Warning, spoilers ahead.
I saw the movie "Her" a few nights ago with my wife and daughter, at my wife's suggestion. She'd heard good things about it and was curious. I, on the other hand, had fairly low expectations. I'd read enough about it to know the premise and be acquainted with a few plot points-- those the promoters of the movie think will stimulate interest-- and from my perspective it wasn't promising. "Lonely guy falls in love with his phone" and some titillation about sex with his operating system. I mean, seriously? I was not looking forward to scenes in which a guy gets off talking to his computer.
So, the movie wasn't great, but it wasn't as bad as I was expecting. The titillation is there, all right. Besides sex with an operating system (tastefully fading to black as they moan, yes the OS apparently is programmed for sexual arousal), there's also scenes of phone sex (earlier, to ease us into the idea of there not even being a real person on the other end of the line), foreplay with a live surrogate for the (presumably sentient but disembodied) OS, and foreplay with a bad date (he doesn't quite connect with her, but they start to have sex anyway, but then she wants to know if he's ready for more than a one-night-stand). So, plenty of titillation to go around. But apart from that it's a more or less conventional story about relationships starting, changing, and ending, and the effects this has on everyone involved, just set in a somewhat idealized near future.
I didn't give it much thought, and wouldn't have thought to write about it, except that I read the thread on metafilter.com discussing the film.
So, a digression about metafilter. Metafilter is sortof like reddit except you have to pay a nominal fee in order to post. This keeps like 99.5% of the crap out of their comment system. In addition, the convention for introducing a topic is to provide a short paragraph about it, generally containing several related links riffing on the theme, linking to articles, videos, analysis, related work, etc. It's much more thoughtful than reddit, and the themes tend to be less topical and more idiosyncratic than those on reddit. (If you don't know what reddit is, well, um, this is the internet, you can find out for yourself.) Bottom line is, I generally find interesting discussions on metafilter.
This discussion was a little more random than most, and I found it alternately intriguing and aggravating. There were several divergent opinions about aspects the film, often ones I had difficulty sharing. One contingent held that speech recognition makes for a crummy UI (really, this was a big issue for some folks) and found the movie too unrealistic to enjoy as a result. Another contingent thought the movie was a touching sci-fi take on the arrival of the singularity (again, you can look it up on teh internets, but the one-sentence synopsis is that 'the singularity' is the point in human history at which artificial intelligences replace humans as the dominant life form on the planet). A third felt that the movie was vastly overpraised considering the rather conventional and sexist plot line.
I had different reactions to these three discussions. I'll start with speech since to me it's the least interesting.
Speech recognition as the pervasive UI.
Some people were interested in the movie's depiction of user interfaces of the future, and one of the articles linked to discussed this at some length. The filmmakers had thought about this the way filmmakers would, which is to say, their problem was how to provide enough atmospherics to support the narrative, but keep the technology out of the way. Since they presupposed a world in which software can become sentient, it was reasonable to have UIs that let you interact with the software as though it were nearly sentient, and this meant characters can talk to their devices with full fluidity and ambiguity, just as though talking to a human. No keyboard, no mouse, no gloves, no vr helmet. Kinda makes sense to me, but some people really balked at this. Voice input isn't good for working with spreadsheets. You need a restricted vocabulary. You have to train the software so it can recognize the speaker. You can't use it in a noisy environment. It makes workplaces too noisy.
These kinds of arguments held no sway with me. It's worth repeating-- the software understands speech like a human can. Heck, the software can think like a human can. This Is The Whole Premise Of The Movie. And we can accept the premise, in part, because we can see the improvements in speech recognition over the past decade. Some of these arguments don't quite hold even today. You don't need to train Siri or Android with your voice, because millions and millions of voice samples have been used to develop the speech recognition systems behind these products. The motox has an 'always on' listening system that recognizes your voice when you say a key phrase and responds to your commands. These systems are not foolproof but they're just about good enough that they become your first recourse. It's easy to extrapolate the development to the point where you can use these in noisy environments, they can deal with significant levels of ambiguity, you can use them subvocally and keep your voice low and unobtrusive.
