ipsa

But who artificially created that problem? The artists. There is no marketing company that has intelligent billboards scanning the public for prostitutes. There are no researchers seriously using CV to classify convicts (at least not in the West, and not with ImageNet). That could be a malicious usage problem. This is either a non-problem or Armaggedon for all ML CV datasets, because you certainly can use most CV datasets to train a crappy classifier to output offensive labels. If I train a people photo tagger using a dataset used for combating poaching monkeys in Africa, then who is at fault? Certainly not the researchers who published that dataset with the idea that the data would be used with common sense and scientific rigor, not adversarially -- to make a political point attacking the very existence of that data. The "exposed" bias is trivial.

It is the technical justification that should be all that matters for a canonical academic dataset. Science does its best to be apolitical, but then politics ("red bull drinking white men train racist and sexist classifiers") is forced upon it, and we can't really have a productive conversation about bias and ethics anymore.

AI needs common sense knowledge of the world to improve. Censorship so science does not offend our sensibilities, would only make it so Google Image Search (a machine learning algorithm) does not return any images of people when you search for "prostitute". Heck, the AI would never learn the difference between a male and a female prostitute. Destruction of accessible knowledge so we (aka: people on Twitter who think AI is the terminator, or the director of the internet) don't get offended by some primitive ML model-as-art-project forced to make errors or awkward classifications. That's a sentence the academic ML community could do entirely without. No benchmarks or duckface selfie would be hurt. No unfortunate third-world souls hired to scan 20 million + internet crawled images for wrongthink, only for the machine to do the unsupervised learning in a hug box, not a black box. Oi mate, you got a loicense fer that label?

Just wait until the activists find out who wrote the first 100 8's added to MNIST. Nobody but MIT would be associated with her, if they found out what she did.

There are ~32.000 tags. No surveillance system is using ImageNet tags to classify people into Buddhists or Not-Buddhist. Most researchers ignore these tags and focus on a 1000 classes, and know that 32k performance is not good (and these artists have no intention of making it work at all). What they are uniquely trying with this Art Project is as much research as it is activism. Note that "mantrap" is defined in synset as "A trap for catching trespassers", and that you are bound to find weird stuff among over 30k categories (imagine what you can say with the 32% most popular words in French...).

This is a photo in question: https://memepedia.ru/wp-content/uploads/2019/09/imagenet-1.p...

This is the route the network took:

person, individual, someone, somebody, mortal, soul (6978) > female, female person (150) > woman, adult female (129) > smasher, stunner, knockout, beauty, ravisher, sweetheart, peach, lulu, looker, mantrap, dish (0)

So it was (politically) correct on the first three categories, and the last one was either a crapshoot (and she could also have gotten to the subcategory of "prostitute" > "streetwalker, street girl, hooker, hustler, floozy, floozie, slattern") or she really is posing in a common "beautiful woman"-way. (The global description for this route is "A very attractive or seductive looking woman" and often triggers for females with tilted heads and lip curls).

You can turn any faces dataset into a labeled face color dataset, so if a black person being subclassified as "negro" is problematic bias or encoded racism, then all such datasets are suspect. Noisy labeled data is the norm, not some horrible exception to be avoided at all costs.

There are so many things just so plain wrong about this (I attempted to respond, then had to stop), that I feel this post is more of an attempt to instill the frustration felt when the author attempted to compete and promptly got run over by some SotA- hungry boost-junkies from countries where the p-test is not part of the curriculum in schools. I really don't know how to constructively salvage this... Talk about the role of luck in games?

> Calling a search engine AI seems a little strange.

It's standard in industry. Search is an AI problem.

https://www.sciencedirect.com/science/article/pii/S000437029...

> The international norm (Microsoft, Apple, every other big company) is to obey China's command.

No that's what companies without AI guidelines do. Violating international norms is when the US, Germany, and Japan would complain when their governments would surveil as much and as invasive as China is doing.

China's surveillance apparatus is NOT the international norm!

> From the article it sounds like that happened in 2017, before the AI principles existed.

Yes. So Sundar Pichai introduced those guidelines, knewing full well that they were dead in the water.

> It wouldn't make it impossible, it's already impossible.

One of the scariest conclusions. Really pause and take it in. How significant is this?

> What grave harm would be caused that doesn't already exist?

This is a weird reasoning for me. It reads to me as similar to: People are going to die anyway, what grave harm would be caused by murdering them by your own hands?

