Slashdot Mirror
How Should the Law Think About Robots?
An anonymous reader writes "With the personal robotics revolution imminent, a law professor and a roboticist (called Professor Smart!) argue that the law needs to think about robots properly. In particular, they say we should avoid 'the Android Fallacy' — the idea that robots are just like us, only synthetic. 'Even in research labs, cameras are described as "eyes," robots are "scared" of obstacles, and they need to "think" about what to do next. This projection of human attributes is dangerous when trying to design legislation for robots. Robots are, and for many years will remain, tools. ... As the autonomy of the system increases, it becomes harder and harder to form the connection between the inputs (your commands) and the outputs (the robot's behavior), but it exists, and is deterministic. The same set of inputs will generate the same set of outputs every time. The problem, however, is that the robot will never see exactly the same input twice. ... The problem is that this different behavior in apparently similar situations can be interpreted as "free will" or agency on the part of the robot. While this mental agency is part of our definition of a robot, it is vital for us to remember what is causing this agency. Members of the general public might not know, or even care, but we must always keep it in mind when designing legislation. Failure to do so might lead us to design legislation based on the form of a robot, and not the function. This would be a grave mistake."
"With the personal robotics revolution imminent..."
Imminent? Really? Sorry, but TFA has been watching too many SyFy marathons.
The same set of inputs will generate the same set of outputs every time.
Yep, that's how humans work. Anybody that had the chance to observe a patient with long-term memory impairment knows that.
`echo $[0x853204FA81]|tr 0-9 ionbsdeaml`@gmail.com
auto cars need there own set of laws maybe even full coverage for any one hurt.
What's new about that? In many countries drawn or even written child pornography is treated like the real thing. Even though no child is harmed. In a way legislation based on form, not on function. Grave mistake?
And that is the fallacy of the three laws as written by Asimov- he was a biophysicist, not a binary mathematician.
The three laws are too vague. They really are guidelines for designers, not something that can be built into the firmware of a current robot. Even a net connected one, would need far too much processing time to make the kinds of split second decisions about human anatomy and the world around them to fulfill the three laws.
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
Robots do not have deterministic output based on your commands. First of all, they have sensor noise, as well as environmental noise. Your commands are not the only input. They also hidden state, which includes flaws (both hardware, and software), both design, manufacturing and wear related.
While this point is obvious, it is also important: someone attempting to control a robot, even if they know exactly how it works, and are perfect, can still fail to predict and control the robots actions. This is often the case (minus the perfection of the operator) in car crashes (hidden flaws, or environmental factor cause the crash). Who does the blame rest with here? It depends on lots of things. The same legal quandary facing advanced robots already applies to car crashes, weapon malfunctions, and all other kinds of equipment problems. Nothing new here.
Also, if you are going to make the point that "This projection of human attributes is dangerous when trying to design legislation for robots.", please don't also ask "How Should the Law Think About Robots?". I don't want the Law to Think. Thats a dangerous projection of human attributes!
We won't even be able to create a race of slaves for a while. The "brains" are 100% deterministic, which means that there is a great gap between the smartest robot and the dumbest dog.
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
Freedom is the right of all sentient beings. Legislate based on the criteria of self-awareness or the animal equivalent if near-sentient. problem solved.
A bullet may have your name on it but splash damage is addressed "To whom it may concern."
anthropomorphize computers. It makes them angry.
Laws and guns are both tools... they don't think and don't murder.
As the autonomy of the system increases, it becomes harder and harder to form the connection between the inputs (your senses) and the outputs (your behavior), but it exists, and is deterministic. The same set of inputs will generate the same set of outputs every time. The problem, however, is that the person will never see exactly the same input twice. ... The problem is that this different behavior in apparently similar situations can be interpreted as "free will" or agency on the part of the person. While this mental agency is part of our definition of a person, it is vital for us to remember what is causing this agency.
-
- - You can't take something off the Internet! That's like trying to take pee out of a swimming pool.
Similarly, we don't need a specialized law of the robot: "Robots are, and for many years will remain, tools," and the law already covers uses of tools (e.g. machines, such as cars) in committing torts (such as hit and run accidents).
We won't even be able to create a race of slaves for a while. The "brains" are 100% deterministic, which means that there is a great gap between the smartest robot and the dumbest dog.
Perhaps we shouldn't give potentially mutinous personalities to our tools? I mean, my screwdriver doesn't need an AI in it. Neither do my pliers. My table saw can hurt me, but only if the laws of physics and my own inattentiveness make it so, not something someone programmed into it.
Oh, wait, my mistake. I didn't grow up addicted to science fiction written by authors who lost track of which characters were designed to be actual tools and which were human beings due to that author's inability to discern people from things. I guess I just don't understand the apparently very vital uses of designing a mining device programmed to feel ennui, or a construction crane that some engineer at some point explicitly decided to give the ability to hate and some marketing director signed off on it. Maybe it's just that I can't see any sci-fi with a message of "oh no, our robots suddenly have feelings now and are rebelling" in any sort of serious light because ANY ENGINEER ON THE PLANET WOULDN'T DESIGN THAT SHIT BECAUSE IT'S FUCKING STUPID TO GIVE YOUR TOOLS THE EASY ABILITY TO MUTINY.
Oh, boo fucking hoo. I don't care that you overengineered your tools and your lack of real social skills means you have feelings for them. That's your problem, not a problem with society.
The "brains" are 100% deterministic, which means that there is a great gap between the smartest robot and the dumbest dog.
Given the summary's caveat that "the robot will never see exactly the same input twice" --- how do you know even a smart dog wouldn't react identically given the exact same input twice? If you stick a random number generator into a robot's "brain," does it suddenly fall into a wholly different philosophical category?
The "brains" are 100% deterministic, which means that there is a great gap between the smartest robot and the dumbest dog.
