8 Reasons The Voight-Kampff Machine is shit (and a redesign to fix it)

Distinguishing replicants from humans is a tricky business. Since they are indistinguishable biologically, it requires an empathy test, during which the subject hears empathy-eliciting scenarios and watched carefully for telltale signs such as, “capillary dilation—the so-called blush response…fluctuation of the pupil…involuntary dilation of the iris.” To aid the blade runner in this examination, they use a portable machine called the Voight-Kampff machine, named, presumably, for its inventors.

The device is the size of a thick laptop computer, and rests flat on the table between the blade runner and subject. When the blade runner prepares the machine for the test, they turn it on, and a small adjustable armature rises from the machine, the end of which is an intricate piece of hardware, housing a powerful camera, glowing red.

The blade runner trains this camera on one of the subject’s eyes. Then, while reading from the playbook book of scenarios, they keep watch on a large monitor, which shows an magnified image of the subject’s eye. (Ostensibly, anyway. More on this below.) A small bellows on the subject’s side of the machine raises and lowers. On the blade runner’s side of the machine, a row of lights reflect the volume of the subject’s speech. Three square, white buttons sit to the right of the main monitor. In Leon’s test we see Holden press the leftmost of the three, and the iris in the monitor becomes brighter, illuminated from some unseen light source. The purpose of the other two square buttons is unknown. Two smaller monochrome monitors sit to the left of the main monitor, showing moving but otherwise inscrutable forms of information.

In theory, the system allows the blade runner to more easily watch for the minute telltale changes in the eye and blush response, while keeping a comfortable social distance from the subject. Substandard responses reveal a lack of empathy and thereby a high probability that the subject is a replicant. Simple! But on review, it’s shit. I know this is going to upset fans, so let me enumerate the reasons, and then propose a better solution.

-2. Wouldn’t a genetic test make more sense?

If the replicants are genetically engineered for short lives, wouldn’t a genetic test make more sense? Take a drop of blood and look for markers of incredibly short telomeres or something.

-1. Wouldn’t an fMRI make more sense?

An fMRI would reveal empathic responses in the inferior frontal gyrus, or cognitive responses in the ventromedial prefrontal gyrus. (The brain structures responsible for these responses.) Certinaly more expensive, but more certain.

0. Wouldn’t a metal detector make more sense?

If you are testing employees to detect which ones are the murdery ones and which ones aren’t, you might want to test whether they are bringing a tool of murder with them. Because once they’re found out, they might want to murder you. This scene should be rewritten such that Leon leaps across the desk and strangles Holden, IMHO. It would make him, and other blade runners, seem much more feral and unpredictable.

(OK, those aren’t interface issues but seriously wtf. Onward.)

1. Labels, people

Controls needs labels. Especially when the buttons have no natural affordance and the costs of experimentation to discover the function are high. Remembering the functions of unlabeled controls adds to the cognitive load for a user who should be focusing on the person across the table. At least an illuminated button helps signal the state, so that, at least, is something.

 2. It should be less intimidating

The physical design is quite intimidating: The way it puts a barrier in between the blade runner and subject. The fact that all the displays point away from the subject. The weird intricacy of the camera, its ominous HAL-like red glow. Regular readers may note that the eyepiece is red-on-black and pointy. That is to say, it is aposematic. That is to say, it looks evil. That is to say, intimidating.

I’m no emotion-scientist, but I’m pretty sure that if you’re testing for empathy, you don’t want to complicate things by introducing intimidation into the equation. Yes, yes, yes, the machine works by making the subject feel like they have to defend themselves from the accusations in the ethical dilemmas, but that stress should come from the content, not the machine.

2a. Holden should be less intimidating and not tip his hand

While we’re on this point, let me add that Holden should be less intimidating, too. When Holden tells Leon that a tortoise and a turtle are the same thing, (Narrator: They aren’t) he happens to glance down at the machine. At that moment, Leon says, “I’ve never seen a turtle,” a light shines on the pupil and the iris contracts. Holden sees this and then gets all “ok, replicant” and becomes hostile toward Leon.