I'd like it if you were a bit more specific :).

This will remain a matter of interpretation and, while my reasoning may be sound, your interpretation may differ (much like those employees that pose that designing and developing a censored and spying China search engine app is consistent with "organize the world's information...").

First off, some premises:

- Information Retrieval, Ranking, Spam filtering, etc. are part of AI. Dragonfly applies to these principles.

- Publishing the AI at Google Principles and packaging it the way they did, allows me as an outsider to hold Google accountable to these principles, question their leadership, and critique them if they apparently skirt these principles.

- China's government spying on political dissidents violates international norms on surveillance.

- Google shut out privacy and security teams from evaluating project Dragonfly.

- Shutting out security teams makes it harder to build projects designed and tested for user security.

- Shutting out privacy teams makes it harder to build projects designed and tested for user privacy. Censored search terms are not transparent. You don't control which data of yours get shared with the government.

- Sundar Pichai lied to employees when he said the project was just an innocent proof-of-concept. There was no room for many voices in that conversation, because people lacked moral authority to form an opinion on the matter (they were kept in the dark).

- A fully operational Dragonfly project would make it impossible for Chinese users to use Google to find information about this very controversy (AKA: Google and its behavior itself becomes part of censorship)

- Human Right Organizations were correct in denouncing Dragonfly for its potential to do damage to Human Rights.

- Getting in trouble with the government over search terms that may denote a political preference opposed to the government causes an unjust impact.

- Censored search terms (without showing a notice: "Some results may have been censored it accordance with Chinese law") remove control from humans without any recourse or opportunity for feedback (or choosing another company). By facilitating a censored search engine Google can't point at a government and say: It was entirely their fault.

- A Chinese Google Search Engine which leaks user data to the government is easily adaptable to harmful usage (with little power for Google to push back/notice/monitor once deployed).

- A Chinese Google Search Engine will have significant impact.

- Google is deeply involved, making a custom solution, which enlarges their duties and responsibilities.

- The (user) benefits do not substantially outweight the potential for grave harm

- A censored and spying government-controlled search engine can be viewed as an information warfare weapon.

With these premises in mind, I see them violating all the principles, but one.

- A mea culpa. Drop Dragonfly, admit that the way it was managed goes against AI guidelines, have a third party, like ACM evaluate missteps, and show your commitment to thought leadership on responsible use for AI tech in the future. Or stop the hypocritical canvassing (outwards promotion of "do no evil"), scrap or de-emphasize those guidelines (which currently unfairly attract idealist AI researchers).

- Fire Beaumont.

- Make secrecy and shutting out privacy and security teams against process and punish violators. Inform key decision makers, such as Larry Page, of controversial projects, and punish violators. Make lying to/obscuring your employees at an all hands meeting a fireable offence.

- Align incentives and OKRs. Promote and reward core values and those that make it sticky. Protect whistle blowers and conscientious objectors. Periodically review (and have subordinates review) managers and people in key positions, not for the profit or projects they launched, but for creating an inclusive collaborative working environment. Put less focus in hiring for skill and more focus in core value alignment.

- Appoint an employee ombudsman and objective ethics audit team. Give them enough authority, visibility, and power to make changes for the better. Make sure concerns of lower level engineers make it to the top. Make management justify putting profit over user safety.

- Offer a few golden handshakes to people high up in management, that are directly or indirectly responsible for the public erosion of Google's core values. Be wise to the fact that the best and most productive/profitable leaders are also prone to shrewd and unethical behavior, and guard against this.

My personal conclusions:

- Google can't be trusted on anything to do with building responsible AI (they violated ACM Code of Ethics and their own AI at Google Principles).

- Google has no authority to talk about ethical use of technology and human resources. The main manager responsible for this kerfuffle brands himself as promoting diversity and responsible use of technology.

- Google can't lay claim to being a transparent company, both to its users and outsiders, and to its employees and insiders. Even Larry Page was blissfully unaware of this controversial project that directly goes against his motivations for leaving China in the first place.

- When you go work for Google, you'll have colleagues and managers that won't speak up if they get to work on another unethical project. That will eschew core values for making their stock options grow. That want to build their own empire and positive performance reviews at all cost (even if this costs Google dearly in PR and culture damage).

- Google can't be trusted to be self-regulating, putting the user first, and to clean up any damage done from a top-level ethics violation. There is no objective ethics commission or employee Ombudsman to keep the bulls in check.

