How Hot is Ukraine Gonna Get?

[sburke]

3 hours ago, The_Capt said:

I found out this spring that my private sector daily rate is 750-1000 per day (depending what we are talking about). Now that is not even close to high end, pretty much just respectable. But of course I am here playing internet reindeer games for free, like a rube.

750-1000 peanuts?  Hell man, I'll toss in a bag for ya!  

[Probus]

Does anyone remember this automatic shotgun weapon system that never quite made it in the early 80s:

 https://en.m.wikipedia.org/wiki/Atchisson_AA-12

Wouldn't one of these loaded with "bird shot" be the perfect weapon for squads to have to help fend off drones in Ukraine?

Edited Friday at 04:04 PM by Probus

[Maciej Zwolinski]

1 minute ago, Probus said:

Wouldn't one of these loaded with "bird shot" be the perfect weapon for squads to have to help fend off drones in Ukraine?

Looks good.

I am wondering about a somewhat related issue. For the purpose of shooting at drones, is it possible to create a simple predictor sight app, using smartphone's camera functionality - the smartphone to be placed over the barrel of PKM or AK, so that one could track the drone for a couple of seconds in flight, and then see the required lead distance & direction indicated on the screen? I have no idea of the physics involved, but am routinely surprised with new wonders that the smartphones can do, so why not this?

[Artkin]

IIRC the problem with these automatic shotguns was that the shells often melted inside the chamber. 

Edited Friday at 04:30 PM by Artkin

[ArmouredTopHat]

Some good news at least

[The_Capt]

1 hour ago, sburke said:

750-1000 peanuts?  Hell man, I'll toss in a bag for ya!  

Clearly I got into the wrong business. This is a lot more than I am being paid now, I can tell you that.

[dan/california]

2 hours ago, Battlefront.com said:

I agree.  Due to the rapid expansion of munitions production across a broad spectrum of types and countries, it's just as likely that these are accidents related to new staff, equipment, procedures, etc. being rushed into place.

That said, I am sure some of these things are not accidents.  There's some evidence of this in Bulgarian and Czech Republic IIRC, there are intercepts that the Russians are interested in these sorts of ops, they certainly are capable of doing them, and it's logical for them to do this sort of thing.

Just a couple pages ago I said we should be prepared for Russia tilting more towards terrorist state behavior as things get worse and worse for them.  Sabotage related to Ukraine aid should be expected.

Steve

 

Quote

 

https://www.bbc.com/news/articles/ce5564x3167o

French police arrest Russian-Ukrainian man on bomb-making charges

 

This is the the most unequivocal case of a Russian agent getting caught. But there has been very little information since the initial report. I can't quite decide if that is because he didn't even know the name of the guy that paid him, or if has proven so valuable they are trying to make everybody forget about him. 

Edited Friday at 05:10 PM by dan/california

[A Canadian Cat]

22 hours ago, dan/california said:

Quote

What’s That Bird Song? Merlin Bird ID Can Tell You

 

Yeah that tech is already moving right along

It is amazing and super useful. It's not perfect though. In fact, it ties into e-bird which is for reporting bird sightings. However when you do mark a bird sighting down from Merlin sound ID it asks you to verify that you actually saw the bird and confirmed the sighting.

I have had a few times where Merlin has told me a super rare wren that I have never personally seen was the source of the sound. I have not been able to actually find the damn thing. I think there is a even chance that it is just not there at all and the AI is crossing a couple of things up.

Don't get me wrong the Merlin Sound ID is awesome and I am using to help me get better at recognizing bird song. It's just not perfect.

[dan/california]

21 minutes ago, A Canadian Cat said:

It is amazing and super useful. It's not perfect though. In fact, it ties into e-bird which is for reporting bird sightings. However when you do mark a bird sighting down from Merlin sound ID it asks you to verify that you actually saw the bird and confirmed the sighting.

I have had a few times where Merlin has told me a super rare wren that I have never personally seen was the source of the sound. I have not been able to actually find the damn thing. I think there is a even chance that it is just not there at all and the AI is crossing a couple of things up.

Don't get me wrong the Merlin Sound ID is awesome and I am using to help me get better at recognizing bird song. It's just not perfect.

