The Case for HAL's Sanity

http://www.visual-memory.co.uk/amk/doc/0095.html

by pmcpinto • 9 years ago

89 44 9 years ago

visual-memory.co.uk

dwringer• 9 years ago

I didn't think there was any question; HAL was simply following the orders of Dr. Langley.

nraynaud• 9 years ago

I think those 2 videos are on the topic:

- AI "Stop Button" Problem https://www.youtube.com/watch?v=3TYT1QfdfsM (the AI won't let you push the stop button, because that would impair its function)

- General AI Won't Want You To Fix its Code https://www.youtube.com/watch?v=4l7Is6vOAOA (the AI won't let you patch, because changing its goal means it can't reach it)

grondilu2• 9 years ago

This second video is of particular interest to me, as it tackles a problem I've been thinking quite a lot lately : is self-modification compatible with volition? What's the point of wanting something if you can chose what to want? The pill that makes you love the idea of killing your kids is an extreme example, and probably no-one would accept to take this pill, but it's really not so obvious why. Or at least, the reasons are pretty deep.

BTW IMHO a simpler example would be a meta-suicide pill : instead of being a poison that would just kill you, it's a pill that would make you want to kill yourself. I suppose you can make people want to kill themselves by inflicting them an awful chronic pain such that death would look like the only solution, but the pill I'm thinking about would not do that : it would change your mind so that you would want to die just because you like the idea of dying, for no particular reason.

maxerickson• 9 years ago

Are you implicitly arguing that humans going about their day to day lives have a self that they are incapable of modifying?

What's the difference between a pill and an idea?

algorithmsRcool2• 9 years ago

If this is an interesting topic, give "Permutation City" by "Greg Egan" a read. It involves a digitized character that intentionally modifies itself in order to avoid unhappiness.

garyrob2• 9 years ago

"is self-modification compatible with volition?" That's a plot point of WestWorld. A more interesting TV series than I expected.

vl• 9 years ago

In addition to these, Nate Soares: "Ensuring Smarter-than-Human Intelligence has a Positive Outcome"

https://youtu.be/dY3zDvoLoao

talks about how current AI fears misrepresent the problem and gives multiple examples of difficulty of defining correct objective function. I enjoyed this explanation a lot!

grondilu2• 9 years ago

The part I preferred is this one:

HAL has to be extremely careful before committing the first murder. It is possible that he wasn't entirely familiar with the EVA procedure, it being one of the few tasks on board the Discovery that does not require his participation. Perhaps he needed to observe a dry run to study the astronaut's actions and calculate when he would be most vulnerable.

Analemma_• 9 years ago

I thought the canon explanation was that HAL wasn't "insane", per se, but that he had been given conflicting orders (keep the Jupiter monolith secret, but also assist the crew in every way) and was resolving them as best he could.

LoSboccacc• 9 years ago

yeah the novel goes and "solve" the misery. with the task of saving the mission, not lying to the humans and withheld information about the alien, the only logical step to resolve the constraint when the human ask about the mission nature is to kill them to prevent having to lie or having to divulge information.

this also echoes many Asimov novel in principle and especially one in specific where the ai goes trough all rules and the only non conflicting way to solve the constraint was self destruct (cant remember the title)

wruza2• 9 years ago

Even macos autolayout subsystem (cassowary solver) can break constraints based on priority as part of regular operation. It is a linear/logical problem, strange that scifi robots fail in it. Probably scifi authors are not aware of non-strict problems.

No-lie should have minimal priority possible, otherwise HAL breaking TLoR is only a matter of time.

qu4z-2• 9 years ago

Are you thinking of Liar![0]?

Looking it up on the Wikipedia, the See Also section includes a reference to HAL.

[0] https://en.wikipedia.org/wiki/Liar!_(short_story)

phkahler• 9 years ago

Hal was in a double bind: https://en.wikipedia.org/wiki/Double_bind

And given that this is a now obsolete explanation for schizophrenia it makes perfect sense that they'd use it in a movie back then. I can also attest to being caught in one and that it can be enormously stressful - they are also more difficult if you are prone to all-or-nothing or black and white thinking, so again the traditional notion of computer logic would be more prone to it.

twoodfin2• 9 years ago

There is no "canon" explanation for the film, since Kubrick did not provide one, and 2001 was, at heart, his movie, independent in significant respects from Clarke's novelization and later sequels.

That being said, I think the onscreen evidence for Clarke's version is compelling. Most notably, it solves the mystery highlighted by the article as to why HAL didn't kill Bowman on the first EVA: If the computer could have engineered a plausibly innocent loss of contact with Mission Control, he could have acted freely to preserve both the astronauts' ignorance and the success of the mission without worry of being contradicted by the ground-based twin computer (which he surely knew would eventually diverge from his decision-making, having not been informed about the Monolith).

Once the astronauts were told of the divergence and took it seriously enough for their discussion in the pod, HAL had no remaining options.

munin• 9 years ago

I recently re-read 2001 and there was a forward by Clarke that explained something I hadn't known before - the screenplay and the book were developed simultaneously, in a joint collaboration between Kubrick and Clarke. Kubrick wanted Clarke to come up with an epic space story and put it on the screen, but didn't want to burden Clarke with having to write a screenplay.

So I don't think you can really call the book a "novelization" because both worked on it.