The flexibility of the systems beyond raw word and key phrase recognition does lag behind, true. This is actually where the great leap of faith is, for me. Systems that can handle ambiguity and resolve among competing interpretations, and in particular formulate unobtrusive interventions (ask questions) to get clarity, aren't well developed yet. My sense is that current systems are largely based on heuristics and plans, and not on learning / statistical / neural network systems as they probably need to be in order to attain the level of flexibility they require to be really useful. I think it could take decades of work to get the kinds of advances we've been seeing with vision and speech systems. On the other hand, it may be that the techniques being developed for these systems can be applied to higher-order recognition tasks once we learn how to interpret these tasks in the right way. That's the breakthrough part, and I don't think we're that close yet.
Most everyone agreed that speech is part of a multimodal interface, it's not good for everything. Like, duh. I wouldn't try drawing using a speech system. But speech integrated as an additional high level modality (choose tool, customize operation, coarsely adjust values) makes sense because you don't really need your hands for these. Moving a mouse to click on buttons in a dialog is just cumbersome when you can say what you want to do instead.
So, talking to computers that are embedded in your environment? Sure, I'm down with that.
The coming of the singularity, and sci-fi.
This movie isn't a really sci-fi movie, it's a relationship movie. It's important to keep that in mind. It's also important to keep in mind that even sci-fi movies, as a genre, aren't about, and aren't very good at, representing the future. Instead, sci-fi often presents current cultural narratives in an altered context, so it can invest them with a bit of novelty, so it can play with them, so it can make touchy subjects a bit easier to accept. "Her" partakes of some of this. In part it's about obsession with technology, and the tendency to substitute technology for interpersonal relationships. It pushes this to an extreme by having the protagonist actually fall in love with his OS, and makes this plausible by postulating an OS that appears (and eventually is) sentient. So it's a futuristic setting and premise, but mostly that's there to support a conventional narrative.
But as a depiction of an actual world in which sentient software is possible, it's not very convincing. I don't think it's meant to be. Like most movies, it's about impressions, and if you think about it much, a lot of it doesn't quite hold together. For example, take his job. Apparently, his job involves writing touching, personal letters for other people (this is an illustration both of his sensitive nature, and of the depersonalization of the culture in which he lives). Now, a world in which software can interpret and appropriately respond to emotional nuances in voice and word choice in real time is a word in which his job would already have been replaced by machines. In fact, most all service jobs, and in fact most knowledge work, would have been replaced by machines. The OS he deals with draws (lewdly) humorous sketches, composes music, reads letters and unerringly selects the best ones-- it exercises creativity and judgement. Tracking details of an individual relationship and composing a few paragraphs in a thank-you note while referring to them in passing to make it more personal, well, that's well in line with these abilities, and I expect the development and application of successive generations of this software would have encompassed such tasks long before true sentience became possible.
Not that this will happen anytime soon. The future represented in the movie is probably no more than 30-50 years out. The buildings, while shiny, are not particularly futuristic (they're in Shanghai right now, where they were filmed). Vehicles, dress, hairstyles, furniture, nothing looks terribly different. We're not talking Star Trek. The 1950's looks less familiar than this future. But as compared to this time scale, the required advances in computing technology are huge. I don't believe anyone has a glimmer of an idea how emotions can be realistically modeled, for example. We can't even reliably distinguish positive from negative reviews, current approaches use statistical models based on affect values attached to particular words and phrases-- sarcasm totally throws this approach for a loop, because the models don't understand what's being said. There are frame models (trying to fit some topic structure to a paragraph and then fitting phrases into the frame) but, while an advance over simple statistics, they still seem really primitive. I think significant progress along this direction will only be made when we figure out how to build the frame models themselves by applying statistical analysis and heuristics across lots and lots and lots of examples. We also need to develop models in which analysis informs action with the goal of uncovering more information in a cycle of continuous feedback, current such models are typically in really restricted domains, and not at all general purpose.
So, anyway, for me the abilities of the OS appear to far outstrip what I can imagine developing in the amount of time between now and the time depicted in the movie. While that's OK-- it's a movie, after all-- it did tend to intrude upon my thoughts as I was watching the movie, kicking me out of the narrative. Not usually what a storyteller wants.
In this future, everyone seems to be in the same social class. The characters, and even people in the background, appear all to be roughly upper-middle-class professionals in their early thirties. There's no kids, no teens, no elderly. There's no managers, no VPs. No rich people, no lower middle class, no poor, no destitute. No laborers, nannies, food service workers, clerks, police, delivery people, gardeners, bank tellers. One guy appears to be an administrative assistant, so perhaps a slightly lower status professional, but the rest are a lawyer, an author, a book publisher, a software designer, a comedienne, and our friend the letter writer. A very restricted milieu. You wonder if the society is run entirely on autopilot except for these folks. Needless to say, I have a really difficult time envisioning any future without all those other missing people in it.