Perfect is not the relevant standard in a full up shooting war. Ukraine might actually be the most complicated case for tech like this though. Both sides are fielding large numbers of the same Soviet vehicles, and speaking similar languages. So IFF by some other method would be needed. If your in the command center at firebase nowhere, it is 0300, and the system flags people speaking Pashtu in a ditch full of cold water that runs in a tactically convenient spot....

[poesel]

20 hours ago, Vanir Ausf B said:

Lithuania is the head-scratcher. A country with no tanks decides in 2023 it wants tanks. Either they know something we don't or we know something they don't.

Lithuania also asked and got a German tank brigade stationed in Lithuania. Buying a few German tanks may have helped that decision.

[Vanir Ausf B]

Could be. Then again, they also bought 91 Boxer IFVs in 2016 and 21 PzH 2000s in 2015 without any brigades in return

[A Canadian Cat]

40 minutes ago, dan/california said:

Perfect is not the relevant standard in a full up shooting war. Ukraine might actually be the most complicated case for tech like this though. Both sides are fielding large numbers of the same Soviet vehicles, and speaking similar languages. So IFF by some other method would be needed. If your in the command center at firebase nowhere, it is 0300, and the system flags people speaking Pashtu in a ditch full of cold water that runs in a tactically convenient spot....

LOL nice example. Yeah I almost added a bit about perfect not being required for this. I should have.

Clearly taking out someone that absolutely should not be there at zero dark 30 in a war zone is perfectly fine - adding a rare bird sighting to a database used by researchers probably isn't. It's just not the same - may bad.

AI is a bit of a buggabo for me. Yes, it is useful. I'm not suggesting we not use it. My issue is people just don't understand it's strengths and weaknesses and where it really should not be deployed. Hell experts don't get it half the time either. I mean my god we all experienced low level bureaucrats saying "the computer says I cannot do that, so I can't fix that for you" and that was before people believed AI had everything figured out and knows more then we humans do. We are in for a long road of frustration in more places than we currently expect.

I just found out about this: https://www.vice.com/en/article/v7v5xb/a-human-amateur-beat-a-top-go-playing-ai-using-a-simple-trick

We are sort of off topic but the fundamental issue here is some researchers discovered (with the help of another AI for extra irony) that the Alpha Go program don't actually understand the fundamentals of what playing go means. This fundamental short coming can be used by amateur level players to defeat Alpha Go 14 times out of 15 tries. Oops. or Yay.

Why does this matter? Good question. We were lead to believe that Alpha Go learned to play Go and it had learned in a deep way such that it understood. It did learn how to play (clearly) but it did not understand what it was doing. That's a big deal. It means that even creating an AI with a narrow task still does not have understanding. Not enough people actually understand that.

To tie that back to this war or war in general it just reinforces we have to be careful about how we deploy this stuff.

Swarm of AI drones released into a kill box that a human commander determined was appropriate - got for it.

Eye in the sky drone with AI looking for bad guy "ABC" using gait tracking and facial recognition with shoot to kill permission roaming the globe - bloody hell do not let that happen.

Edited Friday at 07:57 PM by A Canadian Cat
spelling or typing who knows :-)

[zinz]

Quote

1/ Badly wounded Russian soldiers, some on crutches, are being sent to fight in Ukraine. Russian milbloggers say it is because of huge losses and shortages of personnel, as well as bureaucratic mismanagement and the military's culture of lying to superiors. 

2/ The Russian milblogger Anastasia Kashevarova (a former adviser to State Duma Speaker Vyacheslav Volodin and then to the LDPR party) has posted an angry denunciation of the army's treatment of the men of the 26th Tank Regiment, based in Mulino in the Nizhny Novgorod region.

A recent video shows visibly injured men on crutches pleading to be taken out of the battlefield. According to Kashevarova, they are from the 26th Tank Regiment of the 47th Guards Tank Division.

3/ The men, who number about 50, address their appeal to Putin and the military prosecutor's office. They say that they are wounded, still using crutches and plaster casts, have not been fully treated and have not undergone the required rehabilitation or 45-day medical leave. Nonetheless they are still being sent on combat missions.

