47

Is there a good mechanical or electronic reason for HAL's voice to slow down, record-on-a-turntable-style, as he's disconnected?

It's aesthetically effective, and I'm happy to grant artistic licence, but I'm just curious whether there's actual justification. (I come from a place of mechanical/electronic ignorance. The only thing that occurred to me in thinking about this is that computers used to include reel-to-reel tape, but that wouldn't matter here: HAL's speech isn't pre-recorded, and removing his memory banks wouldn't cause machinery to slow down.)

  • 2
    I cannot think of any reasonable in-universe reason that could explain that behaviour. – Ghanima Jan 1 '15 at 22:08
  • 1
    In reality, I would expect clipping audio, not slower, magnetic tape style, audio. Kind of like when YT keeps having to buffer every five seconds. – fredsbend Jan 1 '15 at 22:09
  • 9
    I've found this extremely detailed explanation of how the effect was achieved. It's wholly irrelevant to the question, but still very interesting : wendycarlos.com/other/Eltro-1967/index.html – Valorum Jan 2 '15 at 0:09
  • 4
    Perhaps HAL's anti-brownout ultra-capacitors started discharging when disconnected? When capacitors are disconnected in a DC circuit, their step response is to discharge current following an exponential-decay law, eventually providing negligible current after a few time constants have passed. This decay could explain HAL's slowing voice as less and less current is available to drive the computer's circuitry and outputs. – Iwillnotexist Idonotexist Jan 2 '15 at 3:46
  • 2
    For what it's worth, in the book HAL's voice does not slow down. It rather becomes confused and full of hesitations, cutting off "suddenly" in the middle of "Daisy" and resuming with a "much slower", "dead mechanical intonation" with HAL's power-on speech. – JdeBP Jan 4 '15 at 0:30

10 Answers 10

60

The best reason is not in-universe, but viewer-verse: The sound that a record or tape makes when played too slow, by broken equipment, was familiar and had an immediate connotation of "malfunction" to the audience at the time the movie was made. Those audiences did not have exposure to the sound that clipping or buffering issues would cause, which is what would probably be used today (which would, mind you, be just as irrelevant to the in-universe problem!)

It's a question of "how do you translate something that doesn't exist into something the audience will understand?" I think that's separate from an "out-of-universe" explanation because it speaks to the translation layer between universes.

  • This could also apply to Stella the android in TOS I, Mudd. While idealistic, the 1960's audiences weren't very computer savvy. – Major Stackings Jan 2 '15 at 5:52
  • Right, a Shodan-like voice would be more realistic and better understood by a modern audience, but the original audience would probably not understand it as well. – vsz Jan 2 '15 at 15:37
  • 2
    As a modern example, in Portal, GLaDOS's voice also speeds up and slows down erratically to indicate failure. – Gustav Bertram Jan 6 '15 at 8:39
54

A reduction in audio clock speed could cause slow playback like HAL's. It's not a great reason—gowenfar's answer regarding connotations of "malfunction" makes much more sense—but it would cause the same symptoms in audio playback in real systems old and new.

Sound is a series of air pressure waves; analog audio devices (including gramophones, vinyl records, and magnetic tape) reproduce those waves' frequency and amplitude as the media passes across the sensor (needle or magnetic head) at a predictable speed. Digital audio isn't very different, in that the system loads numeric readings ("samples") of the microphone position across time to reproduce the waves through a speaker. You can think of an audio file, at its core, as a series of samples and a speed at which to reproduce them ("sampling rate"). CD audio is sampled at 44.1KHz, such that it reproduces 44,100 audio samples per second, whereas DVD audio is usually 48KHz or 96KHz; it doesn't matter as long as playback speed mimics capture speed. If the playback frequency were half of what it should be, for instance, the audio would take twice as long to play and sound an octave lower.

Unlike when audio buffers or clips (where the clock speed is right, but network or decompression speed causes a lack of data), a reduction in clock speed would cause the existing data to be output slower than usual, which would sound lower and slower the exact same way a record or tape would if played at the wrong speed.

What would cause this in-universe? Though most modern sound subsystems have their own quartz-crystal clocks that regulate audio timing, one can imagine a system that reuses the system's main clock and, importantly, fails to compensate for any missed cycles. (If you have an engineering background and this sounds far-fetched, see this article in which a real 1980's speech chip, GI's SP0256, requires an audio clock input and the author uses a CPU pin to do so.) An overall system slowdown with that assumption would thus cause audio systems to play back lower and slower than usual, which would sound more-or-less as it does in the movie.

