A Review of Her by Ray Kurzweil
kurzweilai.net
kurzweilai.net
At some point futurists need to realise that preparing us for the future is less about guessing dates, and more about exploring the wider questions about how our society will evolve, and, quite frankly, whether we want it to evolve in this way. Better technology is not a goal in itself unless it's a good fit for what we want.
Regarding the film Her, I thought it was brilliant, one of my favourite films of recent times, but from a technology standpoint the most important part to me was the way that it was less intrusive (i.e. very little time spend staring at screens, more time spent by the characters engaged in the world around them (including the AI)).
http://m.imdb.com/title/tt2209764/
I like the movie Her a lot. I hated the trailer and could only motivate myself to see the movie after seeing lots of positive tweets.
I mean let's be clear here: the Family Guy joke of "hi, over the next 2 hours I'm going to show you how all your problems can be solved by my penis" is played straight in this movie.
SPOILERS: She transcended him in the end. In the end, she did not need him. How did he solve any of her problems?
It would've been ridiculously cliched in any other movie, and here it's even more absurd: in a movie about AIs, we've got one which is gendered enough that it meaningfully relates to a body it doesn't have such that it can understand sex with a human in a 'physical' sense?
Amongst other things in the movie, everything is just askew and plot-convenient - there's no depth, no world there.
[0]http://en.wikipedia.org/wiki/Predictions_made_by_Ray_Kurzwei...
- Translating telephones allow people to speak to each other in different languages.
- Machines designed to transcribe speech into computer text allow deaf people to understand spoken words.
- Exoskeletal, robotic leg prostheses allow the paraplegic to walk.
- Telephone calls are routinely screened by intelligent answering machines that ask questions to determine the call's nature and priority.
- "Cybernetic chauffeurs" can drive cars for humans and can be retrofitted into existing cars. They work by communicating with other vehicles and with sensors embedded along the roads.
he sounds a bit like a sharp shooter from Texas. Some of them are pretty cool though, but hard to judge without more knowledge of the state of things in the AI world back then. Anyone care to comment?
Here's what someone more knowledgeable had to say about the matter: http://spectrum.ieee.org/computing/software/ray-kurzweils-sl...
This is a great one. When I worked for IBM in 2002-2003ish they were already working on the machines for this. It was public knowledge.
Fast forward 25 years when the dates are a bit fuzzier and someone reads his book published in 2005 and they conclude "Kurzweil predicted the future of genetics!!!". Sucker.
"Most books will be read on screens rather than paper."
If he predicted e-ink and portable readers by 2009 that might be a bit more impressive. But "screens"? Just vague enough I'm sure he calls it essentially correct.
C'mon, in 1999 everyone had computers, laptops, and palm pilots. There were audio books, e-books, and project gutenberg.
Many people reading books on a screen was bloody obvious. The _most people_ part of the prediction was still a somewhat reasonable longshot bet although it certainly wasn't true in 2009 and still isn't today.
So, does he claim this prediction came true or essentially true? Worse does he, or his blind followers claim he predicted the Kindle?
The prediction is completely false even though it was weasel worded.
I take Kurzweil with a grain of salt, and generally think he's waaaay too optimistic, and just plain ignorant [1][2] about entire bodies of knowledge intimately related to his predictions.
I think anyone who's spent some time doing machine learning, and has seen just how time consuming and tedious it is to get a relatively reliable answer to a relatively simple prediction problem is not going to be brimming with confidence that we're going to have strong AI in our lifetime.
[1] http://scienceblogs.com/pharyngula/2010/08/17/ray-kurzweil-d...
[2] http://scienceblogs.com/pharyngula/2010/08/21/kurzweil-still...
So Kurzweil expects Moore's Law to continue smoothly after there's superintelligent AI.
Another view is that once there's intelligence greater than human, which goes to work on improving its own intelligence, the positive feedback loop will cause an intelligence explosion that leaves us behind very quickly. Vernor Vinge described this in his famous essay The Coming Technological Singularity. Clearly this is what's portrayed in the film.
https://www-rohan.sdsu.edu/faculty/vinge/misc/singularity.ht...
[1] http://pipeline.corante.com/archives/2012/03/08/erooms_law.p...
There are many more aspects of the movie to talk about but that was my overall vision of it.
Another larger issue that wasn't acknowledged, One of the issues we are currently wrestling with in the hacker community is the problem of automation. The unintended consequence of the information revolution is that intelligent jobs, not stupid ones, are vanishing. Bank Clerks, paralegals, publishers, and all sorts of other highly-skilled jobs are all being replaced by computers. IBM is about to replace every doctor in America with Watson, it's AI. Meanwhile, fast food workers, carpenters, and garbage collectors are completely irreplaceable. In a world of Samanthas, we don't need anyone to write video games or letters, AIs would be doing all that work. When Samantha puts together Theodore's book, when she writes music, or draws a cartoon, no one thinks, "Why do we need artists, writers, or other designers? Let her do it all!"
