This is one of those interactions that happens over a few seconds in the movie, but turns out to be quite deep—and broken—on inspection.
When Deckard enters his building’s dark, padded elevator, a flat voice announces, “Voice print identification. Your floor number, please.” He presses a dark panel, which lights up in response. He presses the 9 and 7 keys on a keypad there as he says, “Deckard. 97.” The voice immediately responds, “97. Thank you.” As the elevator moves, the interface confirms the direction of travel with gentle rising tones that correspond to the floor numbers (mod 10), which are shown rising up a 7-segment LED display. We see a green projection of the floor numbers cross Deckard’s face for a bit until, exhausted, he leans against the wall and out of the projection. When he gets to his floor, the door opens and the panel goes dark.
A need for speed
An aside: To make 97 floors in 20 seconds you have to be traveling at an average of around 47 miles per hour. That’s not unheard of today. Mashable says in a 2014 article about the world’s fastest elevators that the Hitachi elevators in Guangzhou CTF Finance Building reach up to 45 miles per hour. But including acceleration and deceleration adds to the total time, so it takes the Hitachi elevators around 43 seconds to go from the ground floor to their 95th floor. If 97 is Deckard’s floor, it’s got to be accelerating and decelerating incredibly quickly. His body doesn’t appear to be suffering those kinds of Gs, so unless they have managed to upend Newton’s basic laws of motion, something in this scene is not right. As usual, I digress.
The input control is OK
The panel design is nice and was surprising in 1982, because few people had ridden in elevators serving nearly a hundred floors. And while most in-elevator panels have a single button per floor, it would have been an overwhelming UI to present the rider of this Blade Runner complex with 100 floor buttons plus the usual open door, close door, emergency alert buttons, etc. A panel that allows combinatorial inputs reduces the number of elements that must be displayed and processed by the user, even if it slows things down, introduces cognitive overhead, and adds the need for error-handling. Such systems need a “commit” control that allows them to review, edit, and confirm the sequence, to distinguish, say, “97” from “9” and “7.” Not such an issue from the 1st floor, but a frustration from 10–96. It’s not clear those controls are part of this input.
I’m a fan of destination dispatch elevator systems that increase efficiency (with caveats) by asking riders to indicate their floor outside the elevator and letting the algorithm organize passengers into efficient groups, but that only works for banks of elevators. I get the sense Deckard’s building is a little too low-rent for such luxuries. There is just one in his building, and in-elevator controls work fine for those situations, even if they slow things down a bit.
The feedback is OK
The feedback of the floors is kind of nice in that the 7-segment numbers rise up helping to convey the direction of movement. There is also a subtle, repeating, rising series of tones that accompany the display. Most modern elevators rely on the numeracy of its passengers and their sense of equilibrium to convey this information, but sure, this is another way to do it. Also, it would be nice if the voice system would, for the visually impaired, say the floor number when the door opens.
Though the projection is dumb
I’m not sure why the little green projection of the floor numbers runs across Deckard’s face. Is it just a filmmaker’s conceit, like the genetic code that gets projected across the velociraptors head in Jurassic Park?
Or is it meant to be read as diegetic, that is, that there is a projector in the elevator, spraying the floor numbers across the faces of its riders? True to the New Criticism stance of this blog, I try very hard to presume that everything is diegetic, but I just can’t make that make sense. There would be much better ways to increase the visibility of the floor numbers, and I can’t come up with any other convincing reason why this would exist.
But really, it falls apart on the interaction details
Lastly, this interaction. First, let’s give it credit where credit is due. The elevator speaks clearly and understands Deckard perfectly. No surprise, since it only needs to understand a very limited number of utterances. It’s also nice that it’s polite without being too cheery about it. People in LA circa 2019 may have had a bad day and not have time for that shit.
Where’s the wake word?
But where’s the wake word? This is a phrase like “OK elevator” or “Hey lift” that signals to the natural language system that the user is talking to the elevator and not themselves, or another person in the elevator, or even on the phone. General AI exists in the Blade Runner world, and that might allow an elevator to use contextual cues to suss this out, but there are zero clues in the film that this elevator is sentient.
There are of course other possible, implicit “wake words.” A motion detector, proximity sensor, or even weight sensor could infer that a human is present, and start the elevator listening. But with any of these implicit “wake words,” you’d still need feedback for the user to know when it was listening. And some way to help them regain attention if they got the first interaction wrong, and there would be zero affordances for this. So really, making an explicit wake word is the right way to go.
It might be that touching the number panel is the attention signal. Touch it, and the elevator listens for a few seconds. That fits in with the events in the scene, anyway. The problem with that is the redundancy. (See below.) So if the solution was pressing a button, it should just be a “talk” button rather than a numeric keypad.