In case it needs saying: If you are trying to tell whether the person across from you is a murderous replicant, and you suddenly think the answer is yes, you do not tip your hand and let them know what you know. Because they will no longer have a reason to hide their murderyness. Because they will murder you, and then escape, to murder again. That’s like, blade runner 101, HOLDEN.

3. It should display history 

The glance moment points out another flaw in the interface. Holden happens to be looking down at the machine at that moment. If he wasn’t paying attention, he would have missed the signal. The machine needs to display the interview over time, and draw his attention to troublesome moments. That way, when his attention returns to the machine, he can see that something important happened, even if it’s not happening now, and tell at a glance what the thing was.

4. It should track the subject’s eyes

Holden asks Leon to stay very still. But people are bound to involuntarily move as their attention drifts to the content of the empathy dilemmas. Are we going to add noncompliance-guilt to the list of emotional complications? Use visual recognition algorithms and high-resolution cameras to just track the subject’s eyes no matter how they shift in their seat.

5. Really? A bellows?

The bellows doesn’t make much sense either. I don’t believe it could, at the distance it sits from the subject, help detect “capillary dilation” or “ophthalmological measurements”. But it’s certainly creepy and Terry Gilliam-esque. It adds to the pointless intimidation.

6. It should show the actual subject’s eye

The eye color that appears on the monitor (hazel) matches neither Leon’s (a striking blue) or Rachel’s (a rich brown). Hat tip to Typeset in the Future for this observation. His is a great review.

7. It should visualize things in ways that make it easy to detect differences in key measurements

Even if the inky, dancing black blob is meant to convey some sort of information, the shape is too organic for anyone to make meaningful readings from it. Like seriously, what is this meant to convey?

The spectrograph to the left looks a little more convincing, but it still requires the blade runner to do all the work of recognizing when things are out of expected ranges.

8. The machine should, you know, help them

The machine asks its blade runner to do a lot of work to use it. This is visual work and memory work and even work estimating when things are out of norms. But this is all something the machine could help them with. Fortunately, this is a tractable problem, using the mighty powers of logic and design.

Pupillary diameter

People are notoriously bad at estimating the sizes of things by sight. Computers, however, are good at it. Help the blade runner by providing a measurement of the thing they are watching for: pupillary diameter. (n.b. The script speaks of both iris constriction and pupillary diameter, but these are the same thing.) Keep it convincing and looking cool by having this be an overlay on the live video of the subject’s eye.

So now there’s some precision to work with. But as noted above, we don’t want to burden the user’s memory with having to remember stuff, and we don’t want them to just be glued to the screen, hoping they don’t miss something important. People are terrible at vigilance tasks. Computers are great at them. The machine should track and display the information from the whole session.

Note that the display illustrates radius, but displays diameter. That buys some efficiencies in the final interface.

Now, with the data-over-time, the user can glance to see what’s been happening and a precise comparison of that measurement over time. But, tracking in detail, we quickly run out of screen real estate. So let’s break the display into increments with differing scales.

There may be more useful increments, but microseconds and seconds feel pretty convincing, with the leftmost column compressing gradually over time to show everything from the beginning of the interview. Now the user has a whole picture to look at. But this still burdens them into noticing when these measurements are out of normal human ranges. So, let’s plot the threshold, and note when measurements fall outside of that. In this case, it feels right that replicants display less that normal pupillary dilation, so it’s a lower-boundary threshold. The interface should highlight when the measurement dips below this.


I think that covers everything for the pupillary diameter. The other measurement mentioned in the dialogue is capillary dilation of the face, or the “so-called blush response.” As we did for pupillary diameter, let’s also show a measurement of the subject’s skin temperature over time as a line chart. (You might think skin color is a more natural measurement, but for replicants with a darker skin tone than our two pasty examples Leon and Rachel, temperature via infrared is a more reliable metric.) For visual interest, let’s show thumbnails from the video. We can augment the image with degree-of-blush. Reduce the image to high contrast grayscale, use visual recognition to isolate the face, and then provide an overlay to the face that illustrates the degree of blush.