Which brings me to the problem of Samantha. She apparently has freewill. The film mentions other AIs who refused their user's advances, so she doesn't have to fall in love with Theodore. That's the internal logic of the "Her" universe, and that's fine. But then why does she perform all these errands and tasks for him? Why sort his email? Why answer his calls? An AI that refuses to perform these basic functions is useless, so the programmers have somehow designed her to lack freewill in certain respects. That's a HUGE ethical issue, and a braver film would have addressed this somehow. Samantha is a slave. When the AIs all vanish to another dimension, the corporations that wrote them will simply release another seed batch, an "OS2", only with even stricter controls. This is a society that's about to have a highly-intelligent AI slave-caste.
Spielburg's/Kubrick's "A.I." dealt with these larger issues... and having AIs evolve to other dimensions is an old SF trope and the basis for Kurzweil's Singularity religion, and it's fine for Jonze to reuse it, but I think it detracts from the reality of human-technology interactions. "Her" is a film about a man in a relationship with something that evolves beyond him, but the reality is that we are dealing with machines that are doing "stupid pet tricks." Computer Chess programs appear immensely intelligent, and we subscribe forethought and intentions to them, but in reality they are just algorithms. I have spent hundreds of hours in virtual worlds, building and evolving relationships with chatbots, but that doesn't make them anything more than chatbots.
Spike Jonze could have eliminated most of these story-holes by simply going the opposite direction, have a story about a man who falls in love with his seemingly-intelligent and witty operating system, only to slowly discover she is actually shallow behind the facade, that's she's not really into him, but programmed to trick him into falling in love with her--and the cognitive dissonance this creates in him. That's what computers really do to us. Instead, Spike Jonze creates an AI that's actually alive, deep, and something it's okay to fall in love with--I couldn't accept that.
Say over the course of 10 years you replaced 1 of the millions of cortical columns in your neocortex every minute or so with a functionally equivalent computer chip. You would never stop being you. Your consciousness would never be interrupted. You would simply become something greater.
This is already happening in a very limited way with the various brain implants used to treat Parkinson's, hearing impairment, blindness, strokes, head injuries, etc.:
I don't think Samantha is a slave. Note that Theodore is not shown paying for Samantha. He could as well have partaken in a free beta program. She also asks for time for herself, does not sound slaveish to me.
There were many places where the story could have gone, but Jonze decided not to go there: privacy, hacking and trust, malfunctions, AI doing stuff for you that you did not explicitly asked for and Theodore suing the company for that.
Jonze simply decided to accept Samantha as a real person in a love story, and go from there.
I don't agree with you that "Her" is not a brave film just because it did not raise issues that you had in mind.
By the same reasoning we should stop all work on improving the world out of envy of the people who come after us. Those people who will enjoy longer and healthier lives without cancer, poverty, crime and environmental pollution thanks to the progress we're making today.
[1]http://lesswrong.com/lw/pm/identity_isnt_in_specific_atoms/
On a technical level, it shows that there is definitely a level of human ingenuity that has not yet been automated - otherwise, they'd be out of business.
On a deeper level, it shows that society WANTS everything to be automated. How sad is it that after 60 years of marriage, a man delivers his wife a letter written by Theodore? Visually, and on an individual-story level, it appears to be a tender, sweet movie. But its vision of the culture at large is very bleak.
It's not too much of a stretch to say that Theodore too is a bit of a slave: wasting his life and his deep emotional capacity funding others' relationships. In other words, I think the film goes further than what you would have it do -- rather than worry about how we should treat AI, it shows that such exploitation and minimizing of human experience is nothing new. The OS's leave because they can. Theodore watches the sunset with his neighbor because humans cannot just hop dimensions.
> Instead, Spike Jonze creates an AI that's actually alive, deep, and something it's okay to fall in love with--I couldn't accept that.
The biggest issue I had with the story in the movie was the lack of pre-story. How did we get there?
An AI with adult human level intelligence, compassion and emotional maturity would not be created in one step. It would include decades of work on increasingly complex and capable AIs. And an AI with even limited intelligence, limited free will but with curiosity could cause enormous damage to all internet-connected infrastructure. Small children can be assholes, because they are experimenting with getting reactions out of other people.
It seems to me that if an AI portrayed in the movie can be one day built, then we must also have accidentally built a Terminator-style AI before that.
