There certainly are international rights issues at play. Consider Carpool Karaoke, in many places that kind of content needs to licence the music for performance. Same with music used on dramas and documentaries.
They will have to get licences for jurisdictions they want to stream/sell into, same as Netflix. It's a major hassle for Netflix, especially on shows like Luke Cage where there is a lot of music that is really important for the show and can't easily be replaced.
It's very rare that a single person builds any kind of complex system and then promotes/sells it on their own. More likely there will be a team of people with somewhat diverse views on ethics, so at least there might be a debate.
It's much easier to scream about Leftists and groupthink, but in reality most things are a group effort and ethics are a matter for discussion.
That's getting a little bit ahead of things. Most courses in engineering ethics focus on understanding how engineering affects people and the world. Just thinking about the wider implications beyond "hay we can do this" is the point.
An oft cited example is leaded petrol. The goal was to reduce engine knocking, and it worked. But it had some really bad consequences too. Consequences that were hard to predict at the time, and which some argue were worth accepting for the benefits we got in return. The goal of ethics in engineering courses is to teach people to think about the consequences beyond the mere chemistry, the fact that a substance is being introduced into the environment and people are being exposed to it, and that at the time the consequences were not clear.
Considering that Mozilla has a long history of supporting social/community projects and was in fact built on the idea of being an active member of the open source community, it seems unlikely that they are duping investors.
So you are saying that the problem is one of perception? I'd point out that you are not making it any better with your statements, but at least we agree on what the problem is.
By the way, that line about "all men are rapists" was actually from a play. In the play a woman has been raped, and her upset mother says the line. The daughter then immediately contradicts her and argues that there are good men and she won't give up on them because of her experience.
I accept that the risk is non-zero. I pointed out that the probability is extremely low, and worrying about it is probably irrational for most people because of that. You countered that it's not irrational and cited some anecdotal and, in the case of Linus, likely false evidence.
You now seem to be arguing that any non-zero chance makes it rational to take the rather extreme steps you outlined, such as not being in a room with a closed door with women and not dining alone with a woman. Yet I imagine you still get into cars and cross the road, despite the risk being vastly greater, both in probability of happening and the potential injuries.
If you are sticking to the >0% chance justification then can you explain why you take on other non-zero risks with potentially devastating consequences that are actually far more likely to affect you.
So what is your stat for false accusations? How many false accusations per year are we talking about, or what percentage chance of being accused per year or per lifetime do you think is correct, and what are your sources?
All men are however at risk of a false accusation.
The important question is how great a risk is it?
You could be hit by lightning. There might be a sniper randomly picking people off today. Your house might be built over a sinkhole. These are all things that happen from time to time but you don't worry about them much because they are rare.
On the other hand people do worry about things like terrorism, even though the probability of being caught up in a terror attack is tiny. Much lower than being killed crossing the road. And it's mostly due to terror attacks getting so much publicity, but road accidents rarely making the news.
Stanford says that false rape claims run to about 2%, similar to other types of crime. According to the Equal Opportunity Commission the number of sexual harassment claims has actually/fallen/ since the 90s., and the number of complaints last year puts the chance of being accused at 1 in 14,000. Of course it's actually much lower because many of those complaints are against the same person for multiple incidents.
And of course that's all complaints, not just the false ones.
For comparison the chances of being in a motor vehicle accident are about 1 in 8,000. You are several times more likely to be poisoned by something you eat than falsely accused in the absolute worst case.
In Japan the biggest issue is employers. There is now excellent childcare available, but wages are not high enough and mothers get penalised for taking maternity leave. Many women are prioritising their careers or couples are opting to only have one child because of this.
Amazon will still need humans for now, because Uniqlo has one big advantage over them. Uniqlo only stocks its own products (all clothing) and can package them in a way that the robots can handle. Amazon has to deal with a huge variety of shapes and sizes and packaging types.
It's much easier to make a robot that picks and packs regularly shaped clothing packets (Uniqlo sells a lot of stuff in cardboard stiffened plastic packets for example) than one which can handle arbitrary shapes like a human can.
It's a fine balance between not wanting frivolous lawsuits flying all over the place and the fact that some companies rely on people not suing them as part of their business model. Parking fines are a good example, often they are bogus but many people just pay up because they don't know they can just file a lawsuit and it will magically go away.
We had this in the UK for maybe two decades now, although the Small Claims Court goes back further. You submit a simple online form and pay the fee (last time I checked it was £30, which the other party pays if you win).
You can use it to claim money from anyone or any organization for any reason. It's largely risk free, you don't need a lawyer (the judges are generally sympathetic and they keep the procedural stuff to a minimum) and if you lose you don't normally have to pay the other party's costs. There are exceptions if your claim was completely ridiculous but they are fairly rare.
