The general consensus in my professional network, even reaching up to EU SME’s (!), is to just geoblock the UK given the costs to reach compliance are usually much higher than any business brought in by possible UK customers. This is also in light of other impeding UK legislation.
I would also do the same here in this case, if I were you @pushcx. This is a community of tech-literate people who know how to get a VPN. At the same time, the (slight) inconvenience is a constant reminder of this overbearing legislation for all UK visitors, and it might have a stronger effect on the situation than we could imagine.
It shocks me time and time again how much of a postmodern cyberpunk surveillance- and police-state the UK has become. The UK’s people seems to be content with this development given they voted for it, so who I am to judge? Fortunately I don’t have to live there.
Because they actually believe it is a good thing, and would have passed such an act themselves if it didn’t already exist.
It is very easy to repeal an Act which has not yet come in to force. The coalition gov quickly repealed the ID card legislation despite it already being in force, with its active database, and cards already issued.
So by their deeds (and/or inaction’s) you can determine their beliefs.
I suspect it’s more that they simply don’t understand it. My MP is one of them and he is, uh, not the brightest. I’ve never had an interaction with him where I felt he understood the legislation that he was voting on at more than a superficial level. His level of reasoning is probably ‘child abuse bad, law says child abuse bad, law must be good’ rather than having any actual understanding of the impact of the law.
Not sure. There’s a fair overlap between Labour today and Labour 17 years ago, and you may recall the Terrorism Act 2006, which had similarly vague formulations about what exactly constituted “glorification of terrorism” (which in turn provoked a bunch of SF authors to release a book called “Glorifying Terrorism” in order to see what would happen). It’s not unlikely they view having a law that can be inconsistently applied (“oh but we don’t mean you, you’re one of the good ones”, “we know it when we see it”) as an actual feature - after all, they don’t bear the costs and neither does anyone they care about.
To be fair, irrespective of the belief / intelligence of individual MPs, most are backbenchers with very little power within the party and just follow the whip (especially considering how many are new MPs)
In this particular case they probably agree with it but also it’s a general fact (I haven’t actually checked this - it’s just from my experience - so somebody please correct me if I’m wrong - https://en.wikipedia.org/wiki/Ward_Cunningham#%22Cunningham's_Law%22 !) that governments repeal almost nothing. They almost universally prefer to spend their limited energy on passing new laws. I’m not sure why that is - maybe something to do with Pareto optimality.
the primary goal of most politicians is to get re-elected. due to sample bias, this is especially true with ones who’ve been around longer and thus have the seniority to influence agendas at the party level.
passing a new law is something that it’s easy to brag about and take credit for. it presents the public with an achievement that one or more involved politicians can attach their names to, or over-state their role in if they’re so inclined.
repealing a new law is not nearly as easy to convince the public to congratulate people on, since the net result is that there’s nothing to look at, no new results to point to. it’s certainly the case that there are parties that build their brand on repealing laws as their primary activity, but there is an inherent disadvantage to doing so.
Indeed: it is most probably on the new government’s radar, but they might choose to allow shit to hit the fan before “saving the day”.
We can draw a parallel with hero culture in many dysfunctional companies: there’s no glory to gain by avoiding an incident while the employee who solves an incident get credit, even if they caused it in the first place.
The intentional invocation of Cunningham’s law went right over my head. Shame on me for reading before finishing the first coffee.
I wonder if intentional invocation of Cunningham’s law is like intentional invocation of Murphy’s law… such an act might well turn the incorrect statement into a question and therefore place it on the wrong side of the law.
Many websites don’t support a + in email addresses either, and only support some custom subset of characters instead. I’ve already collected more than 25 cases personally, which is extremely annoying.
In this case, the person was supplying an image (presumably of herself with Asian features), so the content of the image should serve as an implicit prompt. Having a person with Asian features having to specify that they are Asian is in itself an example of bias.
If I am a single person looking for a single photo for some reason, sure.
When these generated images are used in mass, thinking about bias in the system (or model, in this case) is important.
