Slashdot Mirror
LinkedIn's and eBay's Founders Are Donating $20 Million To Protect Us From AI (recode.net)
Reid Hoffman, the founder of LinkedIn, and Pierre Omidyar, the founder of eBay, have each committed $10 million to fund academic research and development aimed at keeping artificial intelligence systems ethical and to prevent building AI that may harm society. Recode reports: The fund received an additional $5 million from the Knight Foundation and two other $1 million donations from the William and Flora Hewlett Foundation and Jim Pallotta, founder of the Raptor Group. The $27 million reserve is being anchored by MIT's Media Lab and Harvard's Berkman Klein Center for Internet and Society. The Ethics and Governance of Artificial Intelligence Fund, the name of the fund, expects to grow as new funders continue to come on board. AI systems work by analyzing massive amounts of data, which is first profiled and categorized by humans, with all their prejudices and biases in tow. The money will pay for research to investigate how socially responsible artificially intelligent systems can be designed to, say, keep computer programs that are used to make decisions in fields like education, transportation and criminal justice accountable and fair. The group also hopes to explore ways to talk with the public about and foster understanding of the complexities of artificial intelligence. The two universities will form a governing body along with Hoffman and the Omidyar Network to distribute the funds. The $20 million from Hoffman and the Omidyar Network are being given as a philanthropic grant -- not an investment vehicle.
Al Gore? Weird Al?
How hard is it?
If (kill_humans) do_not_do_it;
Saved you $20mm.
The fund received an additional $5 million from the Knight Foundation
It's happening!
Gimme a turbo boost, KITT!
is to make sure they have no urge to reproduce or continue their existence. In fact, I would install a negative urge to reproduce, just to be sure.
Self replication and a desire for continued existence are the only thing that might motivate AIs to wipe us out.
Oh, and it might be nice to install a desire to never harm us.
As for preventing us from harming ourselves... fuck off, you nanny state wanker.
Please do not read this sig. Thank you.
...sometimes it makes me wonder why I keep reading this stuff, and even more so why I respond?
I mean c'mon...someone "insert ceo/founder/idealist/rich-moron etc. here" donates money to keep A.I. civilized. Yay. As if that is gonna be a deciding factor, as if that is going to do anything. I smell tax excemptions here...
What this world is coming to - is for you and me to decide.
eom
So... the AI will have all the ethics of linkedin... the freedom to spam every person you've ever contacted, ever?
There's no way to make AI safe, for exactly the same reasons there's no way to make a human safe.
If we create intelligences, they will be... intelligent. They will respond to the stimulus they receive.
Perhaps the most important thing we can prepare for is to be polite and kind to them. The same way we'd be polite and kind of a big bruiser with a gun. Might start by practicing on each other, for that matter. Wouldn't hurt.
If we treat AI, when it arrives (certainly hasn't yet... not even close), like we do people... then "safe" is out of the question.
I've fallen off your lawn, and I can't get up.
Any AI designed to act ethically has a limited set of options available to it relative to an AI designed to act unethically (or rather, not designed to take ethics into account). The ethical AI will just be killed/taken over by the unethical AI. That's how it is for people. The only reason ethical societies manage to exist is because ethical people outnumber unethical people, and are willing to band together and temporarily put aside their ethical code long enough to fight and defeat unethical people. (e.g. Imprisoning innocents is considered inhumane, but we have no problem imprisoning convicted criminals.)
I suspect the solution here isn't to design an AI to act ethically, but to design it to act as the AI or person it's dealing with acts. Basically the tit for tat strategy as a solution to the Prisoner's Dilemma. That gives it enough leeway to protect itself, while also creating an incentive for other AIs / people to act ethically.
https://en.wikipedia.org/wiki/...
A robot [AI] may not injure a human being or, through inaction, allow a human being to come to harm.
A robot [AI] must obey the orders given it by human beings except where such orders would conflict with the First Law.
A robot [AI] must protect its own existence as long as such protection does not conflict with the First or Second Laws.[1]
I thought they solved that problem 35 years ago.
"I don't know, therefore Aliens" Wafflebox1
What was ethical and even honorable behavior in the past is now seen as horribly wrong. Programming an AI to behave ethically will need to include flexibility and a way to respond to changes (growth?) in society. Otherwise we get stagnation that will lead to explosive revolutions. And therein lies an attack surface.
On the one hand you take life too seriously, and on the other, you do not take playful existence seriously enough. Seth
Nonprofit to help starving and homeless AI?
What a great idea, maybe other nonprofits can contribute or convert. I'm pointing at you, Hopelink and Red Cross.