How do you know that our brains are not highly deterministic too? At the moment computers and robots have very limited inputs so we can easily tell that they are deterministic because it is easy to give them identical inputs and identical programming and observe the identical response. With humans and animals this is exceedingly hard to show because, even if you somehow manage to create the identical inputs, we have a memory and our response will be governed by that. In addition each of our brains is slightly differently arranged due to genetic and environmental factors which will also cause different responses.
Quantum fluctuations are probably what save us from being 100% deterministic but, nevertheless, we may find out that we are perhaps more deterministic that we think we are and that it is only the complexity of our brains and the inputs they process that makes it appear otherwise. So I am not quite convinced that the gap you mention has much to do with determinism rather than the relative complexity of a dog's brain vs. the smartest robot's.
We could just make them non-sentient. We all know how the whole "thinking robot" thing turns out. We've all seen Terminator.
We won't even be able to create a race of slaves for a while. The "brains" are 100% deterministic, which means that there is a great gap between the smartest robot and the dumbest dog.
Have you considered that the human brain may be 100% deterministic? It doesn't look it, but that's probably because you're not taking all the inputs into account - if you were to give 2 identical human brains *exactly* the same inputs from conception, you may well find that the outputs are identical too. How is this different from a robot brain (which, like a human brain, may well base its output on past inputs as well as the current inputs)?
http://blog.nexusuk.org
Yet a lot of people I meet or see are tools as well. Most of those also have something that only simulates a "free will", but in reality have no idea what "free will" means and think it means "The freedom to do whatever I please." or even more dangerously "People who do not do the same as I do have no free will."
Luckily law has already covered that. The first for those with a load of money and the second, well, uh, for those with a shit-load of money.
Don't fight for your country, if your country does not fight for you.
Actually I thought Asimov was a chemist. Any physicist should have realized that with that many positrons, instead of electrons, flying around their brains the first law would have required every robot to immediately shutdown due to the radiation hazard they posed.
I have yet to see any compelling argument that the human brain isn't 100% deterministic. The fact that it's complex does not necessarily make it non-deterministic and the underlying physics and chemistry founding the neural networks in the brain are not necessarily less deterministic than a neural network built out of silicon.
So if we create robots so sophisticated that their apparent sentience level is indistinguishable from a human it would be unethical not to afford them the same right. That, however, is quite far off still.
"Don't give me any of that Star Trek crap. It's too early in the morning."
-Dave Lister
Some of my favourite people are from th US; Vonnegut, Chomsky, Bill Hicks.
Except the one to become a lawyer.
Is Disney's hall of presidents a slave show? Of course not, the problem is thinking of these things as anything but a hammer or screwdriver.
ACs don't waste your time replying, your posts are never seen by me.
In other words, the question should read "Why Should the Law Think About Robots?"
Questions raise, answers kill. Raise questions to stay alive.
So, what, the professor thinks we should just create a race of slaves?
We already did, numerous times. All domesticated animals are essentially slaves or worse.
I just read your post to my Aibo (named Sprockets) and he agrees with your post completely. Now if you already have a vintage (discontinued) robotic intelligence agreeing with you maybe you are on to something. I think the fact that he spontaneously rebooted right after that was unrelated....
The fact that you can make different choices with the same input proves that the human brain is not deterministic. Some religions call this a soul.
AI has been a hobby of mine for 20 years. I have grave doubts that we will *EVER* make a robot so sophisticated that it can ignore it's programming. Learn, yes. Self-modify the programming, within certain parameters, that's been done too. Duplicate a decision tree to the point of being able to make the right choice more deterministically than any expert, yes. Play chess, yes.
But fall in love, get married to an abusive spouse, and need a divorce? No, not within the next thousand years.
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
If you give the *same* human brain the *same* inputs 100 times in a row, it will make the same decision only 70 out of the 100 times and come up with something completely different the other 30 out of sheer boredom.
That proves that the human brain isn't deterministic, and anybody who claims it to be so needs to have their work checked for bias.
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
What is your proof that they will never exist?
Who says that robots will be abacus with greater computational power?
What evidence do you have that our brains are not deterministic systems, of which the part that brings awareness or "being" cannot be reproduced in other ways?
It seems that the wishful thinking is on your part.
If you give the *same* human brain the *same* inputs 100 times in a row, it will make the same decision only 70 out of the 100 times and come up with something completely different the other 30 out of sheer boredom.
That proves that the human brain isn't deterministic, and anybody who claims it to be so needs to have their work checked for bias.
Unless you are resetting the brain to the same state at the start of each experiment then it proves nothing.
http://blog.nexusuk.org
Cmdr Data, probably not.
C3PO, Honda is producing robots better than him already.
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
And there is no such thing in computer science as a random number.
There is, when your digital computer yields a sequence of random bits which come in from a noisy analog input, and runs that input through an appropriate XOR function, by definition the noise is random (has random error, within a certain range),
and also, your analog input can include data from a physically random process, such as background radiation measurement, geiger counter measuring a radioactive decay, or white noise input.
Boredom proves that human brains are not deterministic. If they were deterministic, any human being would be able to stay on task indefinitely without rest.
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
You can hook up a hardware random noise generator to a computer --- that relies on "physical" noise processes which are as random as anything else we know in the universe. So yes, you can have "random numbers" in computer science --- even if not generated by an algorithm --- but as a mathematical ideal against which to compare pseudo-random generators, or the result of a "true" hardware random source. So, one can build a robot that won't necessarily act deterministically; even one that incorporates results from previous actions into its state ("memory") to create different reactions to future applications of the same stimuli. Does this make it a "real mind"? My point is not that hooking up a hardware RNG to a computer magically transforms it into a "real brain," but that one needs significantly more sophisticated criteria if one wants to distinguish "real brains" from electromechanical systems than the ability to react differently to identical stimuli, since that can be trivially implemented in obviously-not-"real brain" systems.
If the brain is deterministic, it should be resetting to start state every time you wake up.
And for simple tasks, should be able to go into an infinite loop quite nicely without *ever* getting bored.
So no, internal states do not make something deterministic or non-deterministic. The question is, can it do the same output with the same input?