4/ They complain that they are wounded, still using crutches and plaster casts, have not been fully treated and have not undergone the required rehabilitation or 45-day medical leave. Nonetheless they are still being sent on combat missions. They say that field doctors are "shrugging their shoulders" at the situation, which they blame on their commanders. They are currently in the second line but say that they will be sent to the front line in a few days' time.

5/ Kashevarova says this is a systemic problem which has been going on for "quite a long time". She writes:

6/ "This is not a problem with the medics and the military-medical commission (although such problems also exist, I am not hiding it), this is a problem with specific commanders who don't care about doctors' instructions, referrals to hospitals and planned operations - they load guys into cars by deceit (assuring them that they are taking them to the command for some documents and leave) and send them to the front. Why? What will they do there?"

7/ "What is this image of turbulent activity? What incredible value does a fighter on crutches have on the front line? I understand, maybe not feeling sorry for specific people, but turn on logic, turn on practical considerations: these people on the front line are simply useless! Give them a chance to heal, to rest, and they can be useful. What you are doing now is just a useless, cynical waste of people."

8/ Kashevarova says that she is making a complaint to the military prosecutor's office and she has "a record of which of them file complaints, so unscrupulous commanders will not be able to quickly send them to the front line and hide the traces of their dubious activities".

9/ Other Russian milbloggers have also commented on the video. The pseudonymous Vault 8 calls it "the most vile and treacherous phenomenon, undermining the authority of the command and the RF Armed Forces as a whole in our own eyes as servicemen – when individual butcher-commanders scoop out from permanent posts and hospitals everyone who is destined for a long recovery and simply send them for useless disposal."

10/ He asks "how can you trust commanders if:

- You were told that there would be no mobilisation – and you were mobilised.

- They freeze you in place without a reasonable term of service.

- You get seriously wounded in this war...

- And not only are you not demobilised, but in some cases you are sent to your death without being fully healed."

11/ "This practice of individual commanders and chiefs of staff must be firmly stopped, because so far it is the most treacherous and treacherous blow (one can't call it anything else but treachery) against our [soldiers] from their own side."

Natalia Kurchatova comments that "the situation with neurological problems is especially difficult, when, for example, a person’s arms and legs are formally in place, but they do not function properly."

12/ "In my memory, they tried to send a volunteer to the front, who had shrapnel that injured the spinal cord and was stuck in the spine: his legs were paralysed, his blood pressure was going through the roof, and he was told, roughly speaking: go to the front with your shrapnel and don’t show off. It was a whole saga just to remove this shrapnel from the person."

13/ Svyatoslav Golikov blames "crooked bureaucratic collisions. Semi-disabled casualties continue to remain in service due to the current prohibition on discharge, minus exceptional circumstances that still have to be justified, including obtaining the "D" category of fitness. In addition, by remaining in a military unit, a person receives medical treatment, whereas in the case of dismissal, the prospect is blurred. In fact, such people are not able to fulfil their duties properly."

14/ "Furthermore, the early discharge of untreated servicemen is directly affected by hospitals being overloaded. At the same time, people, including those in fitness category "G", tend to be returned to the places of temporary deployment of units in the Special Military Operation zone, which in itself is not very favourable, plus all the facilities of the nearby rear are in the kill zone of enemy remote high-precision weapons."

15/ "Finally, due to excessive losses in active units engaged in active combat operations, the number of personnel (primarily assault units) regularly drops, which cannot be promptly replenished by marching reinforcements [to the front]. Hence the practice of early return of people to the front line on the principle of 'everyone who can somehow hold a weapon'."

16/ "It is quite natural that this practice does not contribute in any way to improving the effectiveness of combat work and motivation of the personnel."

17/ Kashevarova attributes the situation to "false reports that hide the real balance of forces and the number of personnel. Because of reports that everyone went on leave, that more than 500,000 contractors and volunteers came (although we remember how contractors were recruited – they forced mobilised people to sign), because of the real shortage of fighters at the front."

18/ "Will any of the military experts tell us – how many people are in a platoon now? How many are in a company? A battalion? How many are in a regiment, and how many are in a division? We can have a platoon of 3 people, and a hundred in a regiment, and in a division less than a thousand. But in reports they do not write real numbers, they write battalions/brigades/divisions, and no one goes into detail about the fact that the name does not reflect the number [of personnel]."

