Can AI understand, know between, or choose between good and evil?

[The Liturgist]

HAL was doing what he was programmed to do. end of. Any sufficiently complex piece of software is buggy, it's just a matter of how many and what kind of problems they cause. HAL's bugs were particularly nasty.

If you're working in an evironment where a software bug has a very real potential to kill its users, you'd like to know that the test plan is extraordinarily rigorous. Just ask Boeing. Their "anti-stall" system on the 737 Max prevented stalls at low altitude by flying the plane into the ground.

There is an interesting scene in 2010 where Dr. Chandra declares that it was Dr. Heywood Floyd’s fault that HAL malfunctioned, by inserting the instruction to not reveal the true nature of the mission until the entire team was assembled, and to lie if necessary, which caused a malfunction known as an “H-Moebius Loop” affecting advanced autonomous goal seeking programs (which reminds me of the hallucination problem experienced by chatGPT and other AIs and other bugs I have experienced, for example, chat GPT sometimes forgets parts of the conversation), and in this interesting scene, Dr. Floyd vehemently denies having given such an order or having knowledge of it.

Now initially, on first viewing, this seemed like a continuity error, since in 2001 there was the famous scene where after singing “Daisy Bell” while being de-activated (the first song sung by a computer, which was spliced together with the voice of Douglas Rain, who did the rest of HAL, for an eerie effect), a video playback is initiated in which Dr. Heywood reveals to the astronauts the true nature of the mission “which until now has been known only by your onboard HAL-9000 computer.” However, there are several scenes cleverly included in 2010 where Dr. Heywood Floyd lies or distorts the truth in subtle manners, as well as suggesting to his successor a lie to tell the president to get him to approve sending himself, Dr. Chandra and Dr. Chernow to Discovery onboard the Soviet spacecraft Alexei Leonov because the orbit of Discovery was deteriorating for unknown reasons.

There is even a scene where as a mere prank to alarm Dr. Chernow, Dr. Floyd fakes having lost the red calculator he would have used to activate a “guillotine” to disconnect HAL surreptitiously in the event HAL threatened the mission (which Dr. Chandra was not informed of, and Dr. Chernow had installed it surreptitously so that Dr. Chandra wouldn’t find it - but Dr. Chandra did find it, exposing Floyd as a liar, which caused Floyd to laugh).

So the end takeaway of that is that Dr. Floyd was the actual villain, not HAL-9000, although less of a true villain and more of a trickster character. Also, we see this in 2001 where he allows the Soviet cosmonauts to believe there has been an outbreak of an alien disease at the US moon base at Clavius, before discussing how this ruse had been “personally and professionally embarrassing” at a meeting with the leadership on Clavius, after being asked how long the cover story would remain in effect. “I don’t know, Bill, your guess is as good as mine”, then after the bit about the embarrassment, he continues with “But I accept the need for absolute secrecy” until they better understood the situation and how to properly disseminate it to the public given the grave potential for “societal upheaval and mass unrest” the discovery of TMA-1 (the Monolith) might cause.

Nonetheless, it is really the case that Dr. Floyd was responsible for the deaths of Dr. Kaminsky, Hunter and Kimball, and for Dr. Frank Poole being released into the vacuum of deep space, where he would freeze and not be revived until the year 3001, which would doubtless have been very inconvenient. On the other hand, we cannot fault Dr. Floyd for the apparent deification of Dr. David Bowman, which was rather an unusual and unplanned outcome of the mission.

 

[trophy33]

It depends on the definition of "understand" and of "good and evil".

In some contexts, they can surely conclude that something is harmful or untrue, i.e. evil. But it will be a computation, not some intuitive "understanding".

And I am talking about machine learning, because no true AI exists, yet.

 

[The Liturgist]

The abilities of AI like chatGPT really are overrated - I was just using it for some editing of liturgical documents, where it can be easier to do something in chatGPT rather than using, for example, regular expressions, albeit less reliable; and tonight I found myself in bug city. Every mistake it could make, from editing earlier versions of the file which it had previously corrected rather than the latest uploaded file, to mysteriously crashing while working on a PDF after initially handling it well, it made. And it required me to hold its hand at every step of the process. And Grok simply has no editing capability of this sort at all as far as I can tell

 

[hedrick]

At the moment, the types of AI you're talking about (large language models) are basically giving you back text they were trained on. It's pretty flexible just how the model structures the text, so it's not just quoting specific things. It's reformulating things to match your question. But the currrent models don't have what we'd consider judgement, an ability to weigh things based on applying critical principles, to decide which of the things it's been told are true. There are attempts to insert reasoning, but it's still not making a moral judgement in the same sense as humans.

