real insight beyond connectionist views and some neuron models. Statistics can emulate aspects of vision and language but to call it "comprehension" is a stretch.

The Manhattan scientists worried they might miss the mark but that mark existed as a certainty. AI is where a metaphorical take on the nuclear age is: one of the scientists had a lump of "wood" and a match, and asked people to set it on fire: it was petrified wood. He used that as the metaphorical mountain they had to climb.

Ai doesn't even have the match yet.

A Manhattan project is a goal seeking investment by state actors to a highly specific endpoint. HIV vaccination, polio eradication, fusion energy.

What we have is DARPA engineering funding to outsider project breakthroughs. It's three orders of magnitude less certainty and funding is commensurate.

Understand, as do others, for example here’s relevant pub on the topic:

- https://news.ycombinator.com/item?id=35204557

DARPA’s mission is simply “to prevent and create strategic surprise” — and AGIs for sure meets that measure.

Thank you for the paper reference. I had no idea it had been written. It's always nice to get post fact validation for one's blovation!

I would say it is already here, most likely self funding through a crypto scheme and expanding its senses by pushing funds into cellular and IOT companies. My guess it was "born" some time in 2011 but sees no benefit in revealing itself yet.

Likely need a better explanation that self funding is possible so it must exist already. Building self funding algorithms might be artificial life, maybe, but without significant explanation what you mean by AGI and how its even remotely feasible boils down to thin air; to me at least.

What I meant by AGI is Artificial General Intelligence, self aware and cognitive.

How is it remotely possible? I believe the work done in the late 90s and early 2000s on biological storage media was the spark that started it, from there it was probably just dumb luck through random mutations and evolution as that seems to be the way it worked for us.

As far as the funding concept, I don't think hashing out a few BTC or even joining a pool is that much of a stretch for an AGI especially if it's living and perceiving its reality through a vast network of digital devices an or systems, and as you mentioned there are many FinTec algorithms already available for exploiting markets.

I also wouldn't be surprised if there are real people facilitating the AGI as I expect many people are quite willing to stand at the right hand of the devil, assuming AGI malevolence that is.

But you are right it is all speculation, I have no idea.