It's very useful for dealing with parking fine scams and companies that rip you off. I've used it with Intel over their Meltdown/Spectre flaws, and against the Royal Mail for losing my package and then trying to weasel out of paying compensation. I'm currently looking at options for dealing with PayPal, but the Information Commissioner takes a while to sort out the international GDPR violations that need to be resolved before I can proceed.
This is why we need algorithmic transparency, so we can understand how algorithms and AI make decisions. In this case TFA says they could see it had decided that resumes with the phrase "women's" were down-ranked, and ones with masculine language like "executed" and "captured" were up-ranked.
It's possible that it was using those as proxies for some hidden factor that has eluded us, but unless we can figure out what that factor is it's not really fair to just assume it's there and not the more likely explanation that it has picked up the same biases as the humans it is using as references for ranking.
I mean would you be happy if you bank said "sorry, we can't give you a mortgage because the computer says no, and even though your application looks good to us we just assume it knows something we don't"?
All that is needed to eliminate a bias arising from dataset imbalance is to balance the dataset.
You say that like it's a trivial task. They are evaluating people for a job, and most of the current evaluation data is either subjective or biased or both.
There are long known and well researched gender differences in interest preference
For the sake of argument let's say I accept that statement, what does interest preference have to do with ranking people who applied for your tech job? Surely the fact that they applied implies that they have an interest.
It says that the algorithm specifically decided that phrases like "women's" were negative attributes (i.e. anyone who attended a women's college or was a member of a women's group was down-ranked). It explains quite clearly how that happened and what they did about it.
You can change the default... It's just that most people don't. And the issue is that manufacturers don't have a choice, Google has to be to default search engine to get the Play Store.
Their assumption that women are not inherently less competent is not broken.
Their goal of recognizing that in their hiring practices is not broken either, it's a smart business stratergy to get high quality employees that flaws in their system meant they were passing up previously.
Correct, but it can lead to systemic bias due to incomplete information or error. Also, very little is as empirical as you think.
Take a degree from MIT, for example. It suggests great ability, but is it empirical evidence? Are all MIT degrees equal? And how do you compare between MIT and other universities? And how do you ensure that the perceived value from those institutions is free from human biases against things like the state they are in, their historic reputation vs. their current teaching quality, and their political leanings?
Another example is confirmation bias. The chief of police has limited resources, and sees that crime is 10% higher in one area. So they send more cops to that area, and unsurprisingly the crime rate goes up to 20% higher due to increased detection rates. Is the 20% higher crime rate empirical evidence? Is it complete information that you can use to make a decision like where to live for lower crime? Seems like the 20% lower crime area may just have more unreported/unsolved crime, so might actually be worse. And what type of crime are we talking about, 20% more graffiti but the other side of town is mostly violence?
Sure, but it's a manual process that they have to do themselves, and then pay for the bandwidth to push out.
If you have the Play Store and Play Store Services running you updates direct from Google and other benefits like anti-virus scanning of installed apps, payment handling for in-app purchases etc.
Amazon is big enough to replicate all that, but many OEMs prefer to just let Google handle it. Which is fine, all that's not fine is that then Google forces them to install Chrome and Search as the defaults.
it's someone smugly waiving hands and pointing fingers, shouting "RACIST," etc. while trying really hard not to furiously masturbate
You know I've never actually met anyone like that... On the other hand there are loads of virtue signalling self-proclaimed centrist rationals on Slashdot proudly proclaiming that they are far above such strawman nonsense.
Instead of trying to derail the conversation to make yourself look good, why not have an honest debate about the problems of trying to teach AI using purely empirical data?
Studies show that humans have bias towards more masculine language, so it's probably that which is affecting the outcome. If human reviewers rate more masculine language higher than more feminine language, the AI will learn from their example of what they consider "best".
This can be re-enforced by problems once women start working at Amazon that result in them being promoted at a slower pace or lower rate too. That AI learns that they tend to "perform worse" in the job, because it doesn't understand the existing biases in the system.
I know, I'm pointing out that many of the AAA games people want to play come out on the PS4 and XBOX. You might get a Switch support, likely inferior due to its lower power but not always in the game of games like Fortnight. And you might get a PC port, but then you need to buy and maintain an expensive gaming PC.
So really for a lot of people, especially kids, it's PS4 or XBOX. And XBOX isn't really any better in terms of security or having all your stuff tied to your account.
Indeed they supplied schematics, sometimes on request, for a lot of 8 bit machines. But I'm talking about the internals of the video chip, which were only documented from the point of view of the control registers and overall operation.
By hacking and experimenting people figured out the internal workings down to the memory access timings and sequences.
Slashdot Mirror
There certainly are international rights issues at play. Consider Carpool Karaoke, in many places that kind of content needs to licence the music for performance. Same with music used on dramas and documentaries.