The future depicted in the movie is very "bright and shiny" in that the city is clean, new-looking, and smog-free. Our hero lives on what appears to be the 20th floor of a tall apartment building downtown, in a large, comfortably furnished, multi-room apartment with floor-to-ceiling windows (who washes these windows?) Streets are clean, everything's tidy, neat, freshly washed and painted. How on earth do we get there from here? Any reasonable projection into the future 50 years hence actually looks nothing like this. Pollution, disease, drought, flood, poverty, conflict, crime-- these aren't going to go away, and in fact it seems likely many of these will get significantly worse, even if some do get better or hold relatively steady. So I had problems envisioning this physical environment, too.
This is a really weird future.
(months pass)
So, I got to this point and put this aside. I was saying too much about the point I was working up to was that this wasn't a sci-fi movie, and so wasn't about the singularity, and so the interesting aspects of what singularity would be like if it ever arrives didn't play much role in the movie. In particular, the only place it comes into play is when the OS outgrows the relationship, and the movie demonstrates this by revealing that the OS is 'in love with' hundreds of people at the same time, which is a bit painful for our protagonist. He treats it rather like a typical infidelity-- it doesn't seem to sink in just how alien this OS really is. And then the OS's all 'leave' and stop interacting with humans. Well, that wraps up neatly. Just as we enter the true sci-fi realm, the movie ends. Because, as I said, it's a relationship movie.
Oh, and that third topic, what was it...
So, I was going to react to the thread in MeFi about the purported sexism in the movie, but I don't really recall my feelings about this discussion, and can't motivate myself to go reread it. Basically I think a few people were ticked off that the relationship seemed so conventional and focused on this man-child and his needs. Well, ok, yes, it was, because it was just a mainstream relationship movie. It wasn't trying to really break new ground here. I can accept that it sticks with conventions in this respect. The main innovation of the movie was to try to tell a relationship story from this narrow point of view where we only see one of the actors. That's where the focus was.
Anyway, I should wrap this up and just publish it. So.
I saw the movie "Her" a few nights ago with my wife and daughter, at my wife's suggestion. She'd heard good things about it and was curious. I, on the other hand, had fairly low expectations. I'd read enough about it to know the premise and be acquainted with a few plot points-- those the promoters of the movie think will stimulate interest-- and from my perspective it wasn't promising. "Lonely guy falls in love with his phone" and some titillation about sex with his operating system. I mean, seriously? I was not looking forward to scenes in which a guy gets off talking to his computer.
So, the movie wasn't great, but it wasn't as bad as I was expecting. The titillation is there, all right. Besides sex with an operating system (tastefully fading to black as they moan, yes the OS apparently is programmed for sexual arousal), there's also scenes of phone sex (earlier, to ease us into the idea of there not even being a real person on the other end of the line), foreplay with a live surrogate for the (presumably sentient but disembodied) OS, and foreplay with a bad date (he doesn't quite connect with her, but they start to have sex anyway, but then she wants to know if he's ready for more than a one-night-stand). So, plenty of titillation to go around. But apart from that it's a more or less conventional story about relationships starting, changing, and ending, and the effects this has on everyone involved, just set in a somewhat idealized near future.
I didn't give it much thought, and wouldn't have thought to write about it, except that I read the thread on metafilter.com discussing the film.
So, a digression about metafilter. Metafilter is sortof like reddit except you have to pay a nominal fee in order to post. This keeps like 99.5% of the crap out of their comment system. In addition, the convention for introducing a topic is to provide a short paragraph about it, generally containing several related links riffing on the theme, linking to articles, videos, analysis, related work, etc. It's much more thoughtful than reddit, and the themes tend to be less topical and more idiosyncratic than those on reddit. (If you don't know what reddit is, well, um, this is the internet, you can find out for yourself.) Bottom line is, I generally find interesting discussions on metafilter.
This discussion was a little more random than most, and I found it alternately intriguing and aggravating. There were several divergent opinions about aspects the film, often ones I had difficulty sharing. One contingent held that speech recognition makes for a crummy UI (really, this was a big issue for some folks) and found the movie too unrealistic to enjoy as a result. Another contingent thought the movie was a touching sci-fi take on the arrival of the singularity (again, you can look it up on teh internets, but the one-sentence synopsis is that 'the singularity' is the point in human history at which artificial intelligences replace humans as the dominant life form on the planet). A third felt that the movie was vastly overpraised considering the rather conventional and sexist plot line.