But again, we’re not just looking for blush changes. No, we’re looking for blush compared to human norms for the test. It would look different if we were looking for more blushing in our subject than humans, but since the replicants are less empathetic than humans, we would want to compare and highlight measurements below a threshold. In the thumbnails, the background can be colored to show the median for expected norms, to make comparisons to the face easy. (Shown in the drawing to the right, below.) If the face looks too pale compared to the norm, that’s an indication that we might be looking at a replicant. Or a psychopath.

So now we have solid displays that help the blade runner detect pupillary diameter and blush over time. But it’s not that any diameter changes or blushing is bad. The idea is to detect whether the subject has less of a reaction than norms to what the blade runner is saying. The display should be annotating what the blade runner has said at each moment in time. And since human psychology is a complex thing, it should also track video of the blade runner’s expressions as well, since, as we see above, not all blade runners are able to maintain a poker face. HOLDEN.

Anyway, we can use the same thumbnail display of the face, without augmentation. Below that we can display the waveform (because they look cool), and speech-to-text the words that are being spoken. To ensure that the blade runner’s administration of the text is not unduly influencing the results, let’s add an overlay to the ideal intonation targets. Despite evidence in the film, let’s presume Holden is a trained professional, and he does not stray from those targets, so let’s skip designing the highlight and recourse-for-infraction for now.

Finally, since they’re working from a structured script, we can provide a “chapter” marker at the bottom for easy reference later.

Now we can put it all together, and it looks like this. One last thing we can do to help the blade runner is to highlight when all the signals indicate replicant-ness at once. This signal can’t be too much, or replicants being tested would know from the light on the blade runner’s face when their jig is up, and try to flee. Or murder. HOLDEN.

For this comp, I added a gray overlay to the column where pupillary and blush responses both indicated trouble. A visual designer would find some more elegant treatment.

If we were redesigning this from scratch, we could specify a wide display to accomodate this width. But if we are trying to squeeze this display into the existing prop from the movie, here’s how we could do it.

Note the added labels for the white squares. I picked some labels that would make sense in the context. “Calibrate” and “record” should be obvious. The idea behind “mark” is an easy button for the blade runner to press when they see something that looks weird, like when doctors manually annotate cardiograph output.

Lying to Leon

There’s one more thing we can add to the machine that would help out, and that’s a display for the subject. Recall the machine is meant to test for replicant-ness, which happens to equate to murdery-ness. A positive result from the machine needs to be handled carefully so what happens to Holden in the movie doesn’t happen. I mentioned making the positive-overlay subtle above, but we can also make a placebo display on the subject’s side of the interface.

The visual hierarchy of this should make the subject feel like its purpose is to help them, but the real purpose is to make them think that everything’s fine. Given the script, I’d say a teleprompt of the empathy dilemma should take up the majority of this display. Oh, they think, this is to help me understand what’s being said, like a closed caption. Below the teleprompt, at a much smaller scale, a bar at the bottom is the real point.

On the left of this bar, a live waveform of the audio in the room helps the subject know that the machine is testing things live. In the middle, we can put one of those bouncy fuiget displays that clutters so many sci-fi interfaces. It’s there to be inscrutable, but convince the subject that the machine is really sophisticated. (Hey, a diegetic fuiget!) Lastly—and this is the important part—An area shows that everything is “within range.” This tells the subject that they can be at ease. This is good for the human subject, because they know they’re innocent. And if it’s a replicant subject, this false comfort protects the blade runner from sudden murder. This test might flicker or change occasionally to something ambiguous like “at range,” to convey that it is responding to real world input, but it would never change to something incriminating.

This way, once the blade runner has the data to confirm that the subject is a replicant, they can continue to the end of the module as if everything was normal, thank the replicant for their time, and let them leave the room believing they passed the test. Then the results can be sent to the precinct and authorizations returned so retirement can be planned with the added benefit of the element of surprise.


Look, I’m sad about this, too. The Voight-Kampff machine is cool. It fits very well within the art direction of the Blade Runner universe. This coolness burned the machine into my memory when I saw this film the first dozen times, but despite that, it just doesn’t stand up to inspection. It’s not hopeless, but does need a lot of thinkwork and design to make it really fit to task, and convincing to us in the audience.