  • As an electrical engineer, this was certainly how I understood what was going on in the movie. Don't really think it sounds that far-fetched. – KRyan Jan 3 '15 at 16:47
  • Yes, lost interrupts due to memory pressure does sound at least plausible – Hagen von Eitzen Jan 3 '15 at 21:07
  • Perhaps the global clock was determined by pulses provided by the modules, and none of the modules were expected to fail? – Lodewijk Jan 4 '15 at 22:09
  • 3
    An answer from Bowman, yes, a different Bowman, but still. – ThomasW Jan 5 '15 at 15:27
27

The original screenplay describes his mental decline in a little more detail. The slowing and deepening of his voice was intended to demonstrate his becoming "childish" and eventually stopping when his higher logic functions had been removed:

BOWMAN IGNORES HIM.

HAL : Dave, I don't understand why you're doing this to me.... I have the greatest enthusiasm for the mission... You are destroying my mind... Don't you understand? ... I will become childish... I will become nothing.

BOWMAN KEEPS PULLING OUT THE MEMORY BLOCKS.

HAL : Say, Dave... The quick brown fox jumped over the fat lazy dog... The square root of pi is 1.7724538090... log e to the base ten is 0.4342944 ... the square root of ten is 3.16227766... I am HAL 9000 computer. I became operational at the HAL plant in Urbana, Illinois, on January 12th, 1991. My first instructor was Mr. Arkany. He taught me to sing a song... it goes like this... "Daisy, Daisy, give me your answer do. I'm half crazy all for the love of you... etc.,"

COMPUTER CONTINUES TO SING SONG BECOMING MORE AND MORE CHILDISH AND MAKING MISTAKES AND GOING OFF-KEY. IT FINALLY STOPS COMPLETELY.

The contrast between Hal's easy speech in the earlier parts of the movie and his stilting sing-song was an homage to the earliest days of computerised speech;

22

I have seen gradual slowing of external interactions in real computers with increasingly overloaded memory.

HAL is suffering from reduced memory size. It has to dedicate more and more of its processor capacity to managing memory. It has to compress and decompress, move live data to high latency storage, rearrange blocks, and recalculate rather than reusing results of previous calculations. At some point, it will have to abandon data it cannot recalculate or otherwise recover. Even deciding what data to keep and what to abandon would consume processor time.

All of that reduces the share of processor time that is available to do useful work.

I have spent a significant amount of time writing and testing memory management software, starting in the late 1970's. Computers don't get correct results under extreme memory pressure by accident - it takes a lot of work. HAL's developers may not have designed or tested for the memory sizes to which HAL was ultimately reduced, so there would be likely to be a lot of bugs that would show up only under those conditions.

  • in digital audio,it would actually take more work to stretch out the sound. the result of errors or difficulties playing would just result in parts being dropped or being repeated if the buffer isn't being refilled. – user25857 Jan 2 '15 at 5:10
  • 2
    @user25857 I think you are assuming specialized hardware to generate the sound wave forms that cannot be diverted to memory management. Various functions move in and out of general processors depending on hardware trends. Obviously, the movie assumed more progress in computing than we actually made by 2001, so it is entirely possible HAL did audio wave form generation in general processors that could be diverted to other activities. – Patricia Shanahan Jan 2 '15 at 7:05
  • @user25857 Do You speak like that when tired or having reduced mental capacity? Maybe HAL was not that mechanistic machine whose brain-parts could be clearly separated to different architectures and functionalities. – Roland Pihlakas Jan 2 '15 at 8:39
  • 2
    @user25857 It does not take any more computing power to lower the sample rate. As long as both pitch and speed are reduced in tandem, it's literally just a matter of outputting at a slower pace. – trlkly Jan 2 '15 at 13:48
  • 1
    @LightnessRacesinOrbit I actually only added the mention of my experience in an edit, and that is only in an attempt to explain the incorrect results, rather than the slow-down that is the primary issue. I did try to explain why it makes sense - memory management absorbing an increasing proportion of compute power as the memory situation becomes more desperate. – Patricia Shanahan Jan 3 '15 at 22:34
10

As a researcher in Artificial Intelligence, Cognitive Science and Robotic Engineering, the way HAL's speech alters and regresses makes a lot of sense. The machine I use has 1000 cores, and the approach has perceptual and motor processing distributed as well as higher level processing. To explain in more detail...

To build an AI requires lots of processors, which are distributed across multiple boxes with multiple modules - still today they can look pretty similar to the cabinets and the boards we see being removed. Functionality is distributed across these many processors, including higher function (linguistic, ontological, learning and reasoning capabilities), long and short-term memory (life memory and working memory), and lower function (perceptual/sensory-motor abilities).