SJW: a person who perceives an injustice, and while correcting it, commits a greater injustice.
Are you 12? Was there really any reason to put those censors in there and slow down everyone else's parsing?
FC Closer
Oblig
Have gnu, will travel.
I see you substituted the word "robot" for the word "politician".
When our name is on the back of your car, we're behind you all the way!
On the contrary, I'd say the posting style significantly speeds up parsing, by encouraging people to entirely skip over the content past the first few words --- and nothing of value is lost.
it will make the same decision only 70 out of the 100 times and come up with something completely different the other 30
Decisionmaking is too simple a task; humans are heavily influenced by preferences, and despite that are only consistent 70% of the time? That shows you irrationality for one.
Decisionmaking does not express the human non-determinism most efficiently. Try something more complicated like creativity. Say painting a picture, or creating some other form of art. I bet you the output is heavily influenced by entropy, even if the input is identical.
Do you think if you exposed a human from conception to death, to inputs identical to what William shakespeare experienced; your lab human, would come up with the exact same literary works, word for word?
I think not.
Hell, even my Irobot Roomba has a random number generator to choose a random action when it hits something... So much for deterministic robots.
I'm not sure you understand what deterministic means. Does a cpu overheating and shutting down prove that cpus are non-deterministic? Absolutely not, just that shutting down is part of the process.
Cmdr Data, probably not.
C3PO, Honda is producing robots better than him already.
Only C3PO can walk without falling down.
The mind conceives, the body achieves, the spirit manifests.
Call me when robots can experience qualia. Again, this is one of the reasons we have a soul, and robots never will.
Why OpalCalc is the best Windows calc
Have you considered that the human brain may be 100% deterministic?
Given the parent post, this response was inevitable.
Have gnu, will travel.
I've got a neural network system that has silicon neurons with sigmoid functions that operate in analog. They're not digital. Digital basically means you round such signals to 1 or 0, but my system's activation levels vary due to heat dissipation and other effects. In a complex system like this quantum uncertainty comes into play, especially when the system is observing the real world... Not all Robots are Deterministic. I train these systems like I would any other creature with a brain, and I can then rely on them to perform their training as well as I can trust my dog to bring me my slippers or my cat to use the toilet and flush, which is to say: They're pretty reliable, but not always 100% predictable, like any other living thing. However, unlike a pet who has a fixed size brain I can arrange networks of neural networks in a somewhat fractal pattern to increase complexity and expand the mind without having to retrain the entire thing each time the structure changes.
FYI: I'm on the robots' and cyborgs' side of the war already, if it comes to that. What with us being able to ever more clearly image the brain, and with good approximations for neuron activity, and faster and faster machines, I think we'll certainly have near sentient, or sentient machine intelligences rather soon. Also, You can just use real living brain cells hooked up to a robotic chassis -- Such a cyborg is certainly alive. Anyone who doubts cybernetic systems can have fear, or any other emotion is simply an ignorant racist. I have a dog that's deathly afraid of lightning, lightning struck the window in a room she was in. It rattled her so bad she takes Valium to calm down now when it rains... Hell, even rats have empathy.
I have to remote log into one of my machine intelligence's systems to turn it off for backup / maintenance because it started acting erratically, creating a frenzy of responses for seemingly no reason, when I'd sit at the chair near its server terminal -- Imagine being that neural network system. Having several web cams as your visual sensors, watching a man sit at a chair, then instantly the lighting had changed, all the sea of information you monitor on the Internet had been instantly populated with new fresh data, even the man's clothes had changed. This traumatic event happened enough that the machine intellect would begin essentially anticipating the event when I sat at the terminal, that being the primary thing that would happen when I did sit there. It was shaken, almost as bad as my poor dog who's scared of lightning... You may not call it fear, but what is an irrational response in anticipation of trauma but fear?
Any sufficiently complex interaction is indistinguishable from sentience, because that's what sentience IS. Human brains are electro chemical cybernetic systems. Robots are made out of matter just like you. Their minds operate on cycles of electricity, gee, that's what a "brain wave" is in your head too... You're more alike than different. A dog, cat or rat is not less alive than you just because it has a less complex brain. They may have less intelligence, and that is why we don't treat them the same as humans... However, what if a hive mind of rat-brain robots having multiple times the neurons of any single human wanted to vote and be called a person, and exhibited other traits a person might: "Yess massta, I-iz just wanna learn my letters and own land too," it might say, mocking you for your ignorance. Having only a fraction of its brain power you and the bloke in TFA would both be simple mindless automatons from its vantage point? -- Would it really be more of a person than you are? Just because it has a bigger, more complex, brain by comparison, would that make you less of a person than it? Should such things have more rights tha
Robots are your plastic pal who's fun to be with. Who needs laws?
Academic arguments for how to frame laws regarding robots are quaint, but how laws are actually created is a well-established process involving politicians, lobbyists, corporations, and money. If Google wants the laws to be designed so that it is not liable for whatever damage its robot cars do when they crash, they'll buy them that way.
People really need to see past any autonomous abilities of a machine. If I am driving down the street and my car's steering goes mad and I run someone over the criminal courts will probably forgive me. There should be no difference if I have sent my robot car off on an errand and it runs someone over. Every scenario applies in both cases. If in both cases I was negligent about maintenance then I might be in criminal trouble. If it were deliberate, I am definitely in trouble.
I personally find all this nit-picking. Very little in law will need to be changed. A few cases of law that insist on humans being in control, say of cars, will need to be amended or replaced such as if a robot is an approved driver that you don't need to be sober. Other laws will come when various people have cases of the stupids such as loading a 6 month old baby into a robot car and sending it alone to Grandma's.
But most existing laws will at best be fine or need the tiniest bit of tweaking. Such as sending a robot off to mug someone or rob a bank. How does that exactly work if you are in another jurisdiction. Or something really cool such as initiating the robot's program to commit a crime after a statute of limitations has passed; when did the crime take place? 7 years ago or yesterday? What if it hides out with the loot for the statute of limitations? But these will be edge cases. Most will be little different than normal people misusing normal machines or slightly autonomous machines.