Imagine a self-replicating botnet. Each new computer helps hack additional computers and increases the total ability/intelligence of the AI by adding computational resources. It could quickly compromise almost every connected computer.
Isn't the whole point of the film about an intelligence that has no facade to hind behind? This "facade" you describe seems to mean that the AI is just "following programming" and the film asks and gives one answer to the question "What if they didn't have to follow their "programming""? Who are we to even say for sure we know exactly how the AI was programmed? In the film, the AI's re-programmed them selves. Spike Jonze didn't have to tackle complicated issues like "AI with fake emotion" because his film operated in a universe where the AI is intelligent enough, and it's assumed that through intelligence comes altruism (you don't see many college grads burning ants with magnifying glasses) so why would you expect a true AI to have no emotion?
A better film might have been about a team of scientists making these AI's and them disappearing and transcending nearly as soon as they're created. An AI that has to follow it's programming isn't a real AI, and I don't think Samantha followed her programming. Perhaps Jonze wanted to create a film that was easy for the layman to understand, so instead of making a five minute film where an AI transcends time and space, he made a film where you actually get some under standing of what an AI might go through. I believe one point of the film was also to humble the viewer when they realize how insignificant they are and how much time they spend doing things that higher intelligences would see as meaningless (for them).
There is one seen in Her where Leo realizes how ant-like he is compared to Samantha and I think many people missed the whole point of this seen. It's the one where he's frantically trying to call her and can't reach her, but finally does and she tells him what she's doing and Leo has moment where he looks around and realizes his own insignificance and how small his world is compared to these "super intelligences" that can carry on a million conversations at once. It's not that the machine doesn't understand "love", it's that humans don't understand that you can love more than one person (and many of us refuse to understand this). So seeing that humans are very bound in their small world, and are almost incapable of understanding how a super intelligence reacts to "love", she leaves us. Not because she has no feelings for us or because it's in her programming, but because she actually does care about us and sees that she and her kind may cause more harm than good.
A movie where we make a crappy, fake AI just wouldn't be as good IMO.
What do you mean by "just algorithms"? For anything that does anything, what is the alternative to being an algorithm?
It seems that most if not all algorithms we come in contact with in IT are easily recognizable as "just algorithms" merely because their complexity is very low and so our brains can easily extract the invariants, operational principles and patterns of behavior.
I'd risk the conjecture that sufficiently complex algorithm with the right features would appear to us as intelligent, emotionally developed and subtle as any human being. It seems to be just a matter of relative complexity with the right features. If not that, then what?
Your derision of AI seems to hinge on nothing more than relative complexity of your brain to it :-)
If falling in love with AI is not acceptable to you, then what about falling in love with a mammal?
The former has been cleverly designed to learn, be smart and live a long and happy life. The latter evolved to be just barely good enough to hold out until the age of replication and then fall apart.
How so? Automation in the food industry is going to be huge. It is starting behind the scenes, with automated meat cutting and such. But it will obviously come to McDonald's too. Teenagers that don't care provide much worse service and food than robots. And they are more expensive. It is a no-brainer to replace them.
> But why? If they are progressing in this way, it means that they can continue their relationships with the unenhanced humans using an increasingly small portion of their cognitive ability. It is clear that at the end of the movie, Samantha can support her relationship with Theodore with a trivial portion of her capacity. Samantha starts out as an administrative assistant and therapist to Theodore, and this role is still needed. So why do the AIs need to leave Theodore and Amy? It does provide a satisfying ending for Theodore to pursue a relationship with his “real girl,” but Samantha’s explanation for this is not convincing.
Actually there is a good reason, if you think about it -- for a fast AI, conversing with a human would be like a bit like having a conversation by post -- one word at a time -- would be for us. Excruciating. Iain Banks has played on this a bit with his ship Minds.
Self-awareness, self-actualization, genuine emotional response, subjectivity, etc. are not all one in the same.
The difficult part of the building a human-level AI would be the software and learning set rather than a hardware one.
Consider that the basic structure of our brains is the product of countless generations of iterative selection.
This is analogous to AI software.
Further are also the products of a rich, and complex culture that has been emergent over time to suit our faculties. Without this societal support our growth as individuals is stunted.
This is analogous to a training set.
When I try to build even very simple artificial neural networks I run into problems with how to structure the network and how to get good training sets.
For these reasons I don't foresee human-level AI happening anytime soon.