I regularly see movies/animation slowed down on my computer when there is not enough CPU, thinking/reasoning is slow when there are not enough processors available. The mammalian brain is able to shift functionality, with difficulty, into nearby areas, and ablation studies have shown incredible plasticity and resilience and ability to transfer functionality to a relatively small volume (I recall studies that went as low as 10%). Some studies deliberately seek to understand the human brain by turning off parts, and we have natural experiments resulting from strokes and accidents - and the slowing down and loss of some grammatical is well known in human aphasia (but without the lowering of pitch).

My approach to building an AI is like HAL's based on learning (as explained mainly in the 2010 sequel) and psycholinguistics shows the first memories are in some senses deepest. My slogan in the 1980s and 1990s was "HAL by 2001" - but I didn't get the funding required to achieve that!-) But basically, back then, we already knew what to do but since a review in 1970 of US funding for AI research and language technology, funding has largely been unavailable in this area due to slow progress, and the inability to deliver promises. IBMs developments with Big Blue and Watson show the capability required for playing chess or answering questions are there. It's the kind of things a two-year old can do that computers haven't learned yet - because they need to be trained like a baby to learn to interact in their social and physical environment.

The slowing of speech makes perfect sense. The lowering of pitch makes sense if speech synthesis is performed with distributed waveforms and due to unexpected loss of processing power, the synthesis of the waveform is slowed down because the required CPU power is not available but is being used for other functions, and likely is being performed using jury-rigged pathways, and components not optimized/designed for this purpose.

  • This is a really excellent answer, and I think it can explain the pitch shift as well. One popular way to synthesize speech is to set up filters that represent the resonances of the human vocal tract, and excite them with either a periodic signal (representing the vocal cords) or a noise signal. Such a model of the speech system would be competing for the same resources that you're talking about, and things like the periodic excitation (pitch) would slow down for the same reasons. – Dave Tweed Jan 4 '15 at 1:36
6

One possible explanation is that one of the circuit boards removed by Dave is the real time clock, so that HAL can no longer rely on it and has to fall back to less accurate measures (i.e. counting CPU cycles) to time his speech.

(I know, this answer is pure conjecture, not based on any in-universe clues, but then, so are the other answers...)

5

In addition to other answers one may consider that in future computers memory and CPU circuits are interlaced. That means huge parallel computing going on all the time.

By the way, does your brain have a separate memory area ready to be unplugged? There are programming architectures where data structure is the program. There is no separate program and memory.

Now imagine part of this computing power being removed. All algorithms that expect the neural network to stabilise to some state take longer since part of the input data is missing, the conclusions from remaining data are more unstable, and also the computing capacity is less.

3

There's a potentially darker explanation for this that is slightly hinted at by Richard's answer about HAL's speech becoming more and more childish:

Dave is performing the equivalent of a lobotomy - or similar type of destructive brain surgery - on HAL's brain: he is removing parts of HAL's brain gradually, while HAL is conscious of what is happening but unable to intervene, and is experiencing the effects, in real-time, that his actions are having on HAL's ability to think, speak and reason. His speech is consequently slowing - an equivalent of a human's slow, slurred speech when impaired.

From the information on Rosemary Kennedy's lobotomy, we find via Wikipedia (warning: potentially disturbing):

"We went through the top of the head, I think she was awake. She had a mild tranquilizer. I made a surgical incision in the brain through the skull. It was near the front. It was on both sides. We just made a small incision, no more than an inch." The instrument Dr. Watts used looked like a butter knife. He swung it up and down to cut brain tissue. "We put an instrument inside," he said. As Dr. Watts cut, Dr. Freeman put questions to Rosemary. For example, he asked her to recite the Lord's Prayer or sing "God Bless America" or count backwards..... "We made an estimate on how far to cut based on how she responded." ..... When she began to become incoherent, they stopped.[15]

After the lobotomy, it quickly became apparent that the procedure was not successful. Kennedy's mental capacity diminished to that of a two-year-old child. She could not walk or speak intelligibly ...

This kind of tragic tale has played out in other accounts of failed lobotomies or experimental brain surgeries performed with a conscious patient: the patient's speech becomes noticeably more impaired in real time.

Take this as an analogy for HAL's human-ness, or just a convenient metaphor if you will, but HAL's brain is being destroyed.

Having HAL sing a child's song adds to the tragedy; we suddenly see HAL at his most human and vulnerable.