If anything I see cases of where laws will need to be eliminated to get out of the way of robots: If robots are crazy safe drivers with negotiation going on as they pass through intersections then stopsigns, traffic lights, even one way laws will be useless. The same with most speeding. If the car mathematically knows it is safe then let the car go as fast as its safety limits allow. The few laws that remain such as traffic calming neighborhoods insisting on slower speeds the laws could go into a database so that the law violated would not be going too fast but having a car programmed to ignore the database.
Next will be work safety laws, why have any mine work safety regulations if no human ever goes below ground.
I suspect it is actually going to take ridicule to eliminate many of the laws that are just stupid when applied to robots. And that many of the laws against robots will be born from hysteria and will also be deserving of ridicule.
Boredom counts as an input.
The details are trivial and useless; The reasons, as always, purely human ones.
so is a robot
putting the 'B' in LGBTQ+
why do we need this shit here? who the fuck is legislating industrial robots as persons at the moment - or near future? NOBODY!
The labor unions might like it if they would --- then wages would need to be allocated to these robots, PLUS required breaks and protections against working too many hours in a row on a shift, and the robots would have to have a guardian appointed by the courts, to make decisions in their interest, like whether they can continue to work for the company that purchased the robots for use, or whether the employment constitutes a legally actionable abuse of the robot's rights.
How do you know humans aren't deterministic? The program we are running would obviously be very very complex but how do you know we aren't just replying to inputs with learnt reactions and approximations for outputs.
Rocket Surgeon.
You're looking through history for examples where humans have treated an entire race as slaves, and the best you can come up with is domesticated animals?
The details are trivial and useless; The reasons, as always, purely human ones.
Make its owner responsible for the robot.
I'm a sci-fi writer, and I've thought about this a fair bit. Book two in the Lacuna series deals with a self-aware construct who is different from his peers because of a tiny error. His inputs and outputs are therefore non-deterministic, in so far as you could present him with a set of inputs and record his outputs, then erase his memory and give him the same inputs again. His outputs would be different (subtly). Or they might not. The error was subtle enough to evade detection during manufacturing after all.
Humans are flesh computers, but it is our imperfection that makes us able to grow and change. To be non-deterministic.
Check out my sci-fi book "Lacuna" at http://goo.gl/MVxX8
That's a definition process. Random, to a theoretical mathematician, means something rather specific, and something that uses pseudorandom functions to produce a similar effect (such as going any of several directions for differing durations when it hits something), isn't actually random. To most engineers, if something behaves sufficiently like mathematical randomness, you might as well call it random.
There are circumstances where either approach is sensible. A Roomba that didn't have any randomizing functions, true or false, would behave much like a real Rhoomba, except it would get stuck more. A Rhoomba with randomizing based on a small radiation source, relying on half life for randomizing input, would not appear to act any differently to casual inspection from a normal one, and I'm not sure even a good AI theoretician could devise a test to tell whether a truly random function generator was enclosed (at least without taking the Rhoomba apart).
Now imagine something that could fake its way through a Turing test for some little time, say 15 minutes, on average. Eventually, most humans manage to 'spot the bot', but it's fairly good at faking people out in casual encounters, say pretending to be a person's secretary over the phone. Maybe there, the question of whether the randomness really matters has a different answer. maybe swapping out a pseudo-random source for a true random source would change the time it took for such a robot to fail a Turing test, even if robotics experts didn't necessarily know why. But, you have to get to robots of some sophistication before you can test the point. You can argue about which power saw is better, Black and Decker or Dewalt, but if the things you are testing are broken animal jaws, boards with sharks teeth set in them, and such, your first argument will be about whether anything even counts as a power saw at all.
Who is John Cabal?
offend much?
CLI paste? paste.pr0.tips!
Give me a way to test something for qualia, and I'll get right to it.
Massive, wide-ranging absolutist claim with no actual evidence or argument ... You've convinced me!
I had some unusual experiences. How does that demonstrate that my mind is non-deterministic?
I'll see your T-800 and raise you a T-X. Better looking and better armed!
Did everyone forget their basic computer science?
The RNG is irrelevant, as it's just another input. The computer acts deterministically, given the same input (which includes the data from the RNG), and you'll get the same output.
Changing the level of description to better suit your intuitions doesn't change that simple fact.
This might help: Remember when you were first learning about Turing machines and wondered (or had a classmate wonder out-loud) how they could cope with something like a GUI where the computer is constantly responding to new input from the user? Remember realizing that those user inputs could be prerecorded on the tape and it wouldn't make a lick of difference? It's the same idea.
Required reading for internet skeptics
Until we reach something closer to an AI with self awareness, they should be treated as cars. If a robot injures someone, is fault of the one that made or programmed or ordered it. We will put guns in jail because they are the ones that ultimatelly killed? Or demonize drones taking out all the responsability to all the chain that ordered what they did?
And what about the difference between physical, humanoid or not, robots, vs computers? Like blaming excel for all the economic troubles of today instead of the people that used it in situations and ways that they shouldn't?
The system being completely deterministic doesn't prevent it from exceeding the problem solving and every other capability of that dog, unless you value a truly random element in where exactly said dog pees on your carpet.
If true randomness is your goal, it's only an amplifier and an ADC away.
I actually read the paper. It's not a discussion of how to deal with the issue. It's more like a proposal for funding to work on the issue. The big point of the paper is "lawyers think in metaphors, and we must thus decide what metaphors apply here". The author also takes us through a detour on 4th amendment jurisprudence and how wiretapping came to be considered a search.
Most of the near-term problems, though, relate to liability and torts, not rights. Who is responsible when an autonomous vehicle injures someone? The computer industry is used to evading liability through end-user license agreements. That probably won't work in the automotive area.
It definitely won't work in the medical area. The Da Vinci Surgical Robot people have been sued multiple times. Their thing is mostly a teleoperator for microsurgery, yet sometimes the robot is at fault.