19/ She suggests that the new leadership of the Russian Ministry of Defence has found itself in a "soap bubble" and does not know what to do about the situation. The underlying cause, she says, is a chronic shortage of manpower, which she blames on the following:

"1. Because at the beginning of the Special Military Operation and for a long time after, the generals and leadership of the Ministry of Defence believed that drones were not serious. And we suffered losses in personnel and equipment."

20/ "2. Because they sent untrained people to the front who could not shoot. Normal commanders tried to save them and they themselves died.

3. Because many assaults take place without artillery support. And somewhere the artillery strikes at its own people, and the tankers do not take into account the terrain and also hit their own side from the hilltops."

21/ "4. Because professionals – artillerymen, tankers, UAV operators, electronic warfare officers, air defense officers, surgeons and so on, were sent on stupid assaults on villages.

5. Because an unreasonable, unprofessional approach leads to the fact that not only untrained, but really tough warriors die in vain. Those who can and want to fight, and can lead people and teach them."

22/ "6. Because the majority of the high command is not even sitting on the territory of the Special Military Operation, but somewhere in Rostov and does not know what is going on."

As a result, she says, "commanders on the ground are receiving reinforcements on crutches, and they don’t know what to do about it. And according to [false] reports, their reinforcements are combat-ready. And a good commander and cripples go on an assault, what is their survival rate?"

23/ "I am not writing about henpecked husbands and cowards, about mama's boys, who have snot hanging down and whine that they cannot fight. I am writing about the facts of sending disabled people to the front, who are on crutches, whose arms and legs do not bend. There are fractures that have not healed, and so on. We have one in captivity with broken legs, because he was sent to the front untreated." /end

Sources:

https://t.me/akashevarova/7385

https://t.me/akashevarova/7388

https://t.me/philologist_zov/1075

https://t.me/vault8pro/51394

https://t.me/vozlevoiny/5434

https://mastodon.social/@ChrisO_wiki/112734936137372372

Russian manpower doesn't look too rosy either 

[zinz]

Quote

1/ After Chinese buggies, the next innovation in Russian battlefield transportation may be electric scooters. Russian MP Maxim Ivanov has proposed putting Russian soldiers on scooters, which he says would also help to rid Russian cities of unwanted scooter riders.

2/ Ivanov, a member of Putin's ruling United Russia party, says that Russian soldiers fighting in Ukraine have requested electric scooters to transport themselves between positions. According to Ivanov, one unit's commander has requested electric scooters for members of a grenade launcher platoon so that they can "silently race between positions." A group of minelayers has also requested scooters to transport anti-tank mines, carrying up to 4 per scooter.

3/ Ivanov says there are too many rental scooters in Yekaterinburg; some could be sent to the military. "Everyone wins. Pedestrians in the cities will breathe easy, because there will be fewer daring scooter riders. And the guys in the Special Military Operation will benefit."

4/ The Russians have been using motorcycles and Chinese-made Desertcross all-terrain vehicles to carry troops rapidly across the battlefield, but these have proved vulnerable to drone attacks. Electric scooters are much slower and less stable on rough ground, so it remains to be seen whether they can be effective. /end

Sources:

https://t.me/maksim_ivanov_deputat/1980

https://www.kommersant.ru/doc/6793361

https://mastodon.social/@ChrisO_wiki/112715687961441423

After golf carts we are now at electric scooters... 

[dan/california]

25 minutes ago, A Canadian Cat said:

LOL nice example. Yeah I almost added a bit about perfect not being required for this. I should have.

Clearly taking out someone that absolutely should not be there at zero dark 30 in a war zone is perfectly fine - adding a rare bird sighting to a database used by researchers probably isn't. It's just not the same - may bad.

AI is a bit of a buggabo for me. Yes, it is useful. I'm not suggesting we not use it. My issue is people just don't understand it's strengths and weaknesses and where it really should not be deployed. Hell experts don't get it half the time either. I mean my god we all experienced low level bureaucrats saying "the computer says I cannot do that, so I can't fix that for you" and that was before people believed AI had everything figured out and knows more then we humans do. We are in for a long road of frustration in more places than we currently expect.

