Leave A.I. Alone | Hacker News Reader

Leave A.I. Alone

nytimes.com

32 pointsgk116 comments

Just a random op-ed (read: Rando with an opinion that happens to be under the auspices of NYT) that provides nothing new to the discussion.

Poo-poos concerns of AIs influence on the economy and human safety, with laughable comparisons.

For example it argues that people don't agree on what AI is therefore somehow regulating it is premature... (Why would that matter?).

The clarity of thought is this piece is negligible.

So let's clarify:

- Some people are [rightly] concerned about technology in general. As it becomes more powerful (knives then guns then nukes) the ability for a few people to ruin the world for everyone increases.

- Some people are [rightly] worried about technology's role in concentration of wealth.

- Some people are [rightly] worried about existential threats "grey goo"/black-mirror/robocop/Ex-machina.

The role in AI is that it makes each of these threats much more dangerous (e.g. putting AI in drones, AI's effect on the uneducated people's employability, Robocop)

[Author doesn't attempt to provide evidence against any of these claims, but just argues the concept is used in a vague way therefore somehow doesn't count]

The takeaway I got was that the author more or less wants the conversation about regulation of AI to be less of a top-down approach to a broad reaching and loosely defined topic. The scenario of a, "grey goo" apocalypse verses, say, a bank hiding otherwise discriminatory lending practices behind a neural network black-box, are radically different in terms of plausibility and likelihood. Given the esoteric nature of the topic it's all too easy to see where this is headed when it comes to representative governance. Knee-jerk legislative umbrellas based on the comments of eccentric megalomaniacs made to the news media is a fool's errand, and there's a hell of a lot of foolish people out there.

Well if his position is there are various real AI threats that need to be discussed and handled differently, then I'd agree with that position. However that's not how I read it, particularly given the title.

Nevertheless, someone on the full-time staff decided to commission this piece, then proof-read it and decided to publish it. Op-ed is no disclaimer really, if it’s in the paper it is by definition endorsed by the paper.

The article is a poorly argued response to a strawman.

The NYC and federal advisory councils would be tasked with providing lawmakers with impartial advice on A.I. writ large. That's reasonable. Bias in algorithmic credit decisions and bias in algorithmic sentencing are different problems, but share a lot of the same underlying technological realities/problems. These areas do have cross-cutting concerns, so a legislative brain trust dedicated to disseminating knowledge about these cross-cutting concerns is a wise use of resources.

No one is suggesting "regulating" general AI, so the author's red herring question about "what is AI anyways?" is completely off-topic. Even "alarmists" like Musk and Hawking are proposing much more concrete legislation, such as banning algorithmically generated kill orders in weapons. And to the extent that those people are interested in safe general AI, they're currently trying to achieve that goal by throwing research money at it (rather than proposing concrete legislation regulating general AI). Which is hardly a reasonable thing to criticize!

Furthermore, the proposed approaches in NYC and at the federal level aren't incongruent with the author's proposed sector-by-sector approach toward regulation! These are, after all, simply proposing advisory councils.

Therefore, the author does not provide a compelling reason to oppose any of the concrete legislation mentioned at the beginning of the article. And the arguments that the author does make betray, IMO, a superficial understanding of the proposed legislation.

I'm working my way through the "Blueprints for Armageddon" series in hardcore history. The attempts to regulate AI remind me of the Russian efforts to regulate war before WW1.

In the first part of BfA (I think) he talks about the peace conferences in Russia whose goals were to limit the progression of technology. Partly because Russia wasn't prepared for a high tech war, and partly because large scale efficient killing was kinda scary. Anyways, they wanted to limit what technologies could be improved by freezing combat technologies where they were at the time.

Dan talks about how it wan't really that far-fetched of an idea to to artificially limit the spread and extension of knowledge at the time. This is in contrast to today, where it seems impossible for technological advancement to stop. Tech changes so fast it's difficult to imagine that it'll be slowing down soon. Could we even slow down the advancement of AI if we wanted to? How?

Is there any dangerous technology whose development could be stopped even if we saw it coming?

IMHO it should be illegal to develop an artificial mind capable of consciousness. There are a lot of good reasons to avoid this, not just that it could be potentially unethical to subject such a mind to torture/pain.

A super-intelligent artificial mind is probably going to come up with its own agenda and convince us to work in its interest and potentially against our own. Given the state of computer security today, can you imagine how easily it could spread as a worm?

The AI industry regulates itself by periodically going bankrupt

About the author: "Andrew Burt is chief privacy officer and legal engineer at Immuta, a data management platform for data science, which enables companies to create and manage A.I. and machine learning models."

Hmm.

>in the summer of 1970, Newsweek ran a cover story titled “Is Privacy Dead?” detailing the “massive flanking attack” of computers on modern society. Growing awareness of that threat led to broad appeals that echo modern proposals to regulate A.I. “Eventually we have to set up an agency to regulate the computers,” Senator Sam Ervin of North Carolina said in 1970.

>But instead of regulating all computers, the government sought a targeted approach tailored to specific problems, passing regulations like the Equal Credit Opportunity Act in 1974.

Thereby solving the problems of digital intrusions on privacy, and credit scores forever. Oh wait, no, I guess those both those things are incredibly broken. Digital privacy is notoriously non-existent, infringed on by the state, criminals, and business. Credit and financial information is impossible to keep safe and available to any cyber criminal who wants it.

It's odd that the author's two examples of when the government got something right by not preemptively intervening both turned out so poorly.

AI is fundamentally different from privacy, credit, or any other invention that has happened before. The implications of this technology are too vast for us not to spend time and money thinking about how to proceed safely. We may only get one shot at developing true AI correctly, and establishing a public office to help figure out how to do that seems like the least we could do.

If anyone finds themselves agreeing with this author, I highly recommend Nick Bostrum's book Super Intelligence. If you don't have the time for a whole book Sam Harris has a good TED talk on the subject.

> Credit and financial information is impossible to keep safe and available to any cyber criminal who wants it.

Which is why legislation should be aimed at preventing law-abiding entities from ever storing it in the first place. Obviously, criminals cannot steal what is not there.

Enforcement of these laws has been mild, and all companies and criminals know this. Only the odd major embarrassment seems to motivate companies to protect user data.

'AI' is such a meaningless buzzword that you could simply rename a product to sidestep any pointless regulation.

Regulations generally start out by defining the thing they're regulating. You can call it whatever you want, but if it fits the law's definition, it's going to be covered.

Sure, but pretty much all software qualifies as 'AI'. However most people think about Hollywood interpretations of AI, which is a thing that doesn't exist. It's strictly a matter of interpretation, and figuring out where on this spectrum a particular piece of software exists is basically the equivalent of throwing darts at a board.

In my private taxonomy, hello world through Excel are weak AI, the state of the art mostly narrow AI but fast becoming more general, and the endgame is general AI.

What worries me is the 80 year old judges (and other clueless participants of the judicial system) who don't understand how computers work and will ultimately be responsible for deciding people's fate regarding this stuff.

Reply on news.ycombinator.com