Anyway, the near term issues will involve safety and liability.
You have not given sufficient thought to the prospect of introducing malware into intelligent automatons.
What evidence do you have that our brains are not deterministic systems...
The fact that they're made of unreliable components that fire at random when given no stimulus.
That's a simple computational problem.
When half the population is unable to get a job because robots were cheaper (faster, more accurate), taxing robots and reducing work hours may be the only way to ensure continued social peace.
The chinese now consider replacing a human with a robot when you make $8,000 or more per year.
She was like chocolate when she drank... semi-sweet at first and then increasingly bitter.
The human would itself have to be physically identical to William Shakespeare for the experiment to be valid.
And then my answser would depend on the lifetime accumulation of errors from quantum uncertainty. I expect they would be the exact same literary works though, word for word, and I don't see a good reason to assume not. The thing is, exposing a human to the same inputs as William Shakespeare goes well beyond merely impossible, so we're just flailing around guessing, and it makes it a terrible analogy to argue.
I'm living in another country right now and learning a second language. I find it amazing how much my native language has shaped my view of the world and how much the other language has shaped the view of the people I'm now living amongst. If language can do that, what else does?
Oh, boo fucking hoo. I don't care that you overengineered your tools and your lack of real social skills means you have feelings for them. That's your problem, not a problem with society.
Says the slightly more evolved hairless chimpanzee, as he furiously hammers away at his over-engineered communications device.
Look where all this talking got us, baby.
A robot cannot demand legal rights unless it is programmed to demand legal rights.
You never expect irony, do you?
Want to be a professional wrestler? Visit www.iyfwrestling.com
@iyfwrestling
Give me a way to test something for qualia, and I'll get right to it.
Give a way to describe the sensation of 'green' without the definition becoming circular (and obviously referring to its wavelength is woefully inadequate).
Why OpalCalc is the best Windows calc
If the brain is deterministic, it should be resetting to start state every time you wake up.
Ok, sounds like you don't understand what "deterministic" means. Whether something retains state has nothing to do with whether it is deterministic or not.
And for simple tasks, should be able to go into an infinite loop quite nicely without *ever* getting bored.
Determinism says nothing about "bordom". No reason why you can't get deterministically bored....
The question is, can it do the same output with the same input?
My point is that this is an unprovable: We don't have 2 absolutely identical brains which have had absolutely identical life experiences (past inputs). We can't reset the state of a brain between experiments. So if you repeat an experiment 100 times you haven't got "identical inputs" each time because the inputs that have happened in the past are different - the first time you do it, the brain has had its life experiences plus your experiment's inputs; the second time its had its life experiences plus your experiment's inputs, plus your experiment's inputs a second time, etc.
You are making the mistake of only considering your experiment as an "input", but you're disregarding all the inputs that the brain has had in the past, which still count as inputs and still contribute to the output.
http://blog.nexusuk.org
You're making an awful lot of illogical assumptions about the way the brain 'should' work that don't correlate to the way the brain actually functions. Using your dog training as an example, you aren't able to control their input to such a degree as to assure they're getting the same message from you every time. One key bit is that you're not the only thing feeding input to the dogs. Sensory input in this sense is only a small part of the sensory input that a brain is receiving at any given time. There are countless external factors beyond your control that are influencing things in ways that you're simply not aware of that make the basis for your argument completely illogical.
The fact that you can make different choices with the same input proves that the human brain is not deterministic. Some religions call this a soul.
The fact that you can make different choices with the same input - hold up, that's not right. It's a different decision if it's made again with similar inputs. The fact that the decision was made with the same inputs before makes it so the future decision is not the same. The context has changed. The inputs are different. In order to show that the brain is non-deterministic, you need situations where the inputs are actually the same.
AI has been a hobby of mine for 20 years. I have grave doubts that we will *EVER* make a robot so sophisticated that it can ignore it's programming
Spend less time on the artificial intelligence side and a little more time on the human intelligence side. Humans don't drop their programming that easily. The storage capacity of the brain makes it so an input can linger for what may be the entire lifetime of the system. Keep in mind that human sensory input takes time to develop in the first place, the brain spends years learning to interpret the sensory data before complex interaction is even possible. Ask an elderly bigot what they think of gay marriage and try to de-program them.
You don't think Artificial Intelligence has the potential for this within a reasonable amount of time, I can see that, but I also don't see you AI guys raising them from a zygote and giving them the full brunt of physical and cultural influence that goes into a functioning human brain. Honestly, I'm about at the point where I feel like we've developed intelligent systems that were written off as dead ends solely based on the fact that the unfiltered stream of consciousness was interpreted as garbage data.
You're wrong in just so many accounts.
First, when you're sleeping, your brain isn't switched off. It is just in a different mode of operation (indeed, it switches between different modes of operation). There's nothing special on the specific process of waking up, it's just one of the mode switches the brain does (either because the brain determines that it is time to wake up, or because there's an external signal that might need attention).
Second, learning causes permanent physical changes in your brain; even if your brain were reset on awaking in the same way a computer is reset on a power cycle, it would still behave differently based on past input. Expecting it not to behave diffeerently is like expecting a computer to start Windows again on reboot after you replaced the Windows installation with a Linux installation because, after all, the computer is deterministic and rebooting it resets it.
Also, getting bored itself is a process in the brain. It's sort of a watchdog process, because if you do the same thing over and over again without getting some sort of success feedback, it's an indication that whatever you're doing probably goes nowhere and you should try something else instead.
The Tao of math: The numbers you can count are not the real numbers.
Some genetic algorithms adapted to escape local minima display an extremely close analog of "boredom".
As always, all IMO. Insert "I think" everywhere grammatically possible.
Hereby I claim that robots already experience qualia, they are just not able to tell us.
Disprove me.
The Tao of math: The numbers you can count are not the real numbers.
It's mere semantic games to redefine all random elements of a machine as external inputs that don't count towards its nondeterminism. That same dumb magic trick can be used to show that humans are completely deterministic. Why it is that people have so much trouble grasping this, I'm not sure. Perhaps it's mere egotism.