I just found out about this: https://www.vice.com/en/article/v7v5xb/a-human-amateur-beat-a-top-go-playing-ai-using-a-simple-trick

We are sort of off topic but the fundamental issue here is some researchers discovered (with the help of another AI for extra irony) that the Alpha Go program don't actually understand the fundamentals of what playing go means. This fundamental short coming can be used by amateur level players to defeat Alpha Go 14 times out of 15 tries. Oops. or Yay.

Why does this matter? Good question. We were lead to believe that Alpha Go learned to play Go and it had learned in a deep way such that it understood. It did learn how to play (clearly) but it did not understand what it was doing. That's a big deal. It means that even creating an AI with a narrow task still does not have understanding. Not enough people actually understand that.

To tie that back to this war or war in general it just reinforces we have to be careful about how we deploy this stuff.

Swarm of AI drones released into a kill box that a human commander determined was appropriate - got for it.

Eye in the sky drone with AI looking for bad guy "ABC" using gait tracking and facial recognition with shoot to kill permission roaming the globe - bloody hell do not let that happen.

AI definitely has a brilliant except when it is terrible problem. Given that there are many tens billions of dollars, if not more, to be made solving that problem, i expect it will probably get better.Form a military perspective the hardest conversation of the next ten years is what do we have to keep a man in the loop for? There are a lot of areas where it simply won't be possible, because the whole thing happens at speeds that are just a blur to a person. Trying keep a person in the loop in these areas is a stone cold loser. But in a lot of other areas we will need very highly trained people keeping track of what is going on and making sure the computer is making sense. Getting this right is going a to hard and important.

I have been playing a bit of a another game. I won't irritate Steve by naming it, just to take my mind of the news.  The hardest part is just trying to keep track of what is going on. I over focus on one unit and get beaten somewhere else. Now partly that just demonstrates that I am old and slow. But everybody has cognitive limits, system designers REALLY need to think about that.

Edited Friday at 08:22 PM by dan/california

[acrashb]

5 hours ago, Maciej Zwolinski said:

Looks good.

For the purpose of shooting at drones, is it possible to create a simple predictor sight app, using smartphone's camera functionality - the smartphone to be placed over the barrel of PKM or AK, so that one could track the drone for a couple of seconds in flight, and then see the required lead distance & direction indicated on the screen?

It's already been done - TrackingPoint.  Although the tech was, IMO, premature, it proves the concept.  And apparently worked/works very well.
 

[zinz]

https://meduza.io/en/feature/2024/07/05/a-new-estimate-from-meduza-and-mediazona-shows-the-rate-of-russian-military-deaths-in-ukraine-is-only-growing

Quote

In any case, our estimate suggests that since the beginning of 2024 — or, more precisely, in the 173 days between the two estimates — the Russian military’s losses have increased by 39,000 (or between 34,000 and 46,000). Even if we take the minimum possible value, these figures are substantial; they show that Russia’s losses have risen significantly in recent months. This is nearly double the average casualty rate observed in the last three months of 2023. At that time, we estimated it at about 120 deaths per day, whereas right now there are about 200—250 deaths per day.

The difference in daily casualty rates reflects the development of Russia’s large-scale offensive. It began last winter with the battle for Avdiivka and continued after the city’s capture by Russian forces in February, with further attacks across the entire front. Because of the time it takes to find and verify individual deaths, we don’t know how this rate has changed in the last few weeks, but Russia’s total number of recent losses undoubtedly exceeds even those from the Battle of Bakhmut, where about 21,000 soldiers were estimated to have died.

Meduza estimates around 200-250 Russian deaths a day currently. The Ukrainian reported casualties hovering a bit over a thousand a day in the last month sound quite plausibel in comparison. 

https://lookerstudio.google.com/reporting/dfbcec47-7b01-400e-ab21-de8eb98c8f3a/page/p_wdrgjv1iyc?s=ouwAsvJ07N8

(1156 deaths on average in the last 7 days) 

[ArmouredTopHat]

I thought the scooters were just a meme...

[zinz]

6 minutes ago, ArmouredTopHat said:

I thought the scooters were just a meme...

