If you are going to be around Palo Alto next Thursday evening, you might consider attending a panel discussion on robotics and law at Stanford Law School. I’ll be on a panel alongside some very interesting and knowledgeable folks taking up varied aspects of robotics (my particular interest is robotics and war, but the panel will be considering many areas of robotics). The particulars are below the fold.
(Update:) Here’s the assigned topic for comments, following up on Laura’s opening comment … should the panel discuss the Three Laws? Are they a useful ethical/legal frame for dealing with robots in various aspects of human life? Did Asimov lead us all astray by proposing them? Should we instead avoid discussing them altogether? What would you propose would be a better set of principles/laws/guidelines for robot-human interactions?
(I’ll also be giving a lunch talk/discussion that same day sponsored by various student organizations at SLS specifically on robotics and armed conflict. And thanks Glenn for the Instalanche!)
Stanford Law School
5:30 p.m.-6:30 p.m. Reception (Student Lounge)
6:30 p.m. – 7:45 p.m. Panel (Room 190)Once relegated to factories and fiction, robots are rapidly entering the mainstream. Advances in artificial intelligence translate into ever-broadening functionality and autonomy. Recent years have seen an explosion in the use of robotics in warfare, medicine, and exploration. Industry analysts and UN statistics predict equally significant growth in the market for personal or service robotics over the next few years. What unique legal challenges will the widespread availability of sophisticated robots pose? Three panelists with deep and varied expertise discuss the present, near future, and far future of robotics and the law.
Panelists:
- Kenneth Anderson, Professor of Law, American University; Research Fellow, Hoover Institution on War, Revolution and Peace at Stanford University
- Paul Saffo, Consulting Associate Professor, Stanford University; Visiting Scholar, Stanford Media X; Columnist, ABCNews.com
- F. Daniel Siciliano, Faculty Director, Arthur and Toni Rembe Rock Center for Corporate Governance; Senior Lecturer in Law and Associate Dean for Executive Education and Special Programs, Stanford Law School
- Moderator: M. Ryan Calo, Residential Fellow, Stanford Center for Internet and Society
Co-Sponsored by the Arthur and Toni Rembe Rock Center for Corporate Governance and the Stanford Program in Law Science and Technology‘s Center for Computers and Law (CodeX). Click here to register.
Laura(southernxyl) says:
Are y’all going to talk about the Three Laws?
My husband and I watched “Colossus, the Forbin Project” last weekend. I hadn’t seen it since I was a girl and didn’t remember it well. I kept asking him – WHOEVER THOUGHT THIS WAS A GOOD IDEA? Inconceivable.
So some science fiction (Asimov) did a fairly good job of anticipating how AI would have to be designed for interaction with humans, and some didn’t.
Here is an amusing and fairly recent conception of human/robot interaction and how it could end up.
November 6, 2009, 11:57 amSome dude says:
Speaking of the three laws. They were flawed. Many a plot in Asimov’s revolved around their brokenness. I fixed them.
1. A robot may not injure a human being
November 6, 2009, 12:07 pmor, through inaction, allow a human being to come to harm.2. A robot must obey any orders given to it by human beings, except where such orders would conflict with the First Law.
3. A robot must protect its own existence as long as such protection does not conflict with the First, Second Law.
Mike McDougal says:
Is the robot going to step on my cat to keep itself from falling over?
November 6, 2009, 12:28 pmfirst history says:
That’s also one of my favorite films–I’ll see it anytime, anywhere. A few years ago I saw a screening of a pristine print with a Q&A of Eric Braeden. There is a whole genre of “when computers go bad” films like Demon Seed (Proteus), 2001: A Space Odyssey (HAL 9000, though the movie’s themes are broader), Dark Star (Bomb 20 and Mother), and of course the Terminator series (Skynet).
Skynet, while fictional, could be the logical culmination of Donald Rumsfeld’s “Network-centric” conception of the battlefield. The current development of autonomous UAVs, vehicles, and missiles could lead to systems that select their own targets. The question then is who is responsible for something that goes wrong.
November 6, 2009, 12:39 pmAnon says:
Some dude — nice “duty to rescue” move!
But seriously, what do Asimov’s three laws have to do with real robots? The Asimov stories all dealt with make-believe — sentient AIs that would almost certainly pass a Turing Test (any flavor). Asimov was interested in exploring the interactions between autonomy, self-awareness, and morals, not the boring engineering problems that we have today.
Real robots are things like automated welding machines on an assembly line. The “brains” are finite state machines — the behavior is deterministic based on the inputs. They don’t think, they just drag a welding rod across predetermined locations.