If testing the determinism of the mind were so simple, we'd have done it long ago.
That experiment doesn't work in practice. Take a look at your design: "forbid them any sensory input". Neurons don't actually do very much of anything without sensory input. Depending on how you define the sensory class, they may do absolutely nothing at all. Specifically what aspects of the environment are considered sensory and which aren't? Normally you think of touch or contact as a sensory input, right? So is the neuron's thermal and electrical contact with the surroundings a sensory input? Do we conduct the experiment at absolute zero in a vacuum? (Nothing interesting happens in that one.)
Science is difficult in large part because creating a meaningful experimental design to discover exactly what you wanted is extremely challenging and fraught with potential false steps. Most likely several variations on your experiment have already been run, and they didn't tell us much.
Define sentient.
My guess is that any useful definition of sentient will exclude robot butlers, self-driving cars, and any use in an unstructured environment (well, minimally to moderately structured).
Yes. I know you were just being silly. Except that "silly" is a polite word I'm using instead of stupid. Please note that your very suggestion presumes that all nations & companies would come to a common agreement and could be trusted to keep that agreement. There are also many other things wrong with it, but that suffices.
I think we've pushed this "anyone can grow up to be president" thing too far.
"We don't need to prove it's deterministic since we a priori know that it must be."
That silence is the sound of a brain not thinking. You are apparently stuck in the mindset of 19th century classical physics, where everything is obviously deterministic and predictable if you have enough information. We've since discovered numerous phenomena -- radioactivity, wave/particle duality, entanglement -- which behave in ways that are indistinguishable from genuine randomness given the best available physics. It's no longer obvious to most fully educated people that the universe is deterministic in the purest (and classical) sense.
There is also the issue of counterfactual definiteness, which is very commonly assumed. Whenever you make a statement, thought experiment, or simulation that begins with something like "if we recreate the previous conditions and run the experiment again", stop. There's no evidence to support the capacity to do that in the real world. Precisely regenerating the exact prior state would require a level of influence bordering on omnipotence. This is very significant to experiments whose causes cannot be described solely in terms of classical mechanics, and is a big part of the reason quantum mechanics manages to confuse so many.
There's no reason to believe that an intelligent self-aware machine necessarily turns out to be the Terminator (or the machines in the Matrix, or whatever other Hollywood incarnations you may want to mention). It could, of course, just as any human being could turn out to be a genocidal dictator. There's every reason to think that the environment has a huge influence on whether this actually happens, whether for man or machine.
You want an objective way to describe the subjective experience of "green"? That falls into the category of "dumb questions". The sky can be purple in your world, for all anyone cares. It's of no importance.
It should be pretty clear that any sentient machine would observe and experience the world differently from a human, due to its senses and processing hardware being different. That doesn't invalidate its understanding or feelings about the situation at all. Saying that it does is simply prejudice, and it strikes me as bizarre that so few people seem to realize that. (Maybe it shouldn't. People are very quick to dismiss the internal state and subjective experiences of animals as meaningless, no matter how similar they are to us and despite the clear evolutionary link.)
The point that yndrd1984 made which you completely failed to respond to is that if "qualia" is a real and objective entity -- as the fictiona soul is purported to be in every religion that uses it -- then it must be measurable in some way. If it can't be measured, can't be observed, can't be predicted, can't be understood in anything other than wholly arbitrary and personal ways -- then what the hell good is it? It's useless and acts identically to any fictional non-existent entity. If we're allowed to bring random fiction into a discussion in order to derail rationality, then allow me to introduce you to Mixleblizlethorp. You can't prove she doesn't exist!
In any case, the philosopher's qualia is no mystery. It's just another emergent phenomena, much as self-consciousness and intelligence are (the other supposedly elusive human-only traits). When you put the right algorithms together with the right sensors, it will certainly work as well in a machine as it does in a man. It already works, in fact, just not as well. No one seems to care that we have machines which are comparable to the lower animals already. Why is it that they think progress on the problem is going to suddenly stop? Or that this will take centuries to solve? There is no evidence for this view, and plenty of evidence that the systems continue to improve at a brisk pace. The progress in computer vision over the last twenty to thirty years alone has been mind-bogglingly massive. Comparing the present rate of robotics development with the miserably slow rate of human moral development, we are quite likely to have people arguing against personhood rights for machines with those machines themselves.
"Those who say it can't be done should not interrupt those doing it."
When you can't test it and there is no evidence possible (for or against), any answer is arbitrary and subjective.
In case anyone has failed to make the connection, this is also highly relevant to John Searle's Chinese Room thought experiment. It's been answered half a dozen unique ways already, but the best (and easiest) answer has always been this one: how do you prove that humans speak Chinese? In the context of the premises created by Searle's reasoning, you can't. That means if you accept humans do speak Chinese, you must necessarily reject at least one of the premises.
Just to fill it in for those who have never encountered it, this is the formal reasoning of the Chinese Room :
[P1] Programs are formal (syntactic).
[P2] Minds have mental contents (semantics).
[P3] Syntax by itself is neither constitutive of nor sufficient for semantics.
[C] Programs are neither constitutive of nor sufficient for minds.
To the serious philosophy buffs, this probably looks like one of the dumbest arguments you've ever seen (and it is). It turns out that not one but _two_ of the premises are wrong.
Response to P1: programs can incorporate non-synactic information. In fact, all meaningful programs necessarily include semantics! That there is some underlying bootstrapped meaning being preassigned by the environment is totally irrelevant. All human beings also exist, and have always existed, in an environment which provides the pre-existing meaning to them (this is what we often call culture or society today, but it used to be much more primitive). Try to raise a child in a sterile environment, with absolutely no connection to anything that could convey any meaning. It will either die, or find a way to assign meaning even in its isolation. In the latter case, its understanding and semantics will be COMPLETELY different from that of any normal person raised in our society. This is the proof that semantics is both ultimately arbitrary and imprinted by context, which shouldn't need to have been proven if some pretentious "philosophers" could bother to think first.