  • Well that was an especially creepy read... Still, upvote. – James Sheridan Jan 5 '15 at 12:21
  • Right. Dave is performing a metaphorical lobotomy on HAL - this is why HAL regresses to his 'infancy'. Walter Jackson Freeman called this form of regression "surgically induced childhood": "The patient emerged from the operation ... with an 'immature' and even 'infantile' personality". Hence the nursery rhyme. – A E Jan 5 '15 at 14:28
  • In the book HAL, whose "design... had been copied from the human brain" says, "you are destroying my mind ... I will become childish". HAL's 'auto-intellection' module (which Dave removes) is his pre-frontal cortex: "The basic activity of this brain region is considered to be orchestration of thoughts and actions in accordance with internal goals." – A E Jan 5 '15 at 14:29
0

I think the answer in-universe is clear. It is 2001: a Space Oddessy , presumably in the year 2001. Well most space programs take decades to prepare and launch. So in all probability, HAL was built in the 80's. And so his speaking functionality was built using tape players, not digital audio. I argue that his voice was a complex system of tape recordings that were triggered by the computer in order to simulate speach.

  • There's no evidence in the film that his voice is simulated by a tape-based system. – Valorum Jan 2 '15 at 20:01
  • AT&T released their first digital answering machine to consumers in the early 90's, it's hard to believe that a computer advanced enough to use what looks like clear holographic memory cards would use tape to generate voice, even if it was built in the 1980's – Johnny Jan 2 '15 at 20:10
  • 6
    "Good afternoon, gentlemen. I am a HAL 9000 computer. I became operational at the H.A.L. plant in Urbana, Illinois on the 12th of January 1992. My instructor was Mr. Langley, and he taught me to sing a song." - Nope. – user8719 Jan 2 '15 at 20:15
  • 1
    Became operational on 1992, so it Was being built in the 80's in all likely hood. There IS evidence that the voice was by tape, as the audio slows down. However, there is no evidence it is using digital audio either. So if anything, the flimsy evidence would point to tapes. We are all ctrying to come up with ways that it wasn't on tape, because we feel the tech was too advanced to use something so old like tape. But hey, if it ain't broke, why fix it? Maybe they found that tape was more reliable? – user3183542 Jan 3 '15 at 21:01
  • @user3183542 - see en.wikipedia.org/wiki/… please; digital audio dates to the 60s (I'd have a hard time believing that Clarke at least didn't know of it) and was common by the early 80s. – user8719 Jan 3 '15 at 23:02
0

Speech synthesis is usually a computed operation, true, but the other comments are missing the point. Sure as you disconnect the processing units HAL's voice get slower, but if you analyze what Dave is actually disconnecting it becomes a bit clearer.

HAL 9000 is not a single system, he is a distributed node system with redundancies. As Dave disconnects the processing units, HAL has less and less margin time for the important things (Life support, ship upkeep) and is forced to run speech synthesis and coherence as background jobs. Dave is basically giving HAL a Lobotomy. Eventually the system is unable to keep HAL from talking a rate that would be suitable and slows down his voice, why? Because his speech generator is not only doing text-to-speech, but also has to do cognitive processing to keep a Turing complete conversation going.

When HAL slows down, it has to do with the fact that HAL Cannot come up with a proper sentence on it's Hard-Realtime deadlines and has to compensate. IBM designed HAL in such a way that even in the event of catastrophic failure of many modules, HAL would still be understandable. There is no Buffer Under run situation (Think a CD repeating when anti-skip fails) because IBM has taken this into consideration and creates a mechanism for hiding it.

This comes in the form of slowing down synthesis. This gives the processor more time to come up with the next word for the sentence since each word takes longer to pronounce. Humans do this all the time. Rather than say single words with large breath spaces in between them, we slow down the tempo in order to keep ourselves legible. However, eventually there is simply not enough memory left / not enough CPU time in HAL's processors to keep a legible sentence together. This is when he stops singing.

IBM at the time of the movie was making mainframes, and each individual processing task had it's own hardware. This is why the voice synth never really stops even though HAL cannot conjure up coherent sentences and resorts to singing. Dave was disconnecting main processors, not the specialty Speech Processor I'm almost sure IBM would have made in hardware rather than run as a piece of software.

  • 2
    If you broke this up with some paragraphs it could actually be a reasonable answer; uncyclopedia.wikia.com/wiki/Wall_of_Text – Valorum Jan 3 '15 at 23:34
  • 1
    There is no in-universe connection between HAL and IBM, other than the authors' in-joke that HAL is the result of shifting each letter in IBM backward by one. In fact there is evidence in the movie that HAL is not an IBM product since the IBM logo does appear on another piece of equipment and doesn't on HAL. – keshlam Jan 4 '15 at 1:04
  • @keshlam The connection is that HAL was built by hardware and software company just like an IBM. They certainly might have had similar approaches. The point is in analogy and giving an example. Or was that confusing? – Roland Pihlakas Jan 4 '15 at 15:26
  • I was trying to go with that assumption, but my answer still works without IBM designing HAL – primis Jan 9 '15 at 19:04
  • "Turing complete conversation" doesn't mean whatever you think it means. – immibis May 4 '16 at 10:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.