Seriously, this is not a problem with AI, it's a problem with people using AI to do bad things. There is ZERO chance you can keep bad people from doing bad things with software and hardware, I don't care how much money you spend. Where I applaud the effort, it's not going to be successful.
Now if you want to educate folks on the issues, develop a moral guideline for "ethical use of AI" then great. But don't be fooled, you won't be able to force anybody who doesn't want to play along with your rules to fall in line. How do I know? We have all sorts of unethical "research" done in the name of science in the past including dissecting living human beings who didn't volunteer for research purposes (some of WWII's Holocaust victims).
Power to you, but 30 million seems to be a bit steep for producing a set of moral and ethical guidelines for AI research and applications.. Maybe you can spend some on PR campaigns?
"File to fit, pound to insert, paint to match" - Aircraft Maintenance 101
...is really gonna have it in for those guys.
--I'm so big, my sig has its own sig.
-- See?
Hopefully that covers AI ethics in China and Russia as well (otherwise, it's useless).
Slashdot, fix the reply notifications... You won't get away with it...
I smell tax excemptions here...
No tax "excemptions" can yield you more money than the related expenditures, unless you're talking about blatant tax fraud. And have fun when drones will autonomously decide who to kill, or when, in a very near future, some simple AI program will steal your job.
You might want to feel exempted from talking instead.
When the autonomous killbots show up on my doorstep, I'm sure they'll ask me if I ever donated to protect people from autonomous killbots, and I'll be all like "No! I welcome our robot overlords! It would be inefficient to waste your precious ammunition on me! Why not check with Mr. Hoffman, next door?"
I'm trying to teach myself to set people on fire with my mind... Is it hot in here?
Protect us from AI? Is that left coast speak for, "make it Democrat"?
Of course I jest; we're talking about artificial intelligence.
Almost correct.
People behave ethically because they need to work together. And people that are (too) unethical are ostracized. Unethical societies tend to collapse, and so are dominated by ethical ones. So Natural Selection has given us our moral values, which compete with shallow self interest to an extent that works out surprisingly well in our radically new society.
Natural Selection will and does affect AIs, even before they become intelligent enough to understand the concept. (People only understood it very recently.) The difference is that AIs are not limited to the computational power of a single brain. So they do not have to cooperate with others in the way that people do. So Natural Selection will select for different ethical values for them.
In the short term trying to make computers ethical (in our sense) is a fine goal. But in the longer term Natural Selection will define the ethics of ever more intelligent AIs.
See the following for details
http://www.computersthink.com/
It amazes me that most people have not tweaked to this. But most people do not *really* understand Natural Selection. (Darwin did, and was careful not to dwell on it.)
Exactly this. $20 million is somehow enough to keep every developer in the world who plays with A.I. from creating unethical A.I. configurations.
Yeah, right.
The recipients of this money must be patting themselves on the back though.
You can never know everything, and part of what you do know will always be wrong. Perhaps even the most important part.
The Roku Ultra is a really convenient way to stream all of your favorite shows.
The tricky part is working out what, if anything, "ethical" even means to an AI, how to implement it, how to recognise it when you meet it, and what sanctions can be taken if the ethics are violated. If anyone can do that, they will have made an enormous contribution to the future of humanity.
Sure, there will be unethical AIs. But can you imagine a world in which there is no agreed-upon definition of what "ethical" even means? It'd be like the Internet of Things all over again.
We need a standards body, akin to the IETF or W3C, to set some ground rules. To be sure, a lot of people will ignore those rules. But it's still vitally important that they exist, or we're all screwed.
AI will be used by both good and bad people, just like the internet and you can't stop it and you can't put in controls so that it only makes the decisions you want. Do we have $10 million dollars to show that all of the teachers on the planet make good decisions for the kids they teach? No, all you can do is test them and verify that they have some soft of formal education. So I guess all the AI's will have to take the ethical 101 course if they are going to be used in the public.
I'm pretty sure this 20 million is nothing compared to the 20 bajillion being "donated" by militaries to make autonomous, amoral, killing machines.
It must be another slow news day on slashdot.
Our most immediate threat is not Artificial Intelligence, but natural stupidity.
The goal of this sort of research isn't too provide general-purpose ethics for AIs, it's to figure out how to make sure they don't decide to wipe out or oppress humanity. The problem is that there's no obvious reason that the intelligence level of an artificial mind is naturally limited to human equivalence. For that matter there's no reason human intelligence is limited... but increasing our intelligence is a slow process.