Well if the roads are good enough or makes sense to be faster with moving things from one trench to the next. I imagine the "frontline" from the first trench to a somewhat safer area to be at least 5 to 10 km deep. That is quite some distance to get water food and supplies on foot. So even if the scooters are not used for attacking they make sense as a form of an ersatz battle taxi. 

[LongLeftFlank]

12 hours ago, The_Capt said:

 

 

19 hours ago, Battlefront.com said:

 

 

18 hours ago, ArmouredTopHat said:

A little unfair to imply that was all I was saying, but fair enough. I was getting fed up with the way some people were debating at this point anyway.

The first Persian King who played the game of 'King' (Shach) was the Persian Pad-Shah Shapur II, who was taught it by his Indian wazir.

The wazir was the better chess player, but the King was always the winner of the game.
The King of Kings attained victory by the ingenious device of overturning the chessboard at a crucial point of the game and declaring himself winner.

This showed an imagination of the sort that the wazir did not have; and it was for this reason that Shapur was the King of Kings, and the wazir would never be anything but wazir.
 

Edited Saturday at 12:43 AM by LongLeftFlank
Freeking editor has gone nuts

[ASL Veteran]

3 hours ago, acrashb said:

It's already been done - TrackingPoint.  Although the tech was, IMO, premature, it proves the concept.  And apparently worked/works very well.
 

I believe the IDF currently uses some sort of tracking sight for their rifles.  It's my understanding that it won't allow the trigger to be pulled unless the sight thinks you will score a hit although I think it still allows for non-accurate fire if the shooter prefers that.  They say it's good for letting normal rifles hit drones out to about 100 meters, but after that it gets a little unreliable.  A quick search didn't come up with anything though so I can't link to anything and I'm just going off memory.

[chris talpas]

Good overview video

 

[chrisl]

7 hours ago, A Canadian Cat said:

LOL nice example. Yeah I almost added a bit about perfect not being required for this. I should have.

Clearly taking out someone that absolutely should not be there at zero dark 30 in a war zone is perfectly fine - adding a rare bird sighting to a database used by researchers probably isn't. It's just not the same - may bad.

AI is a bit of a buggabo for me. Yes, it is useful. I'm not suggesting we not use it. My issue is people just don't understand it's strengths and weaknesses and where it really should not be deployed. Hell experts don't get it half the time either. I mean my god we all experienced low level bureaucrats saying "the computer says I cannot do that, so I can't fix that for you" and that was before people believed AI had everything figured out and knows more then we humans do. We are in for a long road of frustration in more places than we currently expect.

I just found out about this: https://www.vice.com/en/article/v7v5xb/a-human-amateur-beat-a-top-go-playing-ai-using-a-simple-trick

We are sort of off topic but the fundamental issue here is some researchers discovered (with the help of another AI for extra irony) that the Alpha Go program don't actually understand the fundamentals of what playing go means. This fundamental short coming can be used by amateur level players to defeat Alpha Go 14 times out of 15 tries. Oops. or Yay.

Why does this matter? Good question. We were lead to believe that Alpha Go learned to play Go and it had learned in a deep way such that it understood. It did learn how to play (clearly) but it did not understand what it was doing. That's a big deal. It means that even creating an AI with a narrow task still does not have understanding. Not enough people actually understand that.

To tie that back to this war or war in general it just reinforces we have to be careful about how we deploy this stuff.

Swarm of AI drones released into a kill box that a human commander determined was appropriate - got for it.

Eye in the sky drone with AI looking for bad guy "ABC" using gait tracking and facial recognition with shoot to kill permission roaming the globe - bloody hell do not let that happen.

"Explainability" is a thing in AI.  Or more appropriately, Machine Learning.

When I use ML to act as a "remote graduate student" it evaluates a bunch of properties of what it's looking at that are based on real physical models, and then compares what it sees for any data set to other real human-evaluated data sets.  It then reports that "this new thing is x% like set 1, y% like set 2, etc" to make clear why it's making some recommendation.  What most of the AI chat bots (and apparently go bots) isn't this.

