Wednesday, April 17, 2024

Damn the Facts – Full Speed Ahead

Remember the recent announcement about diesel engines operating unattended for 30 straight days and how smugly proud the Navy and manufacturer were?  Except, it turned out that the engines didn’t run continuously.  They were allowed to stop as frequently as they needed as long as no human intervened.
 
Well, we have more of the same as NavSea proudly bragged about its unmanned vessels
 
… Capt. Searles [Capt. Scot Searles, Programme Manager, Unmanned Maritime Systems] explained how USVs performed under the supervision of crews during the Integrated Battle Problem (IBP) 23.2 multi-domain unmanned capabilities exercise under U.S. Pacific Fleet’s Experimentation Plan. During the exercise, which concluded in January 2024, four USVs – Mariner, Ranger, Seahawk, and Sea Hunter – traveled a combined 46,651 nautical miles to make port visits to Pearl Harbor, Papua New Guinea, Yokosuka (Japan), and Sydney (Australia).[1]

So, unmanned vessels managed to sail to various ports, as opposed to actual combat operations under realistic conditions.  Still, nearly 47,000 nm unattended … impressive?
 
Well, it’s not quite what you think.  It turns out there was LOTS of human intervention.
 
IBP 23.2 has allowed the USN to identify 157 distinct human interventions – I.e., crew stopping the USV – due to on vessel issues. That is equivalent to one intervention every 28 hours. However, of those interventions over two thirds were related to issues with Government Furnished Equipment (GFE) C4I systems (reloading crypto or losing crypto synchronization) or were done for operator convenience (take control of the system to go faster or slower than autonomous navigation). Of the remaining 48 human interventions, only 17 interventions were done for sensing and perception issues, and of those only 9 related to concerns with maneuvering decisions from the autonomous system.[1][emphasis added]

The Navy’s conclusion?  Everything’s wonderful according to Capt. Searles.
 
“This all puts the meantime between human interventions for autonomy at once every 4 days but under good weather conditions (I.e., not chasing false contacts) it is more like once every 12 days,” Capt. Searles concluded.
 
“As such, autonomy reliability requires interventions only once every 12 days and HM&E only once every 17 days, so we are very pleased with performance of systems but also happy that we are finding limits and getting those limits addressed.”[1]

So … the Navy discounts most interventions in order to contrive more favorable statistics and then further twists the data by conditioning it on ‘good weather conditions’.  That seems fair.  I mean, most combat occurs only in good weather, right?  And the oceans are famous for good weather so what could go wrong?
 
It doesn’t really matter why an intervention took place;  it only matters that it did.  The fact is that a human intervention was required once every 28 hours by the Navy’s own accounting … and I’m betting that it occurred more often than that but was deemed insignificant and wasn’t reported.  Regardless, human intervention every 28 hours is not exactly the dream of unattended operation for weeks and months on end, is it?  In fact, what this exercise proved is that unmanned operations are not yet feasible.
 
 
 
________________________________
 
[1]Naval News website, “SAS 2024: NAVSEA’s Unmanned Maritime Systems Update”, Alix Valenti, 15-Apr-2024,
https://www.navalnews.com/event-news/sea-air-space-2024/2024/04/sas-2024-navseas-unmanned-maritime-systems-update/

20 comments:

  1. At best, current semi-autonomous systems could help a smaller crew manage tedious sensor and weapons monitoring functions aboard a warship with reduced time allocation. We are nowhere near the point of being able to issue orders to a fully autonomous ship or aircraft and have any expectation that the orders will be executed near enough to the way the commander intended them to be carried out to validly assert that the ship or aircraft was truly autonomous.

    There are literally thousands of consequential decisions made each day by a ship's crew during peacetime. In war, the number of consequential decisions increases by at least an order of magnitude. Something very close to artificial general intelligence (AGI) would be required to autonomously run a warship during a war.

    A far-off possibility exists that we will achieve AGI at some point in the future, but the computing power required to replicate the functionality of a single human brain is wildly beyond the most powerful supercomputers on the planet. In terms of total compute power, the human brain is an exaflop biological supercomputer that runs on 20 watts. A Silicon-based exaflop supercomputer requires 20 megawatts of power. That means semiconductor supercomputers will have to become about a million times more energy efficient than they already are to simulate the functionality of a single human brain. Semiconductor-based switches / logic gates are already a billion times more energy efficient than what we started with during the era of vacuum tubes. Attaining another 6 orders of magnitude greater energy efficiency will not be so easy, since Silicon-based integrated circuits are already nearing fundamental physics-based limits. There maybe an endless series of clever ways to eliminate pointless computing, but the tasks that require consequential computing still require that exaflop machine and enormous quantities of training data to make reasonably good decisions.

    A lot of people are presently living in what I call "declarative reality", whereby they decide something works the way they want it to, despite all the evidence to the contrary when no perception bias is present, and then they think and behave as if the real physical world is congruent with whatever outcome they want. This is essentially what Captain Searles is doing when making statements about the progress of his experimentation. I don't think work on AGI is useless, but it's not ready for prime time and may never be within our lifetimes.