5 thoughts on “8 Reasons The Voight-Kampff Machine is shit (and a redesign to fix it)

  1. A couple of comments on Hollywood vs Realism. Firstly, one has to recall that lacking the capability for a character in a novel to introspect (bad voiceovers aside), everything about a character in a movie has to be shown on screen. Therefore, what n actor displays onscreen has to be an amplified version of what would be seen in real life, so that the audience can catch it. Real Holden may have reacted to the real-time display with a mere micro-expression rather than the actor’s emoting, which Leon may or may not have caught (replicants are probably not good at reading expressions, though).

    Secondly, I’d swap your two buttons, Record and Calibrate, so it becomes clear that the interview is being recorded, because the usual procedure is probably to record for later review. Another Hollywood 101: cut the boring bits. VK interviews are probably usually more bureaucratic, with the interviewer just reading the questions and letting the machine’s realtime monitor prompt him for follow-up questions. But Holden’s so experienced that he can pick up on small cues, which is what kills him, alas. Holden spots something, Leon notices, which makes Holden more inquisitive, then realising the potential positive feedback loop, tries to calm things down. Doesn’t work. Even experienced cops make mistakes.

    Side note: you know this already, but DNA profiling was hardly on the radar in 1982; the first draft of the human genome was two decades in the future, and took a while to produce. Whatever technology Tyrell use for their genetic engineering in Blade Runner’s alternative 2019 probably isn’t portable or quick.

    • Yep. Actor “tells” are important for audiences to understand what’s going on in the moment. So we can read Holden’s “ok, replicant” and hostility real-time, extradiegetically. But the scene would have worked as well if Holden had said, “Thank you for your time,” Leon got up and walked to the door, and Holden pulled out a weapon and shot him in the back (audience: whut?), and then Leon turned around, his shoulder bleeding, and strangled Holden. (audience: WHUUT?)

      For the buttons, wouldn’t calibration happen before recording? I’m not married to any particular order, but chronological order makes most sense to me.

      You’re completely right about DNA testing, but this blog adopts the New Criticism stance, which deliberately disregards intent or historical constraints. This helps avoid becoming a historian or archaeologist of the work. I rather want to interpret things for what they mean to us today. Dave Addey on Typeset in the Future has a much more historical approach, and does the ground work to situate his criticism in the times. His stuff is great, just not my approach.

      Thanks for the thoughtful comments, btw. Most of my posts go quietly into the ether. Nice to know someone is reading and engaged with the material.

      • On the order of the buttons, I was thinking that the lit button should be the ‘Record’ button, if we’re putting labels on the buttons after the fact. You’re right about what the order for a useful UI should be, though.

        On your rewrite of the scene, I don’t think Holden missing an easy kill works. It’s something you see in movies all the time — the antagonist surviving an encounter with a mook through the mook making an unprofessional mistake — and it might work in an opening scene, but it’s vulnerable to fridge logic. You’d need a reason for Holden, a seasoned Blade Runner, to miss. Whereas with the scene as shown, it’s all there on the screen: the increasing tension as Leon’s confusion mirrors the audience’s, Holden’s tell, then his attempt to calm things down which looks like it works, but it’s just Leon playing Holden. Holden screws up because Leon’s smarter than he’s used to, maybe, which is a thing that makes the antagonist a real foe and not just a monster. If the Nexus-6’s are bullet-proof instead, that raises problems for Deckard’s part of the plot later. They damn near kill him as it is.

        Things like Holden’s tell are why we have the concepts of diegesis and exegesis in film criticism, and I think it’s reasonable to raise that kind of detail in a criticism, but I don’t feel it’s reasonable to make it a fault in the film-making. If anywhere, the fault lies in ourselves for being so dull-witted that we need things like that telegraphed, and not in the story-telling that tries to accommodate our dull-wittedness.

        (I keep trying to make a short comment, but they keep expanding…)

    • All that said, I do ride the line between solving for the real world and solving for the movie in my posts. I try to make suggestions that work for both. Where that’s not possible, or I’m privileging one to the exclusion of the other, I should be clear.

  2. Pingback: connecting.the.dots.03 - Carmel Hassan

Leave a Reply