There are larger philosopical questions about whether a program can be responsible as a matter or principle. But the current models have not reached that issue.

They can be extremely useful. However you have to temper that by the fact that the accuracy of what they say depends upon the accuracy of what they were trained on. There are ways that the scientists can try to weight things. But there's still a danger that if they're trained on the whole internet, some of that is garbage.

 

[The Liturgist]

At the moment, the types of AI you're talking about (large language models) are basically giving you back text they were trained on. It's pretty flexible just how the model structures the text, so it's not just quoting specific things. It's reformulating things to match your question. But the currrent models don't have what we'd consider judgement, an ability to weigh things based on applying critical principles, to decide which of the things it's been told are true. There are attempts to insert reasoning, but it's still not making a moral judgement in the same sense as humans.

There are larger philosopical questions about whether a program can be responsible as a matter or principle. But the current models have not reached that issue.

They can be extremely useful. However you have to temper that by the fact that the accuracy of what they say depends upon the accuracy of what they were trained on. There are ways that the scientists can try to weight things. But there's still a danger that if they're trained on the whole internet, some of that is garbage.

Indeed; I use them routinely in my work, and I experience enough bugs, I find myself really wishing they were more capable than they actually are.

Trying to get ChatGPT to write an interesting fictional story is like trying to train a cat to sing the parts of Siegfried in Gotterdamerung.

Also, it is very obvious when forum members generate replies using a large language model in a naive manner. Stilted language, lots of bullet points, excessive organization…

 

[timothyu]

It depends on the definition of "understand" and of "good and evil".

Humans can't even agree on that. But yes, it would take a selfish computer to follow the same paths as us.

 

[Jipsah]

Logical deduction

Multiple software bugs. My "mole" in Boeing (yes, I have one) has verified that. That, and the failure of a critical component, the angle of attack sensor, that should have absolutely been redundant but was not. Pennywise and gigabuck foolish, not to mention the cost in human lives and loss of reputation suffered by the company. Aeronautical Engineers should never be, or be influenced by, bean counters.

 

[The Liturgist]

Multiple software bugs. My "mole" in Boeing (yes, I have one) has verified that. That, and the failure of a critical component, the angle of attack censor, that should have absolutely been redundant but was not. Pennywise and gigabuck foolish, not to mention the cost in human lives and loss of reputation suffered by the company. Aeronautical Engineers should never be, or be influenced by, bean counters.

The 737 (and all Boeing narrowbody jetliners going back to the 707) actually has two Angle-of-Attack (AOA) sensors, one of which is connected to the captain’s instrumentation and one to the first officer’s, however, the way MCAS was configured, it was wired to only the Captain’s AOA sensor, which meant a problem with that sensor could cause it to activate spuriously.

There was also an issue with the interval between MCAS resets, which made it extremely difficult to counteract with the trim switches, especially since MCAS would move the horizontal stabilizer faster than the trim switches would in that configuration.

Finally, there was the fact that pilots were not even told about the MCAS system at first, and after the initial LionAir crash in Indonesia, the information disclosed was inadequate.

Historically, before the merger with McDonnell Douglas (where foul ups like this were routine), this sort of thing seldom happened at Boeing, with one exception, that being the 737 rudder hardover malfunction that caused a 737-200ADV and a 737-300 to crash in the late 1980s and early 1990s due to a defective actuator. On the other hand at McDonnell Douglas there is a veritable litany of mistakes like this, going back to the Douglas DC-6 with its defective fuel vent system (which could dump fuel into an inlet valve for the heater causing a fire) and defective fire extinguishers - note that most DC-6 aircraft built were DC-6Bs where this problem and others had been fixed, and the DC-6B was known for reliability, but that was a bit of a humdinger. Then there was the infamous DC-10 cargo door issue, where Convair warned McD that the door handle could be forced causing the door to show as closed on the pilot’s panel, without actually engaging the locking mechanism properly, but McD tried to suppress this memo - when it came to light after the THY Turkish Airlines crash on CDG-LHR near Paris in 1973 it sparked a furor.

