> This is why, after decades of controversy, Google Translates still renders the gender-neutral “they are doctors” in German as “sie sind Ärtze” (masculine) and “they are nurses” as “sie sind Krankenschwestern” (feminine). Google Translate was not programmed to be sexist. The corpus of texts it received happened to contain more instances of male doctors and female nurses. [...] BriefCam’s spokesperson added that they used “training datasets consisting of multi-gender, multi-age and multi-race samples without minority bias,” but declined to provide any evidence or details.
Okay, will we at some point admit that we want machine learning algorithms to be able to interface with symbolic rules rather than pretending that everything is a data issue as if we're living in the era of 1930s inductionism? It's clear that the misgendering issue here is not a stochastic one that ought to be solved by 'balancing out' data, it's that we want to impose strict linguistic rules and constraints on a system in a clear manner.
There really needs to be more work done in AI that makes it possible to interface with the models we built rather than trying to reframe everything as a data problem and then shove it in some end-to-end black box and then hope that whatever comes out at the other end is correct.
The automated systems used in the article are supposed to make judgements about "Detection of body movements that constitute assault." This requires genuine understanding and high-level capacity to reason rather than just pixel-based inference from some camera.
Lithuanian language has it in a very similar way:
Daktaras(male doctor)
Daktarė (female doctor).
A sentence like 'doctor said no' gets translated into 'daktaras pasakė ne'. Even a human could not translate any better unless wider context is known,which could only be derived from other sentences in the paragraph.
> at some point admit that we want machine learning algorithms to be able to interface with symbolic rules rather than pretending that everything is a data issue
That is a long way off. We find ourselves in a situation where we have models that do useful things, but are also horrendously biased.
If you want improvement now, the quickest way to get it is recasting benchmarks. Most NLP research groups just hill climb on GLUE or whatever. What gets measured gets managed, and bias is not being systematically measured.
Right now language models can't even compare ages that fall outside the 20-100 range, and it is really unlikely that we are going to quickly get to architectures with symbolic reasoning sophisticated enough to squelch bias.
To Google's credit, they are working to measure (and therefore manage) bias [2].
Google translates it that way because German nouns have grammatical gender, and doctors are grammatically male, while nurses are grammatically female. It has nothing to do with training ML models (except insofar as most training input is grammatically correct).
Of course, some people are unhappy about it and experimenting with gender stars and gender gaps and other creative orthography, but this is the status quo.
It's quite strange to demand that some ML model is more "progressive" than society and speakers themselves.
The issue here is that grammatical gender is not useful information when reasoning about the semantics of some sentence. Toothpaste in German is grammatically female, that's not a reason for an ML system to make feminine assumptions about toothpaste after it combs through data, it has nothing to do with progressive values, it's that the ML system cannot distinguish between a spurious correlation and actual meaning. Today many more medical graduates are women, this will change the ratio in the future and the inference from grammatical gender will be wrong. We should be able to tell an intelligent system from the get-go to ignore something we know to be spurious rather than fiddling around with the data.
And it's not strange at all to demand of an automated system that it behaves exactly the way we want it to behave. It is not human, it has to be more precise because it is rolled out at scale and it needs to do what we tell it to do. When we use industrial machinery in manufacturing we don't go "ah well humans are only precise down to a centimetre, guess we'll let it slide". An automated car trained on speeding drivers must not learn to speed. Automated systems are faster than humans, so errors compound, which requires more precision on part of a machine. If a ML system accidentally learns that ignoring someone wearing a green shirt is okay, and that is rolled out to a million cars, you don't have an accident but a big disaster.
We need a way to interface with ML systems in ways that let us put precise limits on when it makes inferences from data, why it made those inferences and when to follow logical rules, and when to dispose of certain data.
Der Arzt/die Ärztin, der Pfleger/die Pflegerin (I think "Krankenschwester" isn't being used formally anymore)
Occupations generally have a male and a female variant and as far as I know you are required to address both in formal speech (so I wouldn't call this an "experiment" anymore).