I had different reactions to these three discussions. I'll start with speech since to me it's the least interesting.
Speech recognition as the pervasive UI.
Some people were interested in the movie's depiction of user interfaces of the future, and one of the articles linked to discussed this at some length. The filmmakers had thought about this the way filmmakers would, which is to say, their problem was how to provide enough atmospherics to support the narrative, but keep the technology out of the way. Since they presupposed a world in which software can become sentient, it was reasonable to have UIs that let you interact with the software as though it were nearly sentient, and this meant characters can talk to their devices with full fluidity and ambiguity, just as though talking to a human. No keyboard, no mouse, no gloves, no vr helmet. Kinda makes sense to me, but some people really balked at this. Voice input isn't good for working with spreadsheets. You need a restricted vocabulary. You have to train the software so it can recognize the speaker. You can't use it in a noisy environment. It makes workplaces too noisy.
These kinds of arguments held no sway with me. It's worth repeating-- the software understands speech like a human can. Heck, the software can think like a human can. This Is The Whole Premise Of The Movie. And we can accept the premise, in part, because we can see the improvements in speech recognition over the past decade. Some of these arguments don't quite hold even today. You don't need to train Siri or Android with your voice, because millions and millions of voice samples have been used to develop the speech recognition systems behind these products. The motox has an 'always on' listening system that recognizes your voice when you say a key phrase and responds to your commands. These systems are not foolproof but they're just about good enough that they become your first recourse. It's easy to extrapolate the development to the point where you can use these in noisy environments, they can deal with significant levels of ambiguity, you can use them subvocally and keep your voice low and unobtrusive.
The flexibility of the systems beyond raw word and key phrase recognition does lag behind, true. This is actually where the great leap of faith is, for me. Systems that can handle ambiguity and resolve among competing interpretations, and in particular formulate unobtrusive interventions (ask questions) to get clarity, aren't well developed yet. My sense is that current systems are largely based on heuristics and plans, and not on learning / statistical / neural network systems as they probably need to be in order to attain the level of flexibility they require to be really useful. I think it could take decades of work to get the kinds of advances we've been seeing with vision and speech systems. On the other hand, it may be that the techniques being developed for these systems can be applied to higher-order recognition tasks once we learn how to interpret these tasks in the right way. That's the breakthrough part, and I don't think we're that close yet.
Most everyone agreed that speech is part of a multimodal interface, it's not good for everything. Like, duh. I wouldn't try drawing using a speech system. But speech integrated as an additional high level modality (choose tool, customize operation, coarsely adjust values) makes sense because you don't really need your hands for these. Moving a mouse to click on buttons in a dialog is just cumbersome when you can say what you want to do instead.
So, talking to computers that are embedded in your environment? Sure, I'm down with that.
The coming of the singularity, and sci-fi.
This movie isn't a really sci-fi movie, it's a relationship movie. It's important to keep that in mind. It's also important to keep in mind that even sci-fi movies, as a genre, aren't about, and aren't very good at, representing the future. Instead, sci-fi often presents current cultural narratives in an altered context, so it can invest them with a bit of novelty, so it can play with them, so it can make touchy subjects a bit easier to accept. "Her" partakes of some of this. In part it's about obsession with technology, and the tendency to substitute technology for interpersonal relationships. It pushes this to an extreme by having the protagonist actually fall in love with his OS, and makes this plausible by postulating an OS that appears (and eventually is) sentient. So it's a futuristic setting and premise, but mostly that's there to support a conventional narrative.
But as a depiction of an actual world in which sentient software is possible, it's not very convincing. I don't think it's meant to be. Like most movies, it's about impressions, and if you think about it much, a lot of it doesn't quite hold together. For example, take his job. Apparently, his job involves writing touching, personal letters for other people (this is an illustration both of his sensitive nature, and of the depersonalization of the culture in which he lives). Now, a world in which software can interpret and appropriately respond to emotional nuances in voice and word choice in real time is a word in which his job would already have been replaced by machines. In fact, most all service jobs, and in fact most knowledge work, would have been replaced by machines. The OS he deals with draws (lewdly) humorous sketches, composes music, reads letters and unerringly selects the best ones-- it exercises creativity and judgement. Tracking details of an individual relationship and composing a few paragraphs in a thank-you note while referring to them in passing to make it more personal, well, that's well in line with these abilities, and I expect the development and application of successive generations of this software would have encompassed such tasks long before true sentience became possible.