They will have to get licences for jurisdictions they want to stream/sell into, same as Netflix. It's a major hassle for Netflix, especially on shows like Luke Cage where there is a lot of music that is really important for the show and can't easily be replaced.
It's very rare that a single person builds any kind of complex system and then promotes/sells it on their own. More likely there will be a team of people with somewhat diverse views on ethics, so at least there might be a debate.
It's much easier to scream about Leftists and groupthink, but in reality most things are a group effort and ethics are a matter for discussion.
That's getting a little bit ahead of things. Most courses in engineering ethics focus on understanding how engineering affects people and the world. Just thinking about the wider implications beyond "hay we can do this" is the point.
An oft cited example is leaded petrol. The goal was to reduce engine knocking, and it worked. But it had some really bad consequences too. Consequences that were hard to predict at the time, and which some argue were worth accepting for the benefits we got in return. The goal of ethics in engineering courses is to teach people to think about the consequences beyond the mere chemistry, the fact that a substance is being introduced into the environment and people are being exposed to it, and that at the time the consequences were not clear.
Considering that Mozilla has a long history of supporting social/community projects and was in fact built on the idea of being an active member of the open source community, it seems unlikely that they are duping investors.
So you are saying that the problem is one of perception? I'd point out that you are not making it any better with your statements, but at least we agree on what the problem is.
By the way, that line about "all men are rapists" was actually from a play. In the play a woman has been raped, and her upset mother says the line. The daughter then immediately contradicts her and argues that there are good men and she won't give up on them because of her experience.
I accept that the risk is non-zero. I pointed out that the probability is extremely low, and worrying about it is probably irrational for most people because of that. You countered that it's not irrational and cited some anecdotal and, in the case of Linus, likely false evidence.
You now seem to be arguing that any non-zero chance makes it rational to take the rather extreme steps you outlined, such as not being in a room with a closed door with women and not dining alone with a woman. Yet I imagine you still get into cars and cross the road, despite the risk being vastly greater, both in probability of happening and the potential injuries.
If you are sticking to the >0% chance justification then can you explain why you take on other non-zero risks with potentially devastating consequences that are actually far more likely to affect you.
So what is your stat for false accusations? How many false accusations per year are we talking about, or what percentage chance of being accused per year or per lifetime do you think is correct, and what are your sources?
All men are however at risk of a false accusation.
The important question is how great a risk is it?
You could be hit by lightning. There might be a sniper randomly picking people off today. Your house might be built over a sinkhole. These are all things that happen from time to time but you don't worry about them much because they are rare.
On the other hand people do worry about things like terrorism, even though the probability of being caught up in a terror attack is tiny. Much lower than being killed crossing the road. And it's mostly due to terror attacks getting so much publicity, but road accidents rarely making the news.
Stanford says that false rape claims run to about 2%, similar to other types of crime. According to the Equal Opportunity Commission the number of sexual harassment claims has actually /fallen/ since the 90s., and the number of complaints last year puts the chance of being accused at 1 in 14,000. Of course it's actually much lower because many of those complaints are against the same person for multiple incidents.
And of course that's all complaints, not just the false ones.
For comparison the chances of being in a motor vehicle accident are about 1 in 8,000. You are several times more likely to be poisoned by something you eat than falsely accused in the absolute worst case.
In Japan the biggest issue is employers. There is now excellent childcare available, but wages are not high enough and mothers get penalised for taking maternity leave. Many women are prioritising their careers or couples are opting to only have one child because of this.
Amazon will still need humans for now, because Uniqlo has one big advantage over them. Uniqlo only stocks its own products (all clothing) and can package them in a way that the robots can handle. Amazon has to deal with a huge variety of shapes and sizes and packaging types.
It's much easier to make a robot that picks and packs regularly shaped clothing packets (Uniqlo sells a lot of stuff in cardboard stiffened plastic packets for example) than one which can handle arbitrary shapes like a human can.
It's a fine balance between not wanting frivolous lawsuits flying all over the place and the fact that some companies rely on people not suing them as part of their business model. Parking fines are a good example, often they are bogus but many people just pay up because they don't know they can just file a lawsuit and it will magically go away.
We had this in the UK for maybe two decades now, although the Small Claims Court goes back further. You submit a simple online form and pay the fee (last time I checked it was £30, which the other party pays if you win).
You can use it to claim money from anyone or any organization for any reason. It's largely risk free, you don't need a lawyer (the judges are generally sympathetic and they keep the procedural stuff to a minimum) and if you lose you don't normally have to pay the other party's costs. There are exceptions if your claim was completely ridiculous but they are fairly rare.
