Americans Want To Regulate AI But Don't Trust Anyone To Do It (technologyreview.com)

An anonymous reader shares a report: In 2018, several high-profile controversies involving AI served as a wake-up call for technologists, policymakers, and the public. The technology may have brought us welcome advances in many fields, but it can also fail catastrophically when built shoddily or applied carelessly. It's hardly a surprise, then, that Americans have mixed support for the continued development of AI and overwhelmingly agree that it should be regulated, according to a new study from the Center for the Governance of AI and Oxford University's Future of Humanity Institute. These are important lessons for policymakers and technologists to consider in the discussion on how best to advance and regulate AI, says Allan Dafoe, director of the center and coauthor of the report. "There isn't currently a consensus in favor of developing advanced AI, or that it's going to be good for humanity," he says. "That kind of perception could lead to the development of AI being perceived as illegitimate or cause political backlashes against the development of AI."

Comments

Simple Solution

[AdamStarks]

Create an AI to regulator!

Re:Simple Solution

[olsmeister]

We need an AI as an arbitrator. By using this AI, you waive any right to sue individually or through a class action lawsuit. You agree that any dispute will be resolved through the AI arbitrator and such resolution will be binding legally in any applicable jurisdiction. You forfeit any right to sue the AI for any reason and agree to be irrevocably bound to whatever the arbitrator AI decides.

Re:Simple Solution

[Holi]

Crypto? Well I guess in a couple of months I'll be able to afford, well all of it.

Re:Simple Solution

[VeryFluffyBunny]

AI doesn't need a specific regulator, they just need to be held legally responsible when they test their products on the public without first reasonably assessing how safe they are & determining what precautions & insurance are necessary if something does go wrong. Being run over by a car driven by a computer or a squirrel should be no different in the eyes of the law. If it was reckless, then the person who ultimately makes the decision to put the AI or the squirrel in charge gets prosecuted & sued. Also, remember that even the most advanced AI is a lot less intelligent than a squirrel.

I know the perfect person!

[zippo01]

My vote goes to John Connor!

Re:I know the perfect person!

[sconeu]

We could have Isaac Asimov (or Dr. Susan Calvin) do it!

Warren G and Nate Dogg

[nwaack]

REGULATORS!!!!

Make Apple Pay for it.

[Oswald McWeany]

I propose that AI be thrown in a walled garden, and let's make Apple pay for it.

Most Americans don't know what AI is

[OffTheLip]

Count me as one of them and I was a CS major. Seems like everything is lumped in with AI now.

Re:Most Americans don't know what AI is

[stevegee58]

Haven't you heard? Elon Musk says we're "summoning the demon"!

Re:Most Americans don't know what AI is

[Shaitan]

There isn't a good definition. What was once AI is now being called strong AI or general AI. What we have is more like a process similar to panning for gold, we've created a logical process whereby we toss chaos into a slurry and agitate it looking for the gold to sift to the bottom. Once that is working we take the concept and scale it up to a sluice. It's no more intelligent than that, but we can keep dumping dirt in and it will take care of separating out the gold for us with some margin of error.

Wait ...what?

[theCat]

Does this mean Americans are getting wise to the con of continuous and benevolent progress?

Pics or it never happened.

Re:Wait ...what?

[The Grim Reefer]

Does this mean Americans are getting wise to the con of continuous and benevolent progress?

That or have watched the Terminator movies too many times.

I just wrote a hello world neural network

[stevegee58]

I just wrote a hello world neural network and didn't register it with the government! Did I just commit a felony?

Re:I just wrote a hello world neural network

[BlackOverflow]

Sir, I believe that is a crime against humanity. Please report to the local commisar.

Re:AI nutters

[nw_rad]

AI has been 20 years in the future since the 1960s.

Not possible

[Alain Williams]

AI is software so is not that hard for someone with a bit of money to get hold of and use. Yes: you need some expertise, software can be copied & modified to taste, the hardware is cheap & readily available. Who do they want to stop getting hold of it ? Easily obtainable & affordable by governments, terrorists & Bond villains.