If you ask AI to generate an image of a board room filled with executives, likely due to bias in training material you’re going to end up with a board room of all white men. This perpetuates the idea that executives can only be white men. If these images are then used socially, it’s going to reinforce these biases in society.
That’s typically not a good thing, and something a lot of people care about addressing.
When I say “used in mass”, I don’t mean a single person generating and using a mass set of images to use. I mean the general population, various companies, random untrained users, etc. Not all of these people are going to notice, understand, or care about the general implications of biased output.
That’s why biases are important - you don’t want the default to be harmful, because most usage will be those defaults. Outside of diversity implications, it’s useful to consider that these very visible biases can lead to better understanding of the less visible biases that might be present in training data - we may not even be realizing some of the significant biases that will result in poor or even incorrect results. There’s a great example in the original article if you search “oil change appointment”.
An individualized approach isn’t a good solution to a problem that’s broader than one user. “Just do this workaround” is a fine approach in the short term for software development, but a shoddy approach long term.
This is why people care what gender or race is generated when you do not specify - not because they’re concerned about the individual image they’re creating, but because they’re concerned about the hundreds of thousands generated by other people and what effect that can have.
If I’m generating a lot of pictures, I expect the underlying distribution to be reflected. If it can’t even clear a low bar like a χ² test, it’s biased.
E.g., if I ask for 100 CEO pics, and get 0 women, it’s failed.
The general consensus in my professional network, even reaching up to EU SME’s (!), is to just geoblock the UK given the costs to reach compliance are usually much higher than any business brought in by possible UK customers. This is also in light of other impeding UK legislation.
I would also do the same here in this case, if I were you @pushcx. This is a community of tech-literate people who know how to get a VPN. At the same time, the (slight) inconvenience is a constant reminder of this overbearing legislation for all UK visitors, and it might have a stronger effect on the situation than we could imagine.
It shocks me time and time again how much of a postmodern cyberpunk surveillance- and police-state the UK has become. The UK’s people seems to be content with this development given they voted for it, so who I am to judge? Fortunately I don’t have to live there.
We voted out the government that passed this stupid law. Unfortunately, the new government has not repealed it.
Just out of curiosity, why haven’t they?
Because they actually believe it is a good thing, and would have passed such an act themselves if it didn’t already exist.
It is very easy to repeal an Act which has not yet come in to force. The coalition gov quickly repealed the ID card legislation despite it already being in force, with its active database, and cards already issued.
So by their deeds (and/or inaction’s) you can determine their beliefs.
I suspect it’s more that they simply don’t understand it. My MP is one of them and he is, uh, not the brightest. I’ve never had an interaction with him where I felt he understood the legislation that he was voting on at more than a superficial level. His level of reasoning is probably ‘child abuse bad, law says child abuse bad, law must be good’ rather than having any actual understanding of the impact of the law.
Not sure. There’s a fair overlap between Labour today and Labour 17 years ago, and you may recall the Terrorism Act 2006, which had similarly vague formulations about what exactly constituted “glorification of terrorism” (which in turn provoked a bunch of SF authors to release a book called “Glorifying Terrorism” in order to see what would happen). It’s not unlikely they view having a law that can be inconsistently applied (“oh but we don’t mean you, you’re one of the good ones”, “we know it when we see it”) as an actual feature - after all, they don’t bear the costs and neither does anyone they care about.
To be fair, irrespective of the belief / intelligence of individual MPs, most are backbenchers with very little power within the party and just follow the whip (especially considering how many are new MPs)
In this particular case they probably agree with it but also it’s a general fact (I haven’t actually checked this - it’s just from my experience - so somebody please correct me if I’m wrong - https://en.wikipedia.org/wiki/Ward_Cunningham#%22Cunningham's_Law%22 !) that governments repeal almost nothing. They almost universally prefer to spend their limited energy on passing new laws. I’m not sure why that is - maybe something to do with Pareto optimality.
the primary goal of most politicians is to get re-elected. due to sample bias, this is especially true with ones who’ve been around longer and thus have the seniority to influence agendas at the party level.
passing a new law is something that it’s easy to brag about and take credit for. it presents the public with an achievement that one or more involved politicians can attach their names to, or over-state their role in if they’re so inclined.
repealing a new law is not nearly as easy to convince the public to congratulate people on, since the net result is that there’s nothing to look at, no new results to point to. it’s certainly the case that there are parties that build their brand on repealing laws as their primary activity, but there is an inherent disadvantage to doing so.