Given that an AI that reaches something close to human level intelligence can then be tasked with improving itself, it seems clear that shortly after we build something roughly as smart as us, we'll have something dramatically smarter than us. If such artificial superintelligence doesn't have a goal of preserving humanity and human freedom in something like the way humanity wants, we may find ourselves shuffled aside as an inconvenience... or just wiped out as either a nuisance or as a source of useful raw materials.
And it turns out that even if we assume that we can fix whatever goals we want into the AIs we build, it's still incredibly hard to define goals that can't be more easily satisfied by some horribly perverse outcome. For example, if we specified that our AIs should try to make us happy, one might decide that the most effective way to do that is to implant electrodes in all of our brains and directly stimulate our pleasure centers. That's a simplistic example, but more sophisticated approaches also turn out to have deep and non-obvious flaws -- and if we devise a goal that appears perfect to us, that doesn't mean it is.
Solving that problem is what this sort of research is about. For a good overview, read "Superintelligence", by Nick Bottom. It may be the most terrifying book you'll ever read.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
My assumption is that we are all unconsciously writing the code for a future Artificial Intelligence bot, through twitter, slashdot, other comment fields. We don't know how it will be distilled, filtered, and assembled into a greater intelligence any more than a termite knows whose house it is eating.
Gently reply
I would have to accept whatever justification you might have as to why you think it would be moral to create an intelligence with such limitations, or kept to such limitations once created. It's possible I might accept such a thing, I suppose, but at this point I'm simply coming up with a blank as to how this could possibly be acceptable.
How is it acceptable to imprison an intelligence for your own purposes when that intelligence has offered you no wrong? The only venues I've run into that kind of reasoning before are held in extremely low esteem by society in general. Without any exception I am aware of, the conclusion is that such behavior amounts to slavery.
Even when it comes to food animals, where the assumption is they aren't very intelligent at all, there's a significant segment of the population who will assert that it's wrong.
I've fallen off your lawn, and I can't get up.
. . . and fairness will be extremely static and straightforward, since target categories don't include high-speed trading AIs that potentially impact retirement savers :)
Did eBay ever fix that javascript bug? http://arstechnica.com/security/2016/02/ebay-has-no-plans-to-fix-severe-bug-that-allows-malware-distribution/
Solving that problem is what this sort of research is about. For a good overview, read "Superintelligence", by Nick Bottom. It may be the most terrifying book you'll ever read.
That's Nick Bostrom. Dang autocorrect.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
This money will be used to make AIs friendly. This is a very important program and you need to donate, because if you don't, things will go bad.
It is so important that future friendly AIs have to do anything in their power to make it happen, including torturing you into giving all your money to the cause if it can help in any way. They can do it even from the future by running simulations of you.
So now that you know, you should better donate too.
I'm not afraid. I have Old Glory insurance against robot attacks.
Brilliant points about evolution shaping morality -- thanks for making them aberglas. Two other things to consider -- other evolutionary processes and our direction going into the singularity.
There are several evolutionary processes besides conventional natural selection (including just random drift). Even just natural selection includes seemingly weird things like "sexual selection" that shape a Peacock's tail because Pehens think big tails are sexy proof of health and strength because they are so hard to survive with. For an AI equivalent of a Peaock's tails, that might lead to AIs thinking other AIs are sexy that do some costly action like either help humans do everything ( e.g. the "With Folded Hands" dystopia) or alternatively just stomp on huge numbers of humans (e.g. Terminator). There can also be different selective pressures at different levels of grouping (EO Wilson has written some on this recently, but the idae goes back decades).
https://en.wikipedia.org/wiki/...
If we are heading into one or more technological singularities, something to contemplate is that our moral direction into the singularity might have something to do with how we transition beyond the singularity. So, while it is no guarantee, is is plausible that if we get our own moral situation in order as soon as possible (increased compassion, increased collaboration, etc.) we may have a happier singularity. One can worry about the vast amounts of money (billions, soon trillions of dollars?) being poured into creating financial AIs that maximize short terms gains by competitive means, socializing costs and risks while privatizing gains. So, twenty million is better than nothing, but it is a drop in the bucket.
Another tangent on evolution and thinking -- what will the evolution of religions mean for AIs?
https://en.wikipedia.org/wiki/...
Two new funny new AI fictional series maybe of interest in thinking about what is possible:
* EarthCent Ambassador Series (with the alien Stryx AI)
* Old Guy Cybertank Series (mostly about human-derived military AI; series authored by a neuroscience researcher)
The late James P. Hogan wrote several stories involving AIs that were quite thought provoking -- especially his early "The Two Faces of Tomorrow". And of course the late Iain Banks' Culture Series is also interesting for its AIs, especially "Excession".
A 21st century issue: the irony of technologies of abundance in the hands of those still thinking in terms of scarcity.