Don’t get caught up in the romance of the word “robot,” these things could be built as an elaborate mechanical clockwork. See Babbage’s difference engine. They’re not, though, because that would be very, very expensive.
It makes as much sense to apply Asimov’s three laws to real robots as it does to apply them to staplers, kitchen knives, and automobiles.
If I stab you with a knife, is that because the knife wasn’t designed with Asimov’s three laws? What if I run you over with my car?
Automobiles, of course, have robot-like features — self-park is an apparent one.
But what is a computer-controlled fuel injector? It is no less robotic than the self-park feature or a robotic welder on an assembly line, once you strip away the romance. Do we expect engineers to design fuel injectors to predict when a human could be harmed and shut off the fuel supply?
November 6, 2009, 12:41 pmFub says:
Well, some of us applications folks had a good laugh when Colossus ordered the systems guys, um, executed for attempted debugging.
November 6, 2009, 12:47 pmDaniel Chapman says:
1. A robot may not injure a human being or, through inaction,
allow a human being to come to harm.So a robot bus driver could continue on his merry way as he plowed over pedestrians along his set route? I don’t think this is as elegant a fix as you think it is.
To Anon the Engineer: http://www.xkcd.com/534/
November 6, 2009, 12:48 pmAnon says:
first history,
Dark Star – very fun movie. Bomb 20: “And I saw that I was alone. Let there be light.”
Don’t get all romantic on the real stuff, though. Network-centric doesn’t mean what you think it does, and nothing we have today is any more autonomous than your Play Station (which isn’t, but can give a nice illusion).
Are you worried that World of Warcraft game play across the internet is going to result in the formation of SkyNet?
November 6, 2009, 12:55 pmdisconnect says:
Non sequitur. The bus driver’s actions (driving) are directly causing injury. “Through inaction” would be (e.g.) a mugging occurring down an alley; this robot would need to ensure that human’s lack of injury (which could
You know what? I’m not going to spend the last five minutes of my lunch arguing this. Instead, what I originally came here to post: Jokes Made By Robots, For Robots.
Knock, knock.
November 6, 2009, 12:57 pmWho’s there?
A robot.
Oh, shit.
Sarcastro says:
Has always been thus.
November 6, 2009, 1:08 pmtraveler496 says:
To keep the discussion useful and grounded you might constrain it to periods during which the leading legal minds are not predominantly artificial.
(Prognostication beyond that time in most any field is difficult; and the leading legal minds of that era will in any case have limited use for what is said during this one.)
November 6, 2009, 1:17 pmSome dude says:
Unless you order it not to.
November 6, 2009, 1:22 pmfirst history says:
No, I am not worried that anything we have today is more autonomous than a PlayStation (which I don’t have), but I am looking at the future 20-30 years from now. See this excerpt from P.W. Singer’s Wired for War at the Wilson Quarterly.
And the danger is real. A robotic anti-aircraft gun (an Oerlikon GDF-005) in South Africa had an “oops” moment in 2007 when something went wrong and it started firing uncontrollably, killing nine SA soldiers and wounded 14:
Oops.
November 6, 2009, 1:27 pmInstapundit » Blog Archive » TALKING LAW AND ROBOTICS at Stanford Law School next week…. says:
[...] TALKING LAW AND ROBOTICS at Stanford Law School next week. [...]
November 6, 2009, 1:59 pmAnon says:
first history:
Assuming arguendo that there was a software glitch that caused Oerlikon “oops” — that’s no different than a mechanical failure. That has nothing to do with Asimov’s three laws and more to do with the fuel injector from my previous example.
Suppose the drive-by-wire throttle sticks wide open on your Lexus and you crash, killing yourself and others. Is that a failure that is reasonably addressed by Asimov’s three laws? Or is it merely a combination of poor design and/or operator error?
BTW, your Wilson cite doesn’t support your case. Read it carefully and figure it out on your own.
November 6, 2009, 2:01 pmTim Oren says:
The Three Laws are as big a rathole as the Turing Test, and for much the same reason: There is little that resembles human ‘reasoning’ going on in the constructed devices, be it computers or robots. The current advance in robotics has more to do with increasingly powerful signal and image processing – and increasing reluctance to pay/risk humans in certain roles – than any higher order logic or intelligence. What there is of the latter is highly task specific, and there is little economic incentive to invest in enhancing it beyond the needs of the task and market. There are few robotics roles, with the notable exception of health and eldercare, that require even a simulacrum of conventional human interaction. Given that, some argument beyond a handwave at the future is required if one is to take the Three Laws seriously.