Easiest Solution Ever

[NicknameUnavailable]

Ban all AI research until we understand how to do it without being shitbags (e.g. after we've reached post-scarcity and won't be "trusting" megacorps not to try to enslave it,) if other countries research it nuke them, without mercy, just wipe out every single citizen they have. There's no way to morally develop AI in the modern world, it simply shouldn't be done. When we're able (psychologically, culturally, economically, AND practically) to do it and treat it no differently from a person we'll be ready, until then slaughter anyone who attempts it.

Re:Easiest Solution Ever

[Paxtez]

This has to be a troll. You know any sort of "real" AI is VERY far off. We are not even a little close. Not in
All our 'AI' falls into one of two categories: 1) If / Then / Else type statements, 2) Rapid trial and error to solve this very specific narrow problem.

Google's Deepmind Go program is no closer to taking over the world than an average 3 year old.

Throwing more CPU cycles or RAM will not make it suddenly "smart".

I would be very surprised if we get anything remotely in the same ballpark as "smart" in our lifetimes.

But machine learning can have awesome positive effects on society.

Re:Easiest Solution Ever

[drinkypoo]

We don't know how close we are to developing intelligence. Neurons are more complicated than transistors, but transistors are a lot faster. And relatively simple animals (in terms of number of neurons) exhibit some pretty complex behavior. A frog's got roughly three orders of magnitude less neurons than we have, but it still manages to have fairly complex behavior, like territoriality. We could conceivably also glue logic onto an artificial intelligence that would effectively make it more intelligent by providing functionality which normally requires a bigger brain.

I'm not proposing any timelines, but do we really have enough data to pooh-pooh the idea that it could happen soon enough that it's worth talking about the ramifications now? Nerds have been doing it for ages already.

Re:Easiest Solution Ever

[NicknameUnavailable]

You speak of morally developing AI while championing a zero-tolerance kill 'em all nuke policy? You make Trump look like 10th grade soy boy pussy with that fucking attitude.

Thanks, I think. But in all seriousness, the threat of making a race of AI that want us dead or even a single AI controlled to enslave Humanity is so extreme that killing 99% of the Human population to stop all AI development before we are ready for it would be the less-evil alternative if it were necessary. You're talking about a comparison of "everyone" or "everyone but the 1 dictator who controls it" vs "some number less than everyone minus 1." The numbers don't lie, morality here is entirely quantifiable.

Ready? Obviously you haven't noticed that we humans act like shitbags towards each other on a daily basis. I'm sure we'll treat AI the same. Or worse.

And thus why the sensible solution is to not make it until we're ready to treat it as a person and not have it working to slaughter everyone as a result.

Re:Easiest Solution Ever

[NicknameUnavailable]

This has to be a troll. You know any sort of "real" AI is VERY far off.

That's not a justification to allow the most corrupt organizations on the planet inch toward it.

Re:Easiest Solution Ever

[NicknameUnavailable]

You're kidding me, AI is being heavily developed by all countries for "defense" e.g. war.

And then? We can nuke them all, it would be less destructive than even a single super-Human AI bent on wiping out Humanity.

How do you even "regulate" AI?

[the_skywise]

There's not even standardized definitions of AI internals yet so I'm not sure how you write regulations that could easily be understood by those implementing the AI to begin with.
Sure there's the old "don't be evil" (we see how that turned out) and non-military AI shouldn't kill but beyond that? So long as they're still subject to the laws and regulations of what a normal human would be doing that the AI replaces I'm not sure there's much else that can be done currently
AIs should not cold call people after 8pm in their timezone.
AIs should always say please and thank you
I'm flashing back to that scene in Robocop 2 after the civics board got done adding 100 "prime" directives to his software.

Re:How do you even "regulate" AI?

[alvinrod]

You really don't regulate it. You just get a bunch of useless chuckle fucks together that don't provide any real value and invariably end up being corrupted by monied interests. So you ultimately end up with the least morally scrupulous companies doing what they want anyway, but now the tax payer has to foot the bill for some additional useless paper pushers.

Re:How do you even "regulate" AI?

[Rockoon]

You missed the part where the companies then use those regulators to create barriers to entry into their industry.

Re:How do you even "regulate" AI?

[the_skywise]

Step 4. Profit!

Re:How do you even "regulate" AI?

[mark-t]

The entirely standard definition of AI is "Artificial Intelligence". This definition doesn't necessarily mean as much as it should to many people only because the definition of what constitutes "intelligence" may be fuzzy.