Indeed: it is most probably on the new government’s radar, but they might choose to allow shit to hit the fan before “saving the day”.
We can draw a parallel with hero culture in many dysfunctional companies: there’s no glory to gain by avoiding an incident while the employee who solves an incident get credit, even if they caused it in the first place.
oh good parallel, indeed
(edit: typo)
You linked
https://en.wikipedia.org/wiki/Ward_Cunningham#%22Cunningham's_Law%22
which talks about how Ward Cunningham, a computer programmer, is credited with the idea that
Is that actually what you meant to link? If so, I’m not quite seeing the relationship to what governments do or don’t repeal.
Parenthesis: it applies to the statement before the link. Let me highlight it:
So it applies to the fact they didn’t check, not to what governments do.
The intentional invocation of Cunningham’s law went right over my head. Shame on me for reading before finishing the first coffee.
I wonder if intentional invocation of Cunningham’s law is like intentional invocation of Murphy’s law… such an act might well turn the incorrect statement into a question and therefore place it on the wrong side of the law.
Haha, no problem, happy to be of help. It took me a close reading as well.
I believe they are saying “I’m saying this and if I’m incorrect someone will surely correct me”, that is - utilizing Cunningham’s law themselves
You are, of course, correct. That went right over my head.
The media ignores it so it’s hard to blame the general populace.
I probably wouldn’t bother.
Many websites don’t support a
+in email addresses either, and only support some custom subset of characters instead. I’ve already collected more than 25 cases personally, which is extremely annoying.I had one that allowed + in the email for signup, but the password reset form did not accept emails with a +
I’ve seen this same problem with unsubscribe forms. Absolutely ridiculous. Straight to the spam bin.
I have not understood why people care what gender or race the picture is generated when you do not specify gender or race.
If you want a picture generated which is a specific gender or race you can just add that information to the prompt.
In this case, the person was supplying an image (presumably of herself with Asian features), so the content of the image should serve as an implicit prompt. Having a person with Asian features having to specify that they are Asian is in itself an example of bias.
If I am a single person looking for a single photo for some reason, sure.
When these generated images are used in mass, thinking about bias in the system (or model, in this case) is important.
If you ask AI to generate an image of a board room filled with executives, likely due to bias in training material you’re going to end up with a board room of all white men. This perpetuates the idea that executives can only be white men. If these images are then used socially, it’s going to reinforce these biases in society.
That’s typically not a good thing, and something a lot of people care about addressing.
if you’re generating a lot of pictures your script can just add the race you want to the prompts at the ratio you want.
When I say “used in mass”, I don’t mean a single person generating and using a mass set of images to use. I mean the general population, various companies, random untrained users, etc. Not all of these people are going to notice, understand, or care about the general implications of biased output.
That’s why biases are important - you don’t want the default to be harmful, because most usage will be those defaults. Outside of diversity implications, it’s useful to consider that these very visible biases can lead to better understanding of the less visible biases that might be present in training data - we may not even be realizing some of the significant biases that will result in poor or even incorrect results. There’s a great example in the original article if you search “oil change appointment”.
An individualized approach isn’t a good solution to a problem that’s broader than one user. “Just do this workaround” is a fine approach in the short term for software development, but a shoddy approach long term.
This is why people care what gender or race is generated when you do not specify - not because they’re concerned about the individual image they’re creating, but because they’re concerned about the hundreds of thousands generated by other people and what effect that can have.
If I’m generating a lot of pictures, I expect the underlying distribution to be reflected. If it can’t even clear a low bar like a χ² test, it’s biased.
E.g., if I ask for 100 CEO pics, and get 0 women, it’s failed.
again you can just prompt it for a “women CEO” if you want that result.
Yes, so? What if I want accurate general results instead?
You’re deliberately ignoring the actual problem at this point, which is not “How do I get Dall-E to generate a female CEO?”