What a lot of the AIs that take enormous training data sets are doing is just picking the most likely move/phrase/sentence/whatever based based on a distillation of some enormous number of games.  Go is a combinatoric nightmare, so the computer has to constrain how far it looks around any point, so it makes sense that a big non-local surround would fake it out.  So it has no model, or an incomplete model.  All the ChatAIs are basically just "fits" to all the garbage typed into the internet, so they come up with some spectacularly bad "hallucinations".  It's because there's no actual model of reality behind them.

[dan/california]

3 hours ago, chrisl said:

"Explainability" is a thing in AI.  Or more appropriately, Machine Learning.

When I use ML to act as a "remote graduate student" it evaluates a bunch of properties of what it's looking at that are based on real physical models, and then compares what it sees for any data set to other real human-evaluated data sets.  It then reports that "this new thing is x% like set 1, y% like set 2, etc" to make clear why it's making some recommendation.  What most of the AI chat bots (and apparently go bots) isn't this.

What a lot of the AIs that take enormous training data sets are doing is just picking the most likely move/phrase/sentence/whatever based based on a distillation of some enormous number of games.  Go is a combinatoric nightmare, so the computer has to constrain how far it looks around any point, so it makes sense that a big non-local surround would fake it out.  So it has no model, or an incomplete model.  All the ChatAIs are basically just "fits" to all the garbage typed into the internet, so they come up with some spectacularly bad "hallucinations".  It's because there's no actual model of reality behind them.

If I can attempt to refine this just a little bit, the older style of machine learning can give answers that very likely to be correct in their are of expertise. But those areas are narrow, and they take an enormous amount of carful refinement. I would argue that modern commercial jet auto pilots/flight control systems  are a version of this. Unfortunately Boeing has given us some epically bad examples of what happens when you try to half bleep this approach. 

The new style AI as Chris said is really just an enormously complicated pile of statistics that guesses at the most likely "average answer" to your question. Its guesses are only as good as its input data base. Since three quarters of the internet is questionable at best, models trained on it are a neat parlor trick, but almost useless if you need to trust the answers.

Having said that the approach can be extremely powerful when used on narrower, validated sets of information. Reading mammograms is the most common example, but there are a ton of other medical ones. A very interesting study just came out that strongly suggested AI could predict cardiac risk five and ten years out from a simple chest X-ray. Google's protein structure prediction work deserves a Nobel Prize, it completely revolutionized several scientific fields. We are just starting to feel that one.

I do have an actual relevant point...this war has generated more images of actual combat than all previous wars put together possibly. Many of those images and videos even have pretty good labeling. That is going to be used to create a database that can be used to train military AIs that I think we be rather useful. It should certainly help make some amazing games once the tech filters down.

[chrisl]

48 minutes ago, dan/california said:

If I can attempt to refine this just a little bit, the older style of machine learning can give answers that very likely to be correct in their are of expertise. But those areas are narrow, and they take an enormous amount of carful refinement. I would argue that modern commercial jet auto pilots/flight control systems  are a version of this. Unfortunately Boeing has given us some epically bad examples of what happens when you try to half bleep this approach. 

The new style AI as Chris said is really just an enormously complicated pile of statistics that guesses at the most likely "average answer" to your question. Its guesses are only as good as its input data base. Since three quarters of the internet is questionable at best, models trained on it are a neat parlor trick, but almost useless if you need to trust the answers.

Having said that the approach can be extremely powerful when used on narrower, validated sets of information. Reading mammograms is the most common example, but there are a ton of other medical ones. A very interesting study just came out that strongly suggested AI could predict cardiac risk five and ten years out from a simple chest X-ray. Google's protein structure prediction work deserves a Nobel Prize, it completely revolutionized several scientific fields. We are just starting to feel that one.

I do have an actual relevant point...this war has generated more images of actual combat than all previous wars put together possibly. Many of those images and videos even have pretty good labeling. That is going to be used to create a database that can be used to train military AIs that I think we be rather useful. It should certainly help make some amazing games once the tech filters down.

one minor quibble - if you're referring to MCAS with Boeing:  that wasn't even ML as far as I can tell.  It was a deterministic control law that got initially approved in an implementation that was probably not too horrible, then Boeing had bad configuration management and didn't review ir when it got updated to apply more extreme corrections where they would have found that it had fatal edge cases.  Bad system engineering and configuration management.