It may be that the elevator is always listening, which is a little dark and would stifle any conversation in the elevator less everyone end up stuck in the basement, but this seems very error prone and unlikely.
This issue is similar to the one discussed in Make It So Chapter 5, “Gestural Interfaces” where I discussed how a user tells a computer they are communicating to it with gestures, and when they aren’t.
Where are the paralinguistics?
Humans provide lots of signals to one another, outside of the meaning of what is actually being said. These communication signals are called paralinguistics, and one of those that commonly appears in modern voice assistants is feedback that the system is listening. In the Google Assistant, for example, the dots let you know when it’s listening to silence and when it’s hearing your voice, providing implicit confirmation to the user that the system can hear them. (Parsing the words, understanding the meaning, and understanding the intent are separate, subsequent issues.)
Fixing this in Blade Runner could be as simple as turning on a red LED when the elevator is listening, and varying the brightness with Deckard’s volume. Maybe add chimes to indicate the starting-to-listen and no-longer-listening moments. This elevator doesn’t have anything like that, and it ought to.
Why the redundancy?
Next, why would Deckard need to push buttons to indicate “97” even while he’s saying the same number as part of the voice print? Sure, it could be that the voice print system was added later and Deckard pushes the numbers out of habit. But that bit of backworlding doesn’t buy us much.
It might be a need for redundant, confirming input. This is useful when the feedback is obscure or the stakes are high, but this is a low-stakes situation. If he enters the wrong floor, he just has to enter the correct floor. It would also be easy to imagine the elevator would understand a correction mid-ride like “Oh wait. Elevator, I need some ice. Let’s go to 93 instead.” So this is not an interaction that needs redundancy.
It’s very nice to have the discrete input as accessibility for people who cannot speak, or who have an accent that is unrecognizable to the system, or as a graceful degradation in case the speech recognition fails, but Deckard doesn’t fit any of this. He would just enter and speak his floor.
Why the personally identifiable information?
If we were designing a system and we needed, for security, a voice print, we should protect the privacy of the rider by not requiring personally identifiable information. It’s easy to imagine the spoken name being abused by stalkers and identity thieves riding the elevator with him. (And let’s not forget there is a stalker on the elevator with him in this very scene.)
Better would be some generic phrase that stresses the parts of speech that a voiceprint system would find most effective in distinguishing people.
Tucker Saxon has written an article for VoiceIt called “Voiceprint Phrases.” In it he notes that a good voiceprint phrase needs some minimum number of non-repeating phonemes. In their case, it’s ten. A surname and a number is rarely going to provide that. “Deckard. 97,” happens to have exactly 10, but if he lived on the 2nd floor, it wouldn’t. Plus, it has that personally identifiable information, so is a non-starter.
What would be a better voiceprint phrase for this scene? Some of Saxon’s examples in the article include, “Never forget tomorrow is a new day” and “Today is a nice day to go for a walk.” While the system doesn’t care about the meaning of the phrase, the humans using it would be primed by the content, and so it would just add to the dystopia of the scene if Deckard had to utter one of these sunshine-and-rainbows phrases in an elevator that was probably an uncleaned murder scene. but I think we can do it one better.
(Hey Tucker, I would love use VoiceIt’s tools to craft a confirmed voiceprint phrase, but the signup requires that I permit your company to market to me via phone and email even though I’m just a hobbyist user, so…hard no.)
Here is an alternate interaction that would have solved a lot of these problems.
- ELEVATOR
- Voice print identification, please.
- DECKARD
- SIGHS
- DECKARD
- Have you considered life in the offworld colonies?
- ELEVATOR
- Confirmed. Floor?
- DECKARD
- 97
Which is just a punch to the gut considering Deckard is stuck here and he knows he’s stuck, and it’s salt on the wound to have to repeat fucking advertising just to get home for a drink.
So…not great
In total, this scene zooms by and the audience knows how to read it, and for that, it’s fine. (And really, it’s just a setup for the moment that happens right after the elevator door opens. No spoilers.) But on close inspection, from the perspective of modern interaction design, it needs a lot of work.
Having to use “Have you considered life in the offworld colonies?” as a password strikes me as a very Dickian thing.
Watching the clip, it strikes me that the dialogue was added after the scene was shot, since Deckard turns his back to the camera just as he speaks. Originally, maybe Ford just acted tired and punched the buttons, but the editor or director felt it needed a bit of voice and just patched in the dialogue later, not trying particularly hard to make it make sense.
Ooh. Nice observation.