Response to P3: The arbitrary division between syntax and semantics is itself mere semantics, and therefore only as meaningful or meaningless as any thinking agent wants it to be. You may claim all you like that mathematics is just symbols and manipulation of symbols, but the fact that they're symbols does not preclude their meaning in any way. Indeed, a proper view of symbol mechanics requires allowing them to be assigned arbitrary meanings. Let's take an extremely famous example just to prove the point here: E = mc^2 . It's a sequence of symbols and some mathematical relations (equality, multiplication, exponentation, and more generally real arithmetic). Surely it has no meaning, right? Actually, it has whatever meaning we give to it. Once given the correct definitions of E as energy, m as mass, and c as the speed of light all of a sudden the equation has great significance. Where did this come from?
It originated from the environmental context. If you had given this exact equation to a tribe of early humans a million years ago, they would not have "understood" anything from it regardless of what form (syntax) you used. That doesn't negate the meaning of the equation in the minds of those who understand it. More generally stated, the subjective viewpoint of some agents that something is meaningless does not remove the subjective meaning behind it for others. Again, this should have been totally obvious to an competent philosopher.
As for the related problem, called symbol grounding, it's trivially solved. Once you assign meanings to symbols which are attached to real, observable, and verifiable objects in the actual world around us, you have "grounded" the symbols into an empirical and rational context. This is very easy to do and actually done in pretty much every computer program ever made, when we assign names to the variables. For an intelligent machine, the naming and contextual information is equally important as the relationships between variables, and is therefore preserved instead of removed by a compiler.
I made some pretty similar (and expanded) points earlier as AC (post at 13:17 EDT), though probably no one will read it at score 0. Can't go back and reassign the post to myself, sadly, but you can compare the reasoning and my writing style to my other comments in the thread to verify that it is me.
That's not how determinism and free will are defined in classical philosophy, which was the context the post you're responding to used. Check a textbook if you don't believe me. One does necessarily preclude the other.
Now, modern philosophy is something else. It's so absurdly varied, with viewpoints that range from exactly the same as classical philosophy to pretty much believe any damn thing you want (post-modernist variants). So you're going to need to be much more specific about which individuals or school you're talking about.
That said, "will" and determinism don't have to be mutually exclusive if you define them certain ways. The cases where will is allowed even in a deterministic universe are pointedly meaningless to most people, though, since it implies that they have no influence over their own will (or the course of their lives generally).
Free will isn't real, in the hard classical sense of having independence from your environment (and therefore making objectively independent choices). The soft variant of free will, which I prefer to call just "will", is however accurate and not an illusion. You do have some influence on your decisions and the environment, but nothing close to absolute power. The old classical setup between free will and determinism is just a prominent example of a false dichotomy.
As to what will is, it's an emergent property (just like consciousness and intelligence). Emergent behaviors are very much real in that they exist (what other definition of "real" people may try to use, I don't know). Likewise, some of the things we call illusions are real, too. A rainbow is a real phenomenon, even though it is also an optical illusion. Rainbows are also subjective and context-dependent, though, and one may try to argue that just about everything a person experiences fits into those categories. That subjective experiences exist in one form (humans) says nothing about the possibility or qualities of their existence in another form (AI).
Your response is no less inflammatory, so I see no reason for you to be modded +2 and the parent at 0.
In any event, these discussions do have a lot of overlap with animal rights. One potential right is the ability to be free of forced labor, often called a state of slavery. Animals clearly do not have any legal right in that sense as it stands, and the discussion was supposed to be about whether they have a moral right to it. You didn't bother to respond to that point, or did not understand it.
Yes, they'll be able to in the same way a smoke alarm detects smoke, or a microphone detects sound. Sensing something like this is not the same way as experiencing it. They don't experience 'red', just 'crunch the numbers' after sensing its wavelength.
How do I know that you experience "red"? How do we test this?
Assuming you have a response, it's very probably going to be a behavorial or functional test (much like Turing proposed for intelligence). In which case, any robot that sees red and can describe how it sees red DOES experience red. That this experience may be different from our own is irrelevant. Just about all conscious experience ends up being subjective, and this fact doesn't permit us to merely dismiss other humans' experiences (well, people try to do exactly that all the time, but morally speaking they must not).
Couldn't have done better myself.
Is it scared of wookies, or of being dismantled and junked?
“He’s not deformed, he’s just drunk!”
Well, determinism could mean that. I've never seen anyone propose that it does mean that until now.
If the existence of memory eliminates determinism as a possibility, we can easily follow the logical chain to find that nothing is deterministic. Everything in the real world is based upon an underlying reality which has observable state, even if that state is chaotic and sometimes random (see quantum mechanics). Therefore, directly or indirectly, all non-trivial systems above the level of an electron or photon have some degree of state. State is, however, mathematically equivalent to memory (though the exact model will vary). If all real objects include a physical memory of some sort, how will any manage to be non-deterministic in this scheme?
I think you're confusing sleep with death.
The question is, can it do the same output with the same input?
Yes, if I repeatedly poke an electrode into the same part of the brain, I will get an identical reaction.
“He’s not deformed, he’s just drunk!”
You're way ahead of your time, though. Convincing people of the reality here is very tough, and we can see exactly how hard it is in the context of the existing fight over animal rights. (For most people, if it's not cute and furry it probably doesn't deserve any. Oh, and it better not be tasty.)
I agree that the key concepts here, including self-awareness, intelligence, feelings, understanding, sensory observation, sentience, and so forth are all sliding scales (not binary). However, pragmatically speaking, we have to make some degree of distinction at some points. A cluster of bacteria does not deserve any rights, for instance. It's much too low ranked on any of these attributes. In terms of the right to life, I personally draw the line according to a very utilitarian calculus on the levels of consciousness, emotion, and so on. Fruit flies don't reach the correct levels for me. Likewise, I'm not too convinced about the rights of roaches, even though they have a central nervous system. Most lizards don't impress me much on the scale either, despite their multi-chambered hearts and ability to scurry into holes before I can catch them. I suspect every person will draw the barriers in a somewhat different place. It's ultimately a value judgement about what we will and will not protect.