The only area where I'd say this isn't implemented yet is informal speech. Few people watch their day-to-day language for gender neutrality.
> Der Arzt/die Ärztin, der Pfleger/die Pflegerin (I think "Krankenschwester" isn't being used formally anymore)
Now you are being sexist. Of course Krankenschwester is used but it has a different meaning than Krankenbruder
> Occupations generally have a male and a female variant and as far as I know you are required to address both in formal speech (so I wouldn't call this an "experiment" anymore).
In many languages there are occupations which are specifficaly related to specific genres.
> The only area where I'd say this isn't implemented yet is informal speech. Few people watch their day-to-day language for gender neutrality.
This has become something disgusting. A mother is a mother. A father is a father. A mother can give birth, a father not. Some things will always be gender specific. Or shall we start calling people gender neutral: Mr./Mrs. Donald Trump
I'm not sure I totally understand what you're trying to communicate but I didn't say we need to invent gender neutral terms for every German word. I said formal German is moving away from describing people using words that only exist as either masculine or feminine. So depending on where you sample your data it's not unusual to encounter these speech patterns. (Of course, if an ML algorithm can pick that up is a different question.)
And (this is an opinion now) I don't think that's "disgusting". Agreed, it's kinda clunky to always mention both variants but stereotypes are subtle and this is one thing we can do that's not too invasive. Of course this isn't an ideal solution either because gender non-conformity exists and the German language has no good way of addressing a lot of that. (But there are deeper cultural problems there.)
Also, I'm not sure what you mean with "Krankenbruder" but it sounds like something from /r/ich_iel.
I think you're right, but you're being downvoted by people who don't speak German and don't understand how "sexist" the language really is, even if Germans themselves are as enlightened on the subject as one could be.
People forget that English has reinvented itself for the last hundred years and it's not showing any signs of slowing down, whereas German-speaking culture is a little more resistant to the mores of newspeak. Until you lived and loved in both languages it's not really clear how equitable German culture is, in comparison to English. A German might say there is a level of respect in calling something feminine, whereas an English-only speaker might react with abhorrence. Such misunderstandings are to be expected across the language barrier.
I have checked Prague claims and it does not look like surveillance. It looks like plan to build system for traffic control in 2022. Source article does not say anything about surveillance.
Also source article is really incoherent I do not know what author tried to say maybe some vague claim about possible bribery? Those newspapers are owned by czech PM and he uses them to attack political opponents. I would not believe a word from them otherwise there is another source.
Prague is now ruled by Pirate party and they are against using cameras for surveillance. They stated it in November 2019 when police wanted to start using face recognition. I am not sure why they added Prague to the list.
France has more than just two political parties. Macron started as a centrist populist with his own party, but has leaned to the right with his actions since coming to office.
I understand that there are more than two political parties. I'm not that blind of an American. Just curious how most Frenchmen view their own leaders on the spectrum.
I suppose it depends who you ask. He's less conservative than the traditional right, but economically he's very much on the right side of the spectrum (at least relative to French standards). The prime minister and minister of economy he designated come from the traditional right party.
Center right yes. The prime minister is a from a right party. On average since the election, Macron's popularity went down with French people voting left, and raised (or didn't went down as much) with French people voting right.
Mannheim is running a "pilot project" called "Mannheimer Weg". It is video surveillance + behavior analysis. The german state of Baden-Würtenberg, where Mannheim is located, has already announced to expand "intelligent video surveillance" to other large cities, starting with Heidelberg this year and focusing on central train and tram stations.
There are many sources in german about this topic, some praising it, some are more critical. Another such "pilot project" is run at Hamburg Hauptbahnhof and that one got a lot more bad feedback, because they also "try" facial recognition and tracking smart phone signals.
It should be noted that the european union invested into this technology starting with the 7th Framework Programm in 2007 under the codenames INDECT, ADABTS and SAMURAI. Expect to be flagged for "abnormal, possibly criminal behaviour" in the future if you run or loiter at a large train station.