There was also the issue with the engine mounting, although to be fair, that was AA’s fault for not following the maintenance instructions from Douglas (United was also deviating from the instructions to speed things up, but was doing so in a less unsafe manner, using a sling rather than a hydraulic lift to prop up the engine while disconnecting it), but McD was definitely at fault for the lack of hydraulic fuses that resulted in the United aircraft crash landing in Sioux City (and just getting the aircraft to crash land so that some of the passengers were able to survive was a heroic accomplishment by the flight crew, who brought the aircraft down with no flight controls other than differential thrust on the wing-mounted engines and a tiny amount of aileron), after the center engine had an uncontained failure (which never happened on any other trijet, since the L1011 Tristar and the Boeing 727 had their engines mounted inside the aft fuselage and connected to the inlet via an S-duct).

Then on the MD-11 there was the defective slat handle position and extreme sensitivity to pitch changes in cruise due to the aft center of gravity that led to many people being severely injured when a copilot on China Eastern inadvertantly activated the slats in 1993, causing the aircraft to enter a sort of fugoid flight like a rollercoaster. Pilots from one major US airline called that aircraft the MD-Lemon (I think it was American, although it may have been Delta, or perhaps both).

When Boeing took over McDonnell Douglas, the Boeing engineering-background managers were forced out by politicking by the McD executives, which is ironic, because McD was struggling when Boeing acquired it, but the people responsible for McD’s woes wound up taking over the superior manufacturer and destroying its engineering-focused culture (a major blow was when corporate headquarters was moved from Seattle to Chicago, away from the engineering department).

 

[The Liturgist]

The 737 (and all Boeing narrowbody jetliners going back to the 707) actually has two Angle-of-Attack (AOA) sensors, one of which is connected to the captain’s instrumentation and one to the first officer’s, however, the way MCAS was configured, it was wired to only the Captain’s AOA sensor, which meant a problem with that sensor could cause it to activate spuriously.

There was also an issue with the interval between MCAS resets, which made it extremely difficult to counteract with the trim switches, especially since MCAS would move the horizontal stabilizer faster than the trim switches would in that configuration.

Finally, there was the fact that pilots were not even told about the MCAS system at first, and after the initial LionAir crash in Indonesia, the information disclosed was inadequate.

Historically, before the merger with McDonnell Douglas (where foul ups like this were routine), this sort of thing seldom happened at Boeing, with one exception, that being the 737 rudder hardover malfunction that caused a 737-200ADV and a 737-300 to crash in the late 1980s and early 1990s due to a defective actuator. On the other hand at McDonnell Douglas there is a veritable litany of mistakes like this, going back to the Douglas DC-6 with its defective fuel vent system (which could dump fuel into an inlet valve for the heater causing a fire) and defective fire extinguishers - note that most DC-6 aircraft built were DC-6Bs where this problem and others had been fixed, and the DC-6B was known for reliability, but that was a bit of a humdinger. Then there was the infamous DC-10 cargo door issue, where Convair warned McD that the door handle could be forced causing the door to show as closed on the pilot’s panel, without actually engaging the locking mechanism properly, but McD tried to suppress this memo - when it came to light after the THY Turkish Airlines crash on CDG-LHR near Paris in 1973 it sparked a furor.

There was also the issue with the engine mounting, although to be fair, that was AA’s fault for not following the maintenance instructions from Douglas (United was also deviating from the instructions to speed things up, but was doing so in a less unsafe manner, using a sling rather than a hydraulic lift to prop up the engine while disconnecting it), but McD was definitely at fault for the lack of hydraulic fuses that resulted in the United aircraft crash landing in Sioux City (and just getting the aircraft to crash land so that some of the passengers were able to survive was a heroic accomplishment by the flight crew, who brought the aircraft down with no flight controls other than differential thrust on the wing-mounted engines and a tiny amount of aileron), after the center engine had an uncontained failure (which never happened on any other trijet, since the L1011 Tristar and the Boeing 727 had their engines mounted inside the aft fuselage and connected to the inlet via an S-duct).

Then on the MD-11 there was the defective slat handle position and extreme sensitivity to pitch changes in cruise due to the aft center of gravity that led to many people being severely injured when a copilot on China Eastern inadvertantly activated the slats in 1993, causing the aircraft to enter a sort of fugoid flight like a rollercoaster. Pilots from one major US airline called that aircraft the MD-Lemon (I think it was American, although it may have been Delta, or perhaps both).