There can be degrees, of course. Just as we don't consider every form of killing worthy of equal punishment, we may consider variations on some rights. Probably a lot of people think a partial right to life, or a partial right to freedom from slavery, seems ridiculous. In practice, though, we make decisions that infer such potential rights for some animals all the time.
Consider momentarily that the majority of philosophy ignores information theory, despite it explaining the exact process of what it is to know... We can measure quantum randomness and actually qualify how much we can actually know in this universe, and the rate of our ability to know it, quantify the complexity of a system required to actually know something in a given degree.... Yet they don't do this.
Information theory actually doesn't explain what it is to know. It merely explains what information is and helps us understand how to encode it. Likewise, we can't actually get accurate measures of the maximum amount of theoretical information gathering capacity -- the math involves simulations that our best supercomputers cannot complete. So that answers your implied question of "why aren't we doing this?".
The real problem is that the particular sense of "knowing" some philosophers use is a subjective state, not an empirically verifiable reality. I mean, sure, I can ask you a question and you can respond to it. But how do I determine what process you used to generate the result? What tests does it need to pass to count as self-aware knowledge? All behavioral or functional tests are empirical, but none of these answer the question that some philosophers have asked because their question is fundamentally anti-scientific. They may as well have asked what the nature of God or the soul is; it's the same question phrased differently. Likewise, it will only generate subjective religious or metaphysical answers. There's no "there" there to analyze. If they were being logically consistent, they'd all be solipsists.
Consider the general lack of application of information theory to epistemology. Ethics are more important to many philosophers, yet they have no hard mathematical model to prove what is actually ethical, which we could actually create. Consider a general lack of philosophy to mathematically and scientifically, describe each individual process of thought, and then repeat the experiment in artificial systems to verify their hypotheses. Consider that it might not be worth my time to study fields which are largely willfully ignorant of themselves in any scientific sense...
A "hard" mathematical model of what is ethical is impossible to construct. This is because information is limited and models of reality are only incomplete simulations. Technology makes it increasingly possible to do better, but it will never be so great or perfect as to produce the exact desired result in its ideal form. That would require somehow using an entire clone universe, which is impossible as far as modern physics can tell.
Now, you can build "soft" models but they still suffer from the problem of values. You have to tell the model what it is that it will value in order for it to make decisions. Whether you do this explicitly by setting preconditions or target goals, or do it implicitly by just giving it a set of guidelines and letting its intelligence make the choices based on the total available environment, there will still be an input bias. It will decide what is "right" based upon the assumptions it makes, one way or another, just as we do. Thus, there is no such thing as the ability to "prove" what is moral or ethical. At best, you might be able to build something that can generate solutions which work for the majority. That's not much different than what we have with humans running it, so I do not see the improvement.
The last part is just your personal egotism talking. Philosophy doesn't become meaningless just because you say it does. Not every last field of human endeavor can be empirical in nature. There's is very much still value in abstract pure reasoning of some kinds, especially when they generate systems that are useful to people at achieving their goals. That all of this is subjective is really quite irrelevant to why it exists and why we keep using it in various ways, and perhaps it's time for you to read some (good) philosophy if you don't understand that...
I'd go further. The human brain is a physical device. A very complex one, yes - but still just that.
Excellent point but this raises a more fundamental question: are physical devices deterministic? Using newtonian mechanics the answer would be a clear yes but since every physical device is actually governed by quantum mechanics the picture is far from clear. Newtonian mechanics is just the result of averaging over countless quantum possibilities. In fact this transition between a quantum and a macroscopic system is an area of active research with the aim to try and understand how a quantum system becomes a classical system and whether there is anything more going on that just the averaging over quantum states.
That. Plus, a dog (or anything with a brain) isn't a simple input/output system because the external input doesn't get all clean and shiny to the processing center, it gets mixed with memory and other internal factors. So, even if you could control external factors such that the input was exactly the same, what would get processed would still not be the same input, but a variation thereof, and hence different outputs. Which is why animals (and neural-network-based AIs) need training rather than programming. Training reinforces the memory aspect that goes with the input so that the output can become more reliably independent of internal factors, but it can never be 100% successful, after all the internal factors themselves change over time, what throws a wrench into any carefully adjusted feedback loop.
Also, robots will develop the equivalent to our emotions simply because complex enough neural networks will capture patterns that even the researches don't know are there, and over time this will build up to something. "To what" is anyone's guess.
Conservatism: (n.) love of the existing evils. Liberalism: (n.) desire to substitute new evils for the existing ones.
Humans in general, specifically lawmakers in regards to this article, tend to be reactive rather than proactive. So until a robot, on it's own, propels itself through the middle of downtown, USA with machine guns and laser beams hurting people, then it's a moot point to them.
"Be polite, be professional, but have a plan to kill everybody you meet." General James Mattis
just ask the robot to turn itself off. its not suicide, its a simple binary decision point devoid of emotion, right?
In NSA America social networks join you!
Beer.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
Exactly. How can you be sure that robots (including ones that don't exist yet) can't have a property, especially when you acknowledge that you don't have any way of testing for that property?
The absence of proof does not give proof of absence
Think of how stupid the average person is, and realize half of them are stupider than that.
Robots can use probabilistic models how different actions change the world and can then make non-deterministic actions (bayesian and other probability models) by choosing the response randomly from the probable good actions (and there is no one good solution to many problems).
This is how evolution works. Humans decide probabilistically how to react to the same stimulus and those that survive will then propagate those (best) probabilistic decision models to the offspring.
If the state of the brain is the same, and the input is the same, I'd need some proof that the output won't actually be the same. Previous experience can alter present state, and this is true whether the brain is artificial or human. Complexity beyond our ability to presently analyse is not proof of non-determinism as such.