The more I think about this, the more it makes sense. Like, maybe test audiences missed the worldbuilding detail of Ford pressing “97” and so Scott went back and added in the voice-overs so he could hit the audience over the head with both that detail and a more futuristic-seeming elevator.
Back of the envelope calc on elevator acceleration. (Wow, get to use spaceship mechanics to analyse a user interface :-))
Assume that each floor has 3m for living space and another 1m for girders and pipes and infrastructure. This is based on a quick google for apartment heights, which turned up the highest skyscraper in Hong Kong being just over 480m and 100 floors.
Fastest way to travel 97 floors is accelerate constantly to the half way point, then “turn over” and decelerate constantly to a stop. The second half is the same as the first but in reverse.
So 97 floors in 20 seconds becomes 48.5 floors in 10 seconds.
distance = average velocity x time, average velocity = accel x time / 2
So distance = accel x time squared / 2
Plugging in distance = 48.5 x 4m and time = 10, I get an acceleration of just 3.88 metres/sec, less than half a G.
For comparison, a car accelerating at that speed will be doing over 130 km/hr or over 80 mph after 10 seconds. Uncomfortable for most of us, but not painful.
Of course, that’s assuming an uninterrupted journey. If someone pushes the button on the 18th floor a few seconds after Deckard starts moving up, the elevator would have to decelerate a lot harder to stop in time. So I suspect it would actually go shooting past all the way to 97 and come back down.
Thank you Hugh. The 18th interruption makes this even funnier/darker.
I feel like the elevator ride took place in Hollywood time, actually taking a minute or two, but Ridley Scott didn’t feel he could hold the audience’s attention for more than 20 seconds; this isn’t Solaris, after all.
Ha. Quite probably true but I can’t start excusing cinematic conceits or it will all fall down like a house of cards.
I’d like to suggest a different interpretation of the elevator interaction. The way it works makes more sense if it’s two systems rather than one, another Blade Runner “retrofitted” building, and the voiceprint system isn’t as advanced.
The older system is the floor selection panel. Key in your floor number, the elevator goes there. It’s redundant for Deckard to key in his floor number when there’s a voice system, but the voice system has no interface into the elevator panel. (The company that made the panel refused to provide access, claiming it would void the warranty, but in reality the engineers didn’t want anyone else breaking their system.)
The voice print system is a later addition to stop random strangers from being able to reach any floor. As a separate system it’s a simple go/no go switch between the panel and the actual elevator motors.
Personally identifiable information is necessary if the voiceprint system cannot identify someone by voiceprint, only confirm it. The current day equivalent is the face recognition on phones and laptops. They can’t be given a photo and reliably determine “this is Kim” instead of someone else. But if it is expecting a photo of Kim and hence knows the allowable variation, it can reliably say yes or no.
So Deckard, like everyone else in the building, has to provide some unique personal identifier. A name and a floor number is a good choice and hardly something that needs to be kept confidential. (Deckard lives alone so just a surname, family members would add an initial.) “Deckard97” is his userid, the way he says it confirms that it is him.
I can buy the backworlding of separate disconnected systems. It’s right in line with the movie’s dystopian technology themes. But I disagree about PII being a-OK. As designers we have to take extreme care and err on the side of protecting PIi even when it seems benign. If Deckard had picked it himself, and the system knew how name, it should reject the password for security reasons.
First and foremost it has to be usable, and I don’t believe the residents of an apartment building (whether in the fictional world of Blade Runner or our own) could be required to learn say 10 phoneme pass phrases.
Think about the range of people who have to use this voiceprint system. Ages from school children to the very elderly. Different levels of education, different languages. As well as blade runners too exhausted to notice a stranger in the elevator, sometimes it will be an early riser who hasn’t had their coffee, a parent with a screaming child, someone coming home from a party who is drunk.
A false positive lets a non-resident in, but it isn’t too serious because the apartments still have locks. A false negative keeping a resident out will be annoying to serious.
If a diabetic grandmother can’t remember / retrieve her passphrase because she’s in the early stages of hypoglycemia and thus can’t reach her apartment with the insulin, I don’t think “we were protecting her privacy” will carry much weight with a jury.
I understand the argument for protecting PII, but the residents already know their names and floor numbers, and almost certainly don’t care if other people can overhear them. If security systems are perceived as more effort than what’s being protected is worth, they get bypassed or removed.
Pingback: Report Card: Blade Runner (1982) | Sci-fi interfaces
Does anyone happen to know who the voice actor is for the elevator computer?
It sounds like Sigourney Weaver or Kathleen Turner or a mashup…
I’m curious if the voice would be regional or localized depending on use and building place; we seem to deploy single one-world voice systems and haven’t had too many customization or options to make the voice your own…