When Boeing took over McDonnell Douglas, the Boeing engineering-background managers were forced out by politicking by the McD executives, which is ironic, because McD was struggling when Boeing acquired it, but the people responsible for McD’s woes wound up taking over the superior manufacturer and destroying its engineering-focused culture (a major blow was when corporate headquarters was moved from Seattle to Chicago, away from the engineering department).

By the way @Jipsah my dear friend I would imagine your friend at Boeing could go into much more detail about what I have discussed than what I mentioned, especially if he worked there in the 1990s on the 777-200 project before the McD merger, because it was that project and its celebrated Working Together management philosophy that old Boeing reached its apex, creating a truly beautiful airliner, the first really good large twinjet that could replace the dangerous DC-10 and the ageing Lockheed L1011 (which was also a beautiful aircraft, ahead of its time, which suffered from inadequate range compared to the DC-10-30, but it deserved to win the trijet wars, but didn’t because of a delay in production due to its engine supplier, Rolls Royce, going bankrupt and being bailed out by the British government, which delayed delivery of the RB 211 engine (which later was refined and became a workhorse which was used on the first Boeing 757-200 aircraft, although the Pratt and Whitney powered 757s are known to be more powerful, and were used by Delta, Northwest and TWA, while Eastern and American used the Rolls Royce variety, and I can’t recall what United, Continental or US Airways used. But you know you’ve gotten into the weeds of avgeekery when the subject of comparative engines comes up, so I shan’t bore you with a deep dive into different engine variants, because even i don’t care that much; I have a young British friend who cares more about the engines than the planes much of the time.

But he missed out on the really cool loud engines like the JT3D and JT8D and Rolls Royce Conway that made that lovely jetblast roaring screaming sound I love to hear, but the uncreative philistine regulators object to because they make too much noise. When it comes to jet engines I like ‘em noisy. When I lived in Las Vegas I was in the flight path for Nellis, which was pure bliss, since my neighborhood was very quiet, and I was serenaded throughout the day by the sweet sounds of military jets flying overhead.

 

[Jipsah]

The 737 actually has two AOA sensors, one of which is connected to the captain’s instrumentation and one to the first officer’s, however, the way MCAS was configured, it was wired to only the Captain’s AOA sensor.

Then I misunderstood my "mole", which isn't unlikely. She and I can speak the same language when it comes to computers, but her knowledge of flying things is, quite properly, light years beyond mine. (She's my youngest daughter, BTW.)

There was also an issue with the interval between MCAS resets, which made it extremely difficult to counteract with the trim switches, especially since MCAS would move the horizontal stabilizer faster than the trim switches would in that configuration.

That's the sort of thing I have to take on faith <Laff>

 

[The Liturgist]

She's my youngest daughter, BTW

Alas that probably means she wasn’t there during the 777-200 development program in the 1990s, and likely her whole career has been spent dealing with a company that would be more accurately called “McDonnell Boeing.”

Then I misunderstood my "mole", which isn't unlikely.

Actually you understood her pretty well, in that the MCAS was non-redundantly connected to just one of the two Angle of Attack sensors, which directly contributed to the failure mode experienced on at least one of the accident flights.

I mean what’s the point of having two sensors if you’re going to connect a safety-critical system to just one of them?

That said, part of the problem was that management and the FAA did not actually regard MCAS as safety critical.

 

[Jipsah]

By the way @Jipsah my dear friend I would imagine your friend at Boeing could go into much more detail about what I have discussed than what I mentioned, especially if he worked there in the 1990s on the 777-200 project before the McD merger

That was way before her time. She started there right out of engineering school, six years ago. She initially worked on the FA-18 before they shipped all that stuff back to St. Louis. They shifted her to FLRAA (aka Defiant, aka "Flora"), and when they lost that, she went to a different project at Vertical Lift.

, because it was that project and its celebrated Working Together management philosophy that old Boeing reached its apex, creating a truly beautiful airliner, the first really good large twinjet that could replace the dangerous DC-10 and the ageing Lockheed L1011 (which was also a beautiful aircraft

I never flew on a DC-10, but I made a few trips on TriStars between St. Louis and LA. The cheapest ticket from Nashville to LA in those days via St.L on TWA.