Germany: the city of Freiburg is also buying 18 new cameras for the area "Bermudadreieck" and "Bertoldstraße". I couldn't find any information on how "smart" that system is going to be.
In Darmstadt the central tram hub "Luisenplatz" is going to get modern video surveillance, too. Bought from Dallmeier Electronic GmbH & Co. KG. The mayor said no facial regocnition is planned, even if the vendor offers integration with such systems. Its unknown if advanced behavior analysis is included. It is likely that all cities buy from different vendors and may get very different quality of automated video content analysis.
In Hannover the local police veiled their own cameras last year during a demonstration against a new police law. The demonstration asked for that and announced to go to a court and the police just complied without waiting for a court order, as they wanted to show clearly that they are not interested in filming and (possibly) personally identifying their political opposition. When the city council in Darmstadt was asked for such a rule for the soon to be installed cameras, they laughed and said both the police and the interior intelligence agency can and will request demonstrations to be filmed on a case by case basis and the people will not be told if the cameras are on or off.
Germany's federal police (BKA) and ministry of interior security ("Ministerium für Staatssicherheit") are unhappy with these fragmented and diversified solutions. They aim for a networked system that can find people nation wide on any camera using biometric recognition when a search warrant is issued.
In the meantime Deutsche Bahn is experimenting with a "facial anti-recognition feature" that replaces faces with fakes because they believe they can easier share the data with third party data anaylsis providers if they do that.
Here's what I don't understand: Apart from Heidelberg central station there is only one other place in Heidelberg – Bismarckplatz – that will be subject to intelligent video surveillance and that place already comes with a police station right in the middle of it. So how about increasing convential police presence instead?
What's more, in my experience police in Heidelberg is outstandingly slow to respond. I once reported a mass brawl of ~100 people not far away from Bismarckplatz and police didn't care to arrive until 20min later (with two officers…).
There are at least two issues, and both are worth discussing. Interestingly, if we prevent the surveillance problem we also prevent the racist surveillance problem, so in this way one could argue that (in this specific situation) the surveillance problem is more fundamental.
Maybe you should be entitled to some respect for your privacy in public, though. Indeed in some places you are today. There is no good reason to extrapolate from the natural situation that "If you go out then someone might see you" to the entirely unnatural "If you go out then you should be subject to arbitrary surveillance, recording and analysis by unknown parties who also have access to unknown additional data about you for unknown purposes."
So you're fine with your everyday movements logged and published online? After all, to go anywhere you'll need to step public roads, so you have no expectation of privacy.
>So you're fine with your everyday movements logged and published online?
This comment will be downvoted, but this is hyperbole. Your everyday movements are more than just your activity in public spaces.
Likewise any website publishing your activity in public spaces is not entitled to linking it to your PII, because if they do deanonymize it, you now have means to take it down.
No expectation is different from your information being automatically stored and analyzed (and easily traced back to you whenever the government feels like it).
The article is about using Machine Learning on the video feeds, not that there are video feeds. And the issue at hand is that ML algorithms can have racial bias in the classifications.
Seems like a red herring. They want you to assume that mass violations of privacy are okay as long as they're not racist.
They're also ignoring classism - It's fine if the rights of the working class are violated, as long as they're all violated equally. Never-mind the people who can afford privacy on the free market.
When it comes to the police force surveillance, prejudicial bias is more a problem than a red herring. It's true that mass surveillance is an issue but how it is implemented can be an issue as well.
I'm curious - how does a wealthy person retain their privacy from this type of mass surveillance but the working class does not?
Transportation and shopping come to mind. If you do most of your shopping online you will be less exposed to this kind of mass surveillance; conversely, if you have to use public transportation and walk around, you'll be more exposed to it: not "retain their privacy", but "retain more of their privacy".
It's also possible that more affluent neighborhoods have fewer cameras.