- There are more than a few rotten apples in the upper echelons of Google. Perhaps such $$-eyes behavior is rewarded by growing the ranks and internal opposition is seen as a necessary evil to be managed.

Would you like it if technology you created was used in disagreement with your personal moral compass? Or are you completely agnostic about this?

Google is free to create a censored and wiretapped search engine app for China. But it should give transparency to everyone contributing to Google (especially those indirectly contributing insights or technology), so they can make an informed decision to work there.

> allowing to give the Chinese access to modern technology

This is not a problem or an issue. Nobody would have a problem with a Chinese Google Search engine. It is about facilitating spying, potentially contributing damage to free speech and human rights, about organizing the world's information vs. a government controlled propaganda machine.

What then is the issue, is that Google is very powerful at search and modern technology. And these systems are thus extra damaging in potential. Slapping Google brand on it carries responsabilities.

I do agree this issue has reached the stage of moral outrage. I think this is largely due to the leaks and surprise revelations. Senior Google AI now silent on this issue, yet vocal against Facebook or changing name of conferences.

I would hope the outrage would be the same or louder when this happened in the US, not China, so I don't think it is very much a thing of priviledged West vs. poor East.

I think he was annoyed/bored. He is basically playing against a computer for an increasing number of moves. By the time the theory ends, the game is near-decided. Chess960 reveals the natural talents over the memorizers.

The chess lovers complain and complain, because everybody's style is starting to mimick closer to the objective best moves.

It is also a psychological move in trolling Caruana (part of the game): "You did not even try to win with white in your last game before the rapids. I'll draw in a superior position because I love my chances there."

You can solve this "if against Z, then what about X"?

Think of the user. Does a user benefit from being able to clear a permanent record, long after punishment has been served?

Does a user benefit if Google facilitates government spying?

You can be for the right to be forgotten (and admit to its abuses by politicians and rich hucksters), and against government censorship.

This distinction is powerful and therefor often abused: "It is for the good of the people that people can not dissent and revolt". But hard to argue that's the case for EU user protection laws.

China will create designer babies with extremely high IQ (150%+). There will be no way for the rest of the world to compete using good old fashioned nature, so they either become irrelevant, serfs, or also start using in-vitro CRISPR editing. This will lead to AGI in a single generation (so about 30-40 years).

It's the webdev equivalent of creating a log-in form tutorial and putting the password in a JavaScript variable.

The article splits in time, not randomly.

You can try out this technique at https://github.com/google/unrestricted-adversarial-examples My guess is it would have the same result as adding noise to the normal images too (resulting in a slightly worse performance overall).

Some of them yeah. There is active research on this. But it is also possible to create adversarial images for soft voting ensembles of the 6 most popular architectures. Those strong adversarial images that beat the consensus, also have a large chance to fool new neural network architectures that the adversarial image creator never had access to.

Read it as: A small amount of carefully constructed noise. Then you are correct to literature and pop-science. No misconception needed. There are 1-pixel attacks now. Randomly shuffling a small amount of pixels around can cause predictions to shift.

The issue is that there is no scene understanding. No common sense. No 3D modeling. Just 10x10 pattern matching on a very large fuzzy database of natural images (which works really really well in most cases).

The hype of ML is driven by 3 things: Big companies vying for AI dominance, militaries that want to finally use neural nets that work, and international competition between the West and the East to be the first to largely automate their economies (or AGI if you want to call it that). Catalysts were big data hoarding, GPU training on ImageNet, and then AlphaGo.

Replace "AI" with "software" (which for all intents and purposes it is).

> Software is the centre of every business today and the market has been waiting for a solution that eliminates technical barriers to build software so that everyone can engage in the new economy,” said Manu Gupta, Partner at Lakestar. “By creating a software powered assembly line combined with the best global human talent, Engineer.ai’s Builder bridges the gap between an idea and a software product to enable it.”

You could automatically encode a KNN model as a set of logical if-then rules: "if x1 > 10 and x2 < 3 then 4 nearest labels are [1, 1, 1, 0]" so the information is there. For KNN you could also train weights for every variable (how much should they count in the distance calculation?). For deep learning you have way more parameters and architecture choices than for nearest neighbors (mostly the distance metric and the number of neighbors to consider). After that, both learn a mapping from input data to a target.