I shan’t bore you with a deep dive into different engine variants, because even i don’t care that much; I have a young British friend who cares more about the engines than the planes much of the time.

Yeah, you kind of ran off and hid from me there. I know how the high-bypass turbofan engines work, but don't ask me for details. I could probably get Meedum (literally "Faith",the Mole's Korean name; all my girls have one, but they're for family use only) to explain it all to me.

When I lived in Las Vegas I was in the flight path for Nellis, which was pure bliss, since my neighborhood was very quiet, and I was serenaded throughout the day by the sweet sounds of military jets flying overhead.

Wow! My only recollection of real, serious screaming engines overhead was when I was, as best I recall, about 11 years old. I was already a little geekboy then, and I'd pretty much memorized every airplane book (at least the ones that had cool pictures in them) in the library. We'd just come in from the grocery store and were walking across the front yard carrying bags of stuff when this awful howl came from beyond the hills behind the house. We all looked to see what was, when four B-58s came blasting over. It seems to me that they were just above the treetops, but I'm sure it just seemed that way. There were obviously sub-sonic because they didn't boom us, which still happened from time to time in those days. They probably have broken windows, and maybe ear drums, if they'd been going that fast, but they were certainly fast enough you suit me. There and gone in scant seconds. But they were big, they were fast, and they were unmistakable for any other airplane.

 

[The Liturgist]

I never flew on a DC-10, but I made a few trips on TriStars between St. Louis and LA. The cheapest ticket from Nashville to LA in those days via St.L on TWA.

TWA - I wept when they were absorbed into American Airlines. That was a great airline. They had problems in their last few years, which were caused by damage inflicted by Carl Icahn - if it hadn’t been for him looting the company they would probably still be around. And that STL hub was really good - it lacked the overcrowding of ORD (Chicago O’Hare) but benefitted from a similar central position for transcontinental flights. The only better location for a hub in terms of pure geography is ATL insofar as it can capture transcontinental traffic in addition to seasonal leisure traffic between the North and Florida and Texas, and one of the more westerly hubs like PHX or LAS that only rarely has weather related delays.

 

[The Liturgist]

I know how the high-bypass turbofan engines work, but don't ask me for details.

The older engines are just like the high bypass engines only less efficient because there is less bypass or in the case of the turbojets, no bypass. But the turbojets (or “straight pipes”) were lacking in power, and had to inject water into the engines to increase thrust on takeoff, which resulted in smokey contrails which made for some awesome photos.

What really confuses me, and a lot of other people, are the nuances of turboprops, where you have a jet turbine driving a propeller.

 

[The Liturgist]

She and I can speak the same language when it comes to computers,

I suspect you and I would as well, since I’m a systems programmer (specializing in OS development, largely for embedded systems) who has also done some network engineering and syadmining of virtualization, as I had a side gig doing managed hosting.

I particularly love UNIX-like OSes such as the BSDs, Illumos/Solaris and the more traditional Linux distros - not a fan of systemd.

 

[Stephen3141]

You need to define what you mean by "spiritual".
(One of the Greek meanings is "logical", which an AI tool should
be able to (sorta) handle.)

By "AI", I assume that you mean an "AI" software tool.

All software, deals with symbol manipulation. It is questionable whether
ANY software tool could "understand" anything, as a human being would.
This brings up, what you mean by "understand".
(A mathematical calculator can spit out an answer to a lot of math problems,
but that does not mean that it understands any of the philosophical/logical
basis of mathematics.)

The NT authors use vocabulary that is commonly translated as "spiritual" or
"fleshly" or "heavenly" or "earthly" in a number of ways. These usages may
correlate with the TOPIC that is being discussed, or the VIEWPOINT that is
being discussed, and not particulalry whether the one doing the thinking has a spirit.

All Moral-Ethical models deal with what different people think is "right" or "wrong".
Some of these models are (sorta) compatible with Christianity. Many of them are
NOT. Many of the differences deal with different concepts of what is a "virtue", and
what is a "vice". This is again, (sorta) the same thing as defining what "good" means,
and what "evil" means.

It is possible that an AI tool has been "trained" on historical ME (moral-ethical)
models. Although, the answers that I have seen so far from the AI tools, seems
to reflect that they do not know the difference between an ME model, and a "model"
of an aircraft carrier. Nor are most Christians conscious of the historical ME models.
---------- ----------

You need to clarify your initial question.