But simply put, if something has intelligence, and that intelligence is not naturally caused, then it is AI, by definition.

Re:How do you even "regulate" AI?

[Holi]

Fuck me, that joke finally worked.

Re:How do you even "regulate" AI?

[ljw1004]

There's not even standardized definitions of AI internals yet so I'm not sure how you write regulations that could easily be understood by those implementing the AI to begin with.

The concern is with ML which I think we can come up with a decent working definition. Here's my first attempt after 30 seconds "a decision procedure determined by a weighting system where the weights have been derived through a training data set and where a human is unable to predict/justify/explain the particular values that were assigned to the majority of those weights".

The concern is that for whatever reason - inexpressiveness of the neural network, or inadequate selection of the training set - the decision procedure ends up with decisions that are reasonable to it, but which a human when faced with this particular decision would evaluate that it violated some basic human right or similar.

I'd start by regulating it in specific areas, e.g. "if a person has been denied a loan, they are entitled to a written explanation for the grounds/criteria that went into that decision". The answer "our neural net said so" isn't an explanation and doesn't provide grounds/criteria. The answer "your credit score is below 500" would be fine. The answer "our neural net said so, and a human reviewed it, and based on the following reasons supports the neural net's answer" would also be fine.

Re:How do you even "regulate" AI?

[presidenteloco]

A good chunk of all computer programs ever written do "algorithmic decision making".
That's a lot to regulate.
We better narrow it down.

AI by any other name...

[eford49]

It doesn't help that the media (and politicians) in their ignorance and quest for hyperbole have begun referring to anything done via software as AI.

Re:I just wrote a hello world pubic network

[PolygamousRanchKid]

I just wrote a hello world pubic network and didn't register it with the government! Did I just commit a felony?

Now that might get you into trouble . . . especially when you display it at your local mall.

It's ok

[meglon]

General Brewster is all over this.

Ever notice that those who want to

[oldgraybeard]

regulate something. Always have studies saying "Everyone wants and agrees that X should be regulated" IE Controlled by government!
So the "Center for the Governance of AI" says Americans want AI regulated.
What a crock, most Americans are trying to figure out how to play some game on their phone or get their Alexa, etc to do something that will be fun for about 5 minutes. And they are often not able to figure it out.

Just my 2 cents ;).

Americans

[The Evil Atheist]

Just get over your irrational fear of regulations. If you don't trust the government you voted for to come up with reasonable regulation, maybe it's time you started looking at yourselves and who you vote for and why you vote (or, don't, in both cases). It's your own fucking fault if you vote based on a candidate's likeability rather than realability.

Re:AI nutters

[suutar]

I don't know if people think "real AI" (insert your definition here) really exists, but there's certainly some "the computer makes the decision, humans just get to see the results" in play, and in some cases the humans don't know enough about the current state of the innards of the black box to be able to predict the decision. I'm not worried about robot revolution, but I am somewhat concerned that folks will get turned down for loans, insurance, what have you for reasons that no human can explain, validate, and/or correct.

Confusing

[Visarga]

The term AI is confusing. They should have asked individually about: face recognition, self driving cars, automated shops, machine translation, voice assistants, ML based genetic research, content recommendation/filtering systems and so on. Then people would have given more precise answers. But if you frame it with the term AI, people think Terminator and Skynet.

Re:Confusing

[iggymanz]

That's okay, none of those things you mentioned actually use any form of intelligence, artificial or otherwise. AI is a just buzzword thrown onto algorythms and techniques invented decades that largely haven't delivered on the promise of AI.

What about Emilio Esteves?

[dkman]

Youtube

Re:What about Emilio Esteves?

[nwaack]

+5 informative. I never knew where that sample came from, now I do. Thank you, sir.

Time machines that rest on the ground

[tepples]

But we need to invent the time machine first, otherwise Kyle Reese will never make it back in time to be the father of John Conner!

Is this heading toward a Roseanne/Terminator crossover?

Some SF works have solved the space part of spacetime travel by requiring the vehicle to rest on the ground, thereby remaining coupled to the destination through gravity and friction. These include at least H. G. Wells's The Time Machine and Shane Carruth's Primer.