November 6, 2009, 2:28 pmTweets that mention The Volokh Conspiracy » Blog Archive » Law and Robotics Panel at Stanford Law School -- Topsy.com says:
[...] This post was mentioned on Twitter by JP Fendrich, Tawana Thomas. Tawana Thomas said: The Volokh Conspiracy » Blog Archive » Law and Robotics Panel at …: My husband and I watched “Colossus, the F.. http://bit.ly/2WBngh [...]
November 6, 2009, 2:55 pmTexas calling says:
Awesome!
Looks like their setting the stage for future Software-Americans to dislike attorneys as intensely as current Wetware-Americans do today.
November 6, 2009, 3:08 pmAnon says:
Tim Oren,
You miss the subtle genius of the Turing tests — Turing was effectively postulating that unless something could engage in higher order reasoning (whatever that is), it could not pass the test.
Remember, your brain is a finite collection of neurons, each with a rather simple electrochemical functionality, in a mappable, albeit large, mapping. But whatever is going on inside your melon — you could pass a Turing test.
A machine wouldn’t have to have “human reasoning.” Just as you are smart enough to learn an alien culture to pass their version of the Turing test, so could a truly intelligent machine.
November 6, 2009, 3:31 pmFub says:
Point understood.
But here’s a not entirely impossible scenario: spambots and malware gone wild.
Currently there are huge numbers of internet connected machines infested with various malwares that spew spam at the behest of their criminal human controllers. Some of these spambots are self-replicating in the limited sense that they probe other machines in some pseudorandom way to find new machines to infest. Some of the malwares even scan their hosts to destroy competing malwares.
Ridding the internet of these bot populations has so far proved impossible, even though they are controlled by (criminal) humans. A huge portion of available network bandwidth is already consumed by the operation of these criminally controlled botnets.
Now, what if they became uncontrollable, even by their criminal creators?
Consider appearance of a bug (or feature) in some population of self-propagating bots, whether introduced accidentally or intentionally, whether by spammers or by security personnel, or even by a random error in an infested machine.
The hypothetical bug causes the newly placed bots to ignore all command and control input from their original sources, and to observe commands from other bots propagated through some other channel than the original.
The self-propagating botnet is now autonomous until humans can discern the new command and control channels and mechanisms.
The botnet already has the capability to reduce available internet bandwidth by generating unwanted traffic. The primary way known to stop it is the power OFF switch on millions of computers, or alternatively to shut down millions of high bandwidth channels.
Through some process akin to natural selection among individual bots, a network of bots could evolve that could (at least temporarily) shut down commerce and communication that relies on the existing internet structure.
When population numbers and replication rates (with some errors) are high, something akin to evolution through natural selection, actually a stripped down model of biological evolution, could occur spontaneously.
I’ll agree that my proposed scenario is far-fetched, but it is not impossible. Bugs happen, even in malware.
November 6, 2009, 3:51 pmuberVU - social comments says:
Social comments and analytics for this post…
This post was mentioned on Twitter by mglickman: Law and Robotics Panel at Stanford Law School http://ff.im/-b5UDL...
November 6, 2009, 4:54 pmLaura(southernxyl) says:
I know, I have robotic things at work and have for some time. But if this were the only kind of robots being contemplated, it would hardly be worthwhile to convene a panel to talk about the legal aspects, would it?
I disagree about taking out “through inaction allowing a human to come to harm”. These laws are supposed to be hard-wired into EVERY robot, not only the high-functioning humanoid things. They are simple so that the rule hierarchy is easy to deal with, without a lot of cogitation. Suppose that you have a kitchen robot that knocks the rat poison over and it gets into the sugar bowl. (I know someone who was accidentally poisoned by rat poison in the sugar bowl, so don’t laugh.) If it has very primitive programming, it won’t necessarily solve that unanticipated problem unless it can’t, through inaction, allow a human to be harmed.
…………
I LOVE Dark Star. I quote that bomb all the time – “Nevertheless.”
…………
One of the funny things about “Colossus” was the briefcase Forbin carried around – nowadays it would be a laptop, but no one thought of that. Another was that when they were worried about Colossus linking up with the Russian model, they were only worried about information being passes – the concept of somebody in Russia hacking into the system, or viruses or worms, never occurred to anyone either.
November 6, 2009, 6:47 pmkumquat says:
Another point about Asimov’s giving his robots a “duty to rescue” – a number of his robot stories mentioned that public opinion was a problem for the robot company. People simply did not trust robots, and the three laws were designed to convince everyone that robots were safe.
Can you imagine what a PR disaster it would be for a robot manufacturer the first time a robot with Some dude’s modified 3 laws let a human die who could have been saved, because it didn’t take the rescue initiative on its own and wasn’t given a rescue order in time?