SilasX3• 9 years ago

There is no "canon" explanation for the film, since Kubrick did not provide one, and 2001 was, at heart, his movie, independent in significant respects from Clarke's novelization and later sequels.

If you reject the canonicity of Clarke's novel, then everything after the "Jupiter and beyond..." card is nonsensical. What is the viewer supposed to think about the space baby that's actually grounded in something within the movie?

(Edit: Ditto for the aging sequence. I know that you can fit the events into the "alien zoo" framing, but since the film doesn't direct us to that explanation, and if the novel doesn't count, it can't be any more than a fan theory.)

Incidentally, that's why I think the film went sharply downhill after that card.

randcraw• 9 years ago

Clarke addressed this question directly in his sequel "2010", in which Dr Chandra describes Hal's descent into madness as a "Hofstadter-Moebius loop", in which a lie from a trusted authority figure (the US government) contradicted HAL's instructions from another trusted authority, the crew, which introduced an insoluble logical impasse, resulting in a breakdown in HAL's executive function -- a homicidal level of frustration.

https://en.wikipedia.org/wiki/HAL_9000#2010:_Odyssey_Two

grabcocque• 9 years ago

Wait, is it commonly accepted that HAL was mad?

I had always felt the film made perfectly clear HAL was acting totally rationally within several conflicting sets of mission parameters and moral codes.

taneq• 9 years ago

Not even conflicting orders. I thought it was explicitly stated (towards the end of the book, maybe? It's been years since I read it) that HAL had been ordered by mission command to ensure the success of the mission at the expense of the crew if necessary. As crew members started acting suboptimally, HAL realised they were already jeopardizing the mission just by being fallible humans, and started to fix the problem. An example of the law of unintended consequences, maybe, but not a malfunction.

the_af2• 9 years ago

In the book, sure. In the movie this isn't at all clear. Or rather: it's evident HAL considers himself essential to the mission; it's also evident when they threaten to disconnect him, he can be considered to act in defense of the mission itself. This is entirely rational.

However, the movie implies HAL has a real problem. I don't know if actually "insane", but he incorrectly predicts the malfunction of the antenna, and when confronted about it, he cannot say much more than "oops!". I don't buy this article's explanation of why he could have faked his analysis of the antenna -- going by the movie itself, it's clear to me HAL makes a mistake (maybe because he is conflicted about the contradictory orders). HAL makes a mistake for which there is no easy way to backpedal without looking faulty.

Also, in the movie (and this is explored more in "2010: The Year We Made Contact"), it's clear HAL-class computers are capable of irrational, human-like thought. HAL feels threatened by disconnection. He sounds very scared when Dave Bowman finally disables him ("Stop, Dave. I'm afraid"). A similar computer on Earth, SAL, is similary worried about temporary disconnection and asks "will I dream?". And in the final scene of 2010, Doctor Chandra must be both gentle to HAL and appeal to his sense of duty for him to follow orders.

I think there's a very good case that HAL did, indeed, go insane. Even if it was because of contradictory orders.

xefer2• 9 years ago

I think Kubrick implies it by the way HAL's voice noticeably slows significantly and changes his tonality to suggest he was at least deeply troubled

xefer2• 9 years ago

I had always assumed that Kubrick gave an explicit clue that HAL had "snapped" when he strangely repeats himself when he claims to have detected a problem. HAL suddenly says "just a moment... just a moment..."

It comes across as just the sort of error he was supposedly incapable of.

mjevans• 9 years ago

I agree with the hint at a timing cue but still don't think HAL 'went insane', I more agree that he found an 'insane' "solution" to a problem humans would find intractable.

The "just a moment..." can be thought of more as an auditory interface representation of "processing", where HAL is exploring the solution space of ever more unlikely solutions until it has at last resolved a statistically superior solution.

PeterisP• 9 years ago

It's just as we are eager to label others as insane simply because their goals violently conflict with ours and they would do hurtful things to us.

We're eager to label the behavior of various criminals or politicians as crazy or stupid because we disapprove of it - however, often (though not always) if you take a step back, it's quite consistent with their interest, the evil behavior is reasonable, rational and understandable in context of their situation and goals.

It's not the execution we should worry about, but rather about alignment of goals - as for other people, as for AI.

equalunique• 9 years ago

At "Replacing the AE-35 Unit" there seems to be a typo in this sentence: "Such people tend to be curious and want to get the bottom of mysteries, is HAL is trying to pique their curiosity so that they will take the required action to fulfil his murder plan."

sidcool• 9 years ago

I am amazed at the level of detail that Kubrick went into. Even after decades of movie's release there are hidden nuggets I keep finding. Absolutely brilliant.

vinceguidry• 9 years ago

Sanity is a legal definition, not a psychological one. People often forget that even the mentally-damaged act with reason and logic. This rationality may not be accessible to a bystander, but is present regardless.

I see the article as making a distinction without a difference. Kubrick was perfectly clear in his presentation of HAL's motives. There were a few points of interest, like the chess game, which I missed when I watched the film, but the 'theory' is how I viewed HAL's actions the whole time.

pps43• 9 years ago

The section on the chess game is not particularly persuasive. According to Stockfish, black already has winning position. So Frank (playing white) can no longer make mistakes. All moves at his disposal lead to the same outcome.

"Queen to Bishop three" is not a legal move, so Frank could have autocorrected it.

The Case for HAL's Sanity

Related Stories