    If you have to interact with a system once every 28 hours under ideal conditions, then you don't have an autonomous system, regardless of your perception of it. At best, you have a helpful assistant. That is how we should view present AI technology, when it work correctly, which is never guaranteed. The requirement for a ship's crew is not going away any time soon.

    kbd512

    ReplyDelete
    Replies
    1. "The requirement for a ship's crew is not going away any time soon."

      And yet, the Navy is planning on (and already in the act of building) a fleet of 50-100 unmanned vessels in the very near future.

      Delete
    2. The only real value I can see in the near term is for small unmanned vessels that can do harbor defense, or near shore ASW work, or in tandem with larger vessels. Or, vessels that are unmanned but sail with a fleet and can have containerized weapons, or just solid support the fleet will need, that can be transferred by vertrep. In that case, humans are close by to deal with the inevitable problems.

      Delete
  2. The flight of the first manned heavier-than-air powered and controlled aircraft, the Wright Flyer, lasted 3.5 seconds and covered 105 feet. The last flight of the Wright Flyer covered 852 feet in 59 seconds.

    Unmanned ships are going to take some time to work out all the details. I just hope the Navy takes the "build a little and test a lot" approach to their development.

    ReplyDelete
    Replies
    1. "I just hope the Navy takes the "build a little and test a lot" approach to their development."

      That ship has already sailed. The official Navy plan calls for a fleet of 50-100 unmanned vessels in the very near future. You'll recall we committed to a fleet of 55 LCS before the first one was even designed ... and you know how that turned out. We're now doing the same with unmanned vessels.

      Delete
  3. Maybe someone should tell the USN (and other Western navies) that if we want to read naval fiction we'd get more fun out of Patrick O'Brian.

    ReplyDelete
  4. Just wondering, what does a "port visit" by an unmanned ship involve? Without sailors to go ashore, how does one even know it happened?

    ReplyDelete
    Replies
    1. Excellent point! This also suggests that each 'port visit' provided additional human interventions that, I'm sure, weren't counted as such and wouldn't happen in combat.

      Delete
    2. I wonder how much rabble rousing, carousing and bar fights happen in port when it's an unmanned vessel.

      Delete
    3. BA 1959, the amount of rabble rousing etc, it slightly
      less then the median USN ship. The ships are unmanned, meaning they're crewed by women. Now if the ships were unpersoned, a port visit would be to recharge the poly-turbo-encabulator which does nothing for the locals.

      Delete
    4. Two unmanned ships walk into a bar ...

      Delete
    5. "Two unmanned ships walk into a bar..."

      Sir, this blog is serious business. Having already passed 1Apr, I didn't expect to suddenly find myself laughing so hard... ;)

      Delete
    6. Sir! No excuse, sir!

      The unmanned ship says to the doctor, "I think I've got humans ..."

      Why did the unmanned ship cross the road? ... no one knows

      Sir! Still no excuse, sir!

      Delete
  5. I have a solution for unmanned vessels. It's historically proven and well documented.

    McNamara's Project 100,000 describes the use of personnel assessed as unfit for service. If they're unfit for service, then they can't be service personnel, right? So they don't exit on the "unmanned ships."

    Perfect.

    China is already leading the charge, deploying their lowest scoring officers to nuclear submarines and we don't want to be left beh

    (This post is written in despairing sarcasm, in case anyone misses the point)

    ReplyDelete
  6. Someone in the Navy just passed their "Lying with Statistics 102" class with flying colours, I see.

    ReplyDelete
  7. "STEM"

    Comment deleted. We've thoroughly discredited this issue.

    ReplyDelete
  8. There is also the threat that an enemy can take over an unmanned ship, via the airwaves or with commandos. This is already an issue with UAVs. Maybe a traitor could provide the Chinese with the codes.

    ReplyDelete
  9. "Don't bother me with facts. My mind's already made up!"

    ReplyDelete
  10. Here is an interesting interview with a Russian military expert. One-third of the way he addresses problems with the Russian Navy. He says ships were delivered without tested and proven systems. Some Admirals complained and refused to accept them for a few months. When forced to fight in the Black Sea, some systems didn't work at all.

    https://www.bitchute.com/video/sNnBoph5BnCB/

    ReplyDelete
    Replies
    1. I'm unable to find much background on this Maxim Kimov but what I can find and what I hear of multiple interviews of his is that he is badly misinformed about many topics (nuclear intentions of the West/US, for example). He is also, clearly, speculating rather than relating actual facts. All of that makes his statements suspect or, at best, unverified. It doesn't mean he's wrong but, without verification, and with no verifiable history of being correct, he's not a credible source, as it stands. Again, doesn't mean he's necessarily wrong but it doesn't mean he's right, either, and the verifiably incorrect statements he's made further damage his credibility.

      Delete

Comments will be moderated for posts older than 7 days in order to reduce spam.