November 7, 2009, 12:51 amAnatid says:
Yes. We already have perfectly good names for this kind of behavior: bugs and viruses. There’s a reason we don’t call them web tigers.
Presumably, if a computer could be as complex as a human brain, then it could be programmed to emulate a human brain. But why would we want to? Our understanding of consciousness, and even of sentience, is the heavily subjective result of a somewhat-efficient gestalt of hundred of interconnected subfunctions, each function somewhat adaptive. Why, for example, would we want to limit a computer’s ability to attend to multiple objects, or cause it to feel jealousy when its programmer interfaces with a different machine, or have its processing performance vary based on its level of stress? Much of the human experience is defined by our limitations, our preconceived notions. Why pass these limitations on to computers?
November 7, 2009, 4:51 ammarkm says:
You really, really don’t get it. Present day robots are incapable of recognizing a human, let alone recognizing that something might harm a human. Nor are they capable of recognizing threats to themselves. For instance, a robotic cart does not have a command against running over humans or running into brick walls – it just has a programmed routine that stops it when its sensors detect an object directly in front of it. Whether that’s effective or not depends on how good those sensors are, and sensors with a wide range and field of view plus enough resolution to not be firing false alarms all the time are expensive. So your cart will stop for a trash can placed in its path but quite likely will not detect a human walking on an intersecting course until too late. As for the 3rd law, it probably won’t have a sensor to detect a hole in the floor, but just follows a set path, and the designer trusts that the users will turn the carts off or re-route them before tearing up the floor.
Even with an unlimited budget, we can install better sensors, but we can’t write a program that makes sense of unexpected items appearing in them. It’s hard enough just to recognize one item at various angles and distances. Beyond that, even if a robot can identify a human and a bus (for instance), there is not even a hint of how to give a computer the reasoning power to recognize that a human standing in the path of a bus is in danger. Yes, I could probably program a robot to rush out and push a human from the path of a bus – but the program would execute without understanding and certainly be either underinclusive (that SUV is not a bus, so go ahead and stand in the road), overinclusive (“rescuing” humans from a newspaper blowing in the wind), and most likely both. All I could do is try to add clauses for all the situations I can imagine, but this will never ensure “sensible” reactions to situations I did not anticipate.
And I see no path from our current software development techniques to software that does understand the world.
November 7, 2009, 8:14 ammarkm says:
Now, assume a Turing-capable robot that really does understand the world around it and so can follow the three laws. In this case, the second clause of Asimov’s 1st law is disastrous. The world is full of dangers. Better collect all the humans and lock them up in padded cells, for their own good. Either the robots are too limited intellectually to recognize most dangers, or you have the dystopia of Jack Williamson’s “Humanoids”.
I’ll also note that, as a practical matter, the 2nd and 3rd laws are in the wrong order. No sane company is going to program a million dollar robot to self-destruct because the janitor tells it to.
November 7, 2009, 8:24 amLaura(southernxyl) says:
Look at the OP:
I’ve known people who have no imagination, or who were made uncomfortable by SF-type speculation. These aren’t the people who move us forward.
Wasn’t it in the 1970′s that the president of IBM said there would never be a need for individuals to have computers in their homes? It’s hard to say what people will never want to do, or see value in. Look at the doll-like robots they are developing in Japan. I wouldn’t want one, but I’m not them. (They.)
November 7, 2009, 10:19 amAnatid says:
That wasn’t a rhetorical “no one will ever want this” “why.” It was an actual, genuine question I was hoping someone here could speculate upon.
November 8, 2009, 5:41 pmVincenti Winchester says:
The artificial intelligence people at MIT are still having major problems with robots even understanding how to open a door.
It sounds easy to open a door until one realizes how many types of doors there are; the push/pull decisions; the level of the handle, etc.
The microtubules that our brains are based on function at the quantum level. At the moment we can’t even understand how these structures can function at all, because they involve superposition and entanglement.
So, the bottom line is that robots are great if we keep them in perspective.
They will always be servants that do simple things.
It is US who will program simple things to do dangerous things, like fly over a field and drop an object that goes boom.
November 15, 2009, 3:35 amTyson F. Gautreaux says:
Zune and iPod: Most people compare the Zune to the Touch, but after seeing how slim and surprisingly small and light it is, I consider it to be a rather unique hybrid that combines qualities of both the Touch and the Nano. It’s very colorful and lovely OLED screen is slightly smaller than the touch screen, but the player itself feels quite a bit smaller and lighter. It weighs about 2/3 as much, and is noticeably smaller in width and height, while being just a hair thicker.
November 17, 2009, 1:43 pm