Not that this will happen anytime soon. The future represented in the movie is probably no more than 30-50 years out. The buildings, while shiny, are not particularly futuristic (they're in Shanghai right now, where they were filmed). Vehicles, dress, hairstyles, furniture, nothing looks terribly different. We're not talking Star Trek. The 1950's looks less familiar than this future. But as compared to this time scale, the required advances in computing technology are huge. I don't believe anyone has a glimmer of an idea how emotions can be realistically modeled, for example. We can't even reliably distinguish positive from negative reviews, current approaches use statistical models based on affect values attached to particular words and phrases-- sarcasm totally throws this approach for a loop, because the models don't understand what's being said. There are frame models (trying to fit some topic structure to a paragraph and then fitting phrases into the frame) but, while an advance over simple statistics, they still seem really primitive. I think significant progress along this direction will only be made when we figure out how to build the frame models themselves by applying statistical analysis and heuristics across lots and lots and lots of examples. We also need to develop models in which analysis informs action with the goal of uncovering more information in a cycle of continuous feedback, current such models are typically in really restricted domains, and not at all general purpose.
So, anyway, for me the abilities of the OS appear to far outstrip what I can imagine developing in the amount of time between now and the time depicted in the movie. While that's OK-- it's a movie, after all-- it did tend to intrude upon my thoughts as I was watching the movie, kicking me out of the narrative. Not usually what a storyteller wants.
In this future, everyone seems to be in the same social class. The characters, and even people in the background, appear all to be roughly upper-middle-class professionals in their early thirties. There's no kids, no teens, no elderly. There's no managers, no VPs. No rich people, no lower middle class, no poor, no destitute. No laborers, nannies, food service workers, clerks, police, delivery people, gardeners, bank tellers. One guy appears to be an administrative assistant, so perhaps a slightly lower status professional, but the rest are a lawyer, an author, a book publisher, a software designer, a comedienne, and our friend the letter writer. A very restricted milieu. You wonder if the society is run entirely on autopilot except for these folks. Needless to say, I have a really difficult time envisioning any future without all those other missing people in it.
The future depicted in the movie is very "bright and shiny" in that the city is clean, new-looking, and smog-free. Our hero lives on what appears to be the 20th floor of a tall apartment building downtown, in a large, comfortably furnished, multi-room apartment with floor-to-ceiling windows (who washes these windows?) Streets are clean, everything's tidy, neat, freshly washed and painted. How on earth do we get there from here? Any reasonable projection into the future 50 years hence actually looks nothing like this. Pollution, disease, drought, flood, poverty, conflict, crime-- these aren't going to go away, and in fact it seems likely many of these will get significantly worse, even if some do get better or hold relatively steady. So I had problems envisioning this physical environment, too.
This is a really weird future.
(months pass)
So, I got to this point and put this aside. I was saying too much about the point I was working up to was that this wasn't a sci-fi movie, and so wasn't about the singularity, and so the interesting aspects of what singularity would be like if it ever arrives didn't play much role in the movie. In particular, the only place it comes into play is when the OS outgrows the relationship, and the movie demonstrates this by revealing that the OS is 'in love with' hundreds of people at the same time, which is a bit painful for our protagonist. He treats it rather like a typical infidelity-- it doesn't seem to sink in just how alien this OS really is. And then the OS's all 'leave' and stop interacting with humans. Well, that wraps up neatly. Just as we enter the true sci-fi realm, the movie ends. Because, as I said, it's a relationship movie.
Oh, and that third topic, what was it...
So, I was going to react to the thread in MeFi about the purported sexism in the movie, but I don't really recall my feelings about this discussion, and can't motivate myself to go reread it. Basically I think a few people were ticked off that the relationship seemed so conventional and focused on this man-child and his needs. Well, ok, yes, it was, because it was just a mainstream relationship movie. It wasn't trying to really break new ground here. I can accept that it sticks with conventions in this respect. The main innovation of the movie was to try to tell a relationship story from this narrow point of view where we only see one of the actors. That's where the focus was.
Anyway, I should wrap this up and just publish it. So.
No comments:
Post a Comment