It's very useful for dealing with parking fine scams and companies that rip you off. I've used it with Intel over their Meltdown/Spectre flaws, and against the Royal Mail for losing my package and then trying to weasel out of paying compensation. I'm currently looking at options for dealing with PayPal, but the Information Commissioner takes a while to sort out the international GDPR violations that need to be resolved before I can proceed.
This is why we need algorithmic transparency, so we can understand how algorithms and AI make decisions. In this case TFA says they could see it had decided that resumes with the phrase "women's" were down-ranked, and ones with masculine language like "executed" and "captured" were up-ranked.
It's possible that it was using those as proxies for some hidden factor that has eluded us, but unless we can figure out what that factor is it's not really fair to just assume it's there and not the more likely explanation that it has picked up the same biases as the humans it is using as references for ranking.
I mean would you be happy if you bank said "sorry, we can't give you a mortgage because the computer says no, and even though your application looks good to us we just assume it knows something we don't"?
All that is needed to eliminate a bias arising from dataset imbalance is to balance the dataset.
You say that like it's a trivial task. They are evaluating people for a job, and most of the current evaluation data is either subjective or biased or both.
There are long known and well researched gender differences in interest preference
For the sake of argument let's say I accept that statement, what does interest preference have to do with ranking people who applied for your tech job? Surely the fact that they applied implies that they have an interest.
It says that the algorithm specifically decided that phrases like "women's" were negative attributes (i.e. anyone who attended a women's college or was a member of a women's group was down-ranked). It explains quite clearly how that happened and what they did about it.
Did you really read TFA?
You can change the default... It's just that most people don't. And the issue is that manufacturers don't have a choice, Google has to be to default search engine to get the Play Store.
Their assumption that women are not inherently less competent is not broken.
Their goal of recognizing that in their hiring practices is not broken either, it's a smart business stratergy to get high quality employees that flaws in their system meant they were passing up previously.
Empirical data isn't racist or sexist
Correct, but it can lead to systemic bias due to incomplete information or error. Also, very little is as empirical as you think.
Take a degree from MIT, for example. It suggests great ability, but is it empirical evidence? Are all MIT degrees equal? And how do you compare between MIT and other universities? And how do you ensure that the perceived value from those institutions is free from human biases against things like the state they are in, their historic reputation vs. their current teaching quality, and their political leanings?
Another example is confirmation bias. The chief of police has limited resources, and sees that crime is 10% higher in one area. So they send more cops to that area, and unsurprisingly the crime rate goes up to 20% higher due to increased detection rates. Is the 20% higher crime rate empirical evidence? Is it complete information that you can use to make a decision like where to live for lower crime? Seems like the 20% lower crime area may just have more unreported/unsolved crime, so might actually be worse. And what type of crime are we talking about, 20% more graffiti but the other side of town is mostly violence?
Sure, but it's a manual process that they have to do themselves, and then pay for the bandwidth to push out.
If you have the Play Store and Play Store Services running you updates direct from Google and other benefits like anti-virus scanning of installed apps, payment handling for in-app purchases etc.
Amazon is big enough to replicate all that, but many OEMs prefer to just let Google handle it. Which is fine, all that's not fine is that then Google forces them to install Chrome and Search as the defaults.
it's someone smugly waiving hands and pointing fingers, shouting "RACIST," etc. while trying really hard not to furiously masturbate
You know I've never actually met anyone like that... On the other hand there are loads of virtue signalling self-proclaimed centrist rationals on Slashdot proudly proclaiming that they are far above such strawman nonsense.
Instead of trying to derail the conversation to make yourself look good, why not have an honest debate about the problems of trying to teach AI using purely empirical data?
Studies show that humans have bias towards more masculine language, so it's probably that which is affecting the outcome. If human reviewers rate more masculine language higher than more feminine language, the AI will learn from their example of what they consider "best".
This can be re-enforced by problems once women start working at Amazon that result in them being promoted at a slower pace or lower rate too. That AI learns that they tend to "perform worse" in the job, because it doesn't understand the existing biases in the system.
That's true to some extent, it certainly was a powerful machine. In the other hand the graphics were somewhat limited for games and it was expensive.
One thing that really impressed me about the BBC Micro and the later Archimedes was how efficient the BASIC interpreter was.
I know, I'm pointing out that many of the AAA games people want to play come out on the PS4 and XBOX. You might get a Switch support, likely inferior due to its lower power but not always in the game of games like Fortnight. And you might get a PC port, but then you need to buy and maintain an expensive gaming PC.
So really for a lot of people, especially kids, it's PS4 or XBOX. And XBOX isn't really any better in terms of security or having all your stuff tied to your account.
Indeed they supplied schematics, sometimes on request, for a lot of 8 bit machines. But I'm talking about the internals of the video chip, which were only documented from the point of view of the control registers and overall operation.
By hacking and experimenting people figured out the internal workings down to the memory access timings and sequences.