'Accessibility, affordability, advertising, anonymity and anomie, are the five cylinders of the engine of mass addiction'
Um Another way to view... Let us take a quick look at a utopian idea: 'Does the government guarants me, that gains come without additional costs ?' (-;
The same way they do now to consume luxury goods and services and avoid papparazo and working folk. Blacked out SUVs and back entrances. You think Kylie Jenner has set foot on an LA sidewalk?
I think as long as your in public then it's fine to assume you're being recorded by CCTV. Obviously the government being able to access cameras in your own house like 1984 is not ok. But CCTV in public does reduce crime and help investigations. For example the Central line in London doesn't have CCTV (unlike most of the other lines) and it has the most assaults. https://www.newhamrecorder.co.uk/news/central-line-trains-ha...
I think wether you can trust your government to not misuse it (eg tracking minorities) is somewhat separate.
The idea that you should lose any form of privacy the moment you step outside your home is an extremely dangerous one. The reasons we value privacy do not suddenly cease to apply because you live a normal life.
Also, the evidence that CCTV is effective at deterring or solving crime is debatable. There is a plausible alternative theory that it simply moves the crime around. That is OK for you if you live and work in "nice" parts of town, but not so much if you are in the districts where the crime and antisocial behaviour get pushed to.
A +1 observation regarding the effectiveness of CCTV being debatable. I used to work in this field and the number of crystal clear views of hoodies and base-ball caps was remarkable. Cameras tend to be mounted up high to deter casual vandalism. As for number plate recognition systems - the ancient trick of stealing a set of number plates from a random vehicle is alive and well. As long as it doesn't ping anything as the car moves along it is effectively a ghost until the theft of the plate is reported. The only time I have heard of this back-firing is when there is a problem with the stolen number plate (e.g. No insurance, Intel markers etc).
The Central Line is also the busiest (most trips, I can't find most passenger-km), so without knowing the number of assaults on the other lines, this is irrelevant.
I've lived in the U.S for over 20 years in multiple regions, I've lived in Denmark for over 20 years, and been around other parts of Europe reasonably much, and I have some experience about how racism works in Brazil.
While all these regions are prone to racism, the experiential reality of racism between two places is not the same, and this article's seeming assumption that the racism of the U.S is directly translatable to these different EU cities is ridiculous.
Of course if someone uses an algorithm trained on the racism of the U.S in Prague they might get something that anyone viewing the reports would find really weird, so that might be an amusing thing to try.
In short the issue at hand is an issue apparently dreamed up by people not realizing that their society's problems are not exactly universal.
Indeed. This is where the discussion should be. The topic of whether we should have cameras everywhere is one conversation. But this one is about how to handle the unintended side effects of ethnic bias based on data and the analysis.
You make an excellent point about an algorithm being trained in the US being used in Prague. They might not be applicable since Czech Republic since training an algorithm in what is a few steps up from an apartheid state will of course be doped with the racial biases of that country.
I'd like to see a breakdown by cameras by government, companies and homes. I think all those are not equal. Its one thing to have a camera looking at your porch for package thieves but a camera in government hands can be used for a lot more.
Is this actually a useful metric? Cities should be compared to cities.
Also, that link doesn’t seem to say whether cameras are owned by the state. Presumably surveillance operated by the government is more totalitarian than various private businesses.
I guess, but I still feel that including NYC and rural Wyoming under the same metric makes it useless. Something like ‘Sq. Feet under surveillance’ might be better.
Okay, will we at some point admit that we want machine learning algorithms to be able to interface with symbolic rules rather than pretending that everything is a data issue as if we're living in the era of 1930s inductionism? It's clear that the misgendering issue here is not a stochastic one that ought to be solved by 'balancing out' data, it's that we want to impose strict linguistic rules and constraints on a system in a clear manner.
There really needs to be more work done in AI that makes it possible to interface with the models we built rather than trying to reframe everything as a data problem and then shove it in some end-to-end black box and then hope that whatever comes out at the other end is correct.
The automated systems used in the article are supposed to make judgements about "Detection of body movements that constitute assault." This requires genuine understanding and high-level capacity to reason rather than just pixel-based inference from some camera.