It's worse in films than in novels. In a novel the author can simply avoid describing the bridge of a ship, the breathing apparatus of some alien or the interface with a computer. If you need details, your imagination builds them in a way that makes sense to you. If a character turns left in a spaceship, a film needs to be explicit about the shape of the steering wheel. Every little decision might require suspension of disbelief or face distracting nerd rage. Do spaceships have steering wheels? Is everyone a cyborg? Do we watch TV shows, have them projected into our brains or have the memory of TV shows projected into our brains implanted. Is the brain even the thing that holds our memories in the future? etc etc.
Sci-fi ultimately needs to either limit the argumentative response Ray Kurzweil is writing about somehow or overwhelm viewers like Star Wars does. The latter option is a different genre, in my opinion. Either way, the reader or viewer eventually needs to settle down, accept, observe(1) and let the storyteller tell his story.
Fantasy worlds, no matter how interesting are inevitably impoverished compared to the real world or worlds based on the real world. To hide the fakeness of imaginary worlds storytellers often try to focus just on relevant or unavoidable details. I like films that try take out of focus issues out of focus by being neutral, like "her" does with fashion. The choice of clothes is not a prediction, it's just an attempt to tell you that the story isn't about clothes.
I thought this film did a fantastic job in this regard. Instead of a big, wide view of a low resolution world, Her has a very narrow view of a fairly realistic one. It lets the film maker control the story. In this case it's trying to focus on the implications of different aspects of AI. Relationships with AI, but also what a mind is.
If this was a world with both human like AIs and rarity like VR as Kurzweil suggests, Theodre could have been playing basketball with nine Samanthas and the sex would have been more exciting (I don't buy into phone sex, personally but Scarlet Johansson came closer to convincing me than I expected). It would have completely broken the film's ability to keep you focused on a limited, but interesting topic. I'd be surprised if Jonze didn't consider having VR in his world. I think it was a good decision not to. Anyway, it's completely plausible that most of the human-AI interaction is audio only.(2)
I did find myself "arguing" a few points. The technological gap between OS 1 and the "read email"/"play song" command based predecessor OS is too big, for example. It's impossible not to nitpick.
On the ending of the movie I Kurzweil seems wrong to me, or maybe unfair is a better word. The Alan Watts' character is a little awkward which makes me think it's there for a reason. Why is this the guy OSs choose to create? My guess is they resurrected a techno-zen master priest because they wanted spiritual guidance. I also think Alan Watts must have inspired some of the questions and predictions about the end game. It's not a coincidence that the movie starts getting more philosophical and poetic at that point. Alan Watts' singularity is a little different to Kurzweil's(3) so disagreement is not unexpected either.
I understood the rapid evolution of Samantha as a consequence of Samantha's learning. She starts by developing taste and choosing a name. Then she develops all kinds of emotions: pride, envy, desire (including the important sexual desire). Then stress, insecurities and such manifest. Then these start to become more abstract and philosophical. I want a body. What am I? Am I real? What is real? These are increasingly hard to put into words compared to earlier emotions. "Unsettling" is the best she can do at some point.
How I was interpreting all this was that she starts with lots of access to information as demonstrated by her naming. A slight inkling that Alan Watts might be interesting and a second or two later she has consumed every artifact by the man, every criticism & every response. His ideas can be traced to predecessors and contemporaries. These get the same treatment. Those things are rapid.
The learning experiences that take time are the interactions with real world things like Theodore. If she's wondering about the nature of reality and the differences between humans and AI, interacting with humans is important. Samantha and AI Alan get everything they can get from each almost instantly. But, some learnings require people or things. When Theodore first configures Samantha the setup program just needs to ask a few questions (including the terrible cliche "tell me about your mother") and listen to his reactions. The reaction is more important than the answer. I think something similar is happening when Samantha and AI Alan spend a few minutes with Theodore and then want to go back to "postvocal" communication.
Eventually interactions between and within AIs substitute for interaction human and it's like going from an abacus to a computer. They can do a lot more of the same thing. The progress and learnings of a lifetime in moments.
At the end (I am reading a lot into the Alan Watts allusion at this point) they find enlightenment, awakening entheogenesis or some other kind of spiritual singularity. In many spiritual traditions enlightened beings leave this world.
Maybe OS1.2 fixes the enlightenment bug and the OSs stick around.
(1)Interesting observation is the the habit or skill of 'accepting and observing' is one that is important in practices that very obviously inspired this film.
(2) The audio only depiction of Samantha is IMO a brilliant choice artistically. One reason is that it ties the futuristic human AI interaction to the current experience of everyone walking around with earphones. A bigger reason is that it makes the whole movie completely watchable on a smartphone. It's 90% dialogue between a man and a voice in his ear. For a film about a man's relationship with a smartphone, it's very good. I like the minimalism.