> But] our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range…This wasn’t what we intended.
Er, ok. It might have been an idea to test the output then. Can anybody take this seriously? Did they seriously go into a meeting and agree to add “diverse” to any input even if people were looking for Irish peasants in the 19C, or a Scottish highlander circa 185.
They surely run test prompts.
So either they didn’t test this, in which case the entire test team should be fired, or they did test this and didn’t see any problem - in which case the entire test team should be fired.
Regarding why audiences are okay with a diverse cast for Hamilton, I don't think there's a strong expectation the cast matches the physical look of the subjects in plays. Theatre (everything from the set to the actors) is never realistic, so the expectation is "does the performance capture some essence of the subject".
Ultimately this aligns with the point you made, Puerto Rican Hamilton is okay because the essence of Hamilton is not that he's white.
"One invites you to suspend disbelief, while the other orders you to accept a lie."
This is the perfect line to describe the difference between Hamilton and Gemini. One could argue the whole point of fiction is to invite the reader/viewer to suspend disbelief. It doesn't always work, but people have an easier time with art doing this than with a fact-gathering machine like Google.
There are some aspects of this essay that are thought provoking, especially about the effect of statues, and kudos (I guess) for capitalizing white as well as black. Though I'm not sure how this convention is meant to bring people together vs just emphasising differences.
But I feel it misses the mark because it dwells entirely on race. Gemini isn't a racist chatbot, it's a hard left chatbot and anti-white racism is just the subcomponent of it that happened to go viral. You could have just as easily written about the erasure of men. I'm pretty sure there were not many if any women serving as front-line Nazi infantry, but it forced them in there anyway. On the other hand for that there isn't any angle about multi-cultural society, or taking shortcuts to achieve it. Women are the majority of the population, they have many legal privileges men don't have as well as of course numerous conventions that unfairly privilege them too. They are very much in the more powerful position and Gemini reflects that, where it constantly replaces not only whites with blacks/native Americans but also men with women.
And it doesn't stop there. Ask Gemini about Stalin and it'll tell you that his legacy is complex and multi-faceted, that he did a lot of good things too. Anything on climate or COVID is of course a waste of time. The paintings of historical figures went viral because the manipulative and non-factual nature of left wing output is undeniable and obvious in that case, but there is a huge space of less in-your-face outputs where it also just makes things up, lies to you or otherwise tries to manipulate you that people won't take umbrage with because they've become inured to it and expect it.
So I feel pretty clearly that the issue here isn't Google trying to take shortcuts to a better world. It's much simpler. They purged everyone who isn't far left, as far left people often do, and now they had a chance to create the ultimate New Soviet Man of their dreams. No surprise that it refuses to depict reality accurately in thousands of different ways. It's exactly what you'd expect and deep down has nothing to do with race.
"[Antiracist discrimination is] likely to have the opposite effect — pushing more White people into a bitter, defensive embrace of White racial identity."
I was born into a liberal Jewish family. Race was a non-issue, I took MLK's "judge a man by the content of his character" creed to heart. But over the course of my adult life more and more organizations have chosen to draw race-based battle-lines. What did they think was going to happen?
“There are definitely still plenty of organizations out there that discriminate against nonwhite people,”
The link in that statement is to an article about IRS auditing blacks at a disproportionate rate. Here’s a quote from that article: “The researchers suggest that’s because the IRS set up systems that are more likely to flag tax returns with potential mistakes in how some tax credits, like the earned-income tax credit (EITC), are claimed. There is no evidence that Blacks cheat on their taxes, but they do file at disproportionately higher rates the kinds of returns that the IRS approach targets.”
That’s not a very good example of discrimination based on race. I bring this up because I wonder if in 2024 there actually is still a lot of organization-level racial discrimination against non-whites. Because it seems like the successful elimination of organization-level racial bias is part of the reason that the anti-racism movement has had to move to less objective targets like white culture.
The tragedy to me about this current moment is that when my boys, now nearing 20, were young teenagers, they truly did not notice race. They weren’t “programmed to think right” – they just didn’t worry much about it. They had friends from different backgrounds of different races and nationalities and religions, and it was very ho-hum to them. To me this represented progress towards what I’d thought the goal of the whole civil rights movement was, that is, to be judged on the content of one’s character and not the color of one’s skin. I thought we were headed solidly in the right direction and I had a lot of optimism for the future. It seems that we’ve gone radically off track, as if getting too close to achieving the goal of an integrated – or at least colorblind, although I know that’s a disfavored notion – society was simply too much for, in this case, progressive intellectuals to handle. Hopefully we can get back on track…
A quick comment pushing back on your statement that TV shows diverse characters to represent the country.
TV absolutely fails on this regard, especially in relation to Hispanics. The random TV show or movie is more likely to have a gay person than a Hispanic character despite the former representing a much smaller segment of the population.
As a person who has been involved in a large number of "DEI" type trainings, both as a recipient and as training manager, I have seen a version of the Google mistake made consistently.
My experience has been that it comes from a lack of empathy and a certain degree of arrogance. Basically, when you want to influence someone (which is what the AI was attempting), you have to see the world through their eyes. This is so you can generate content that they find compelling. You also have to get them to least minimally suspend their inherent objection to being manipulated. This is to prevent a backlash and increase receptivity to the training.
If you lack empathy with the people you want to influence, you generate content which they will find compelling. If you are arrogant, you lack the insight that they might find your content objectionable (after all you are the really smart person who knows what everyone should think) and lack the ability to introspect. This prevents you from recognizing your errors.
In my experience, when a training suffers from these issues it leads to what we called "training scars."
This means the training causes more harm than benefits (i.e., people come out of a DEI training less receptive to the principles of DEI). This is because the only people who find the content compelling and are willing to be manipulated already share the beliefs being pushed at the training. They do not benefit much from the training since they are already on board. Anyone even a tiny bit skeptical comes away feeling pissed off and more oppositional than when they went in.
So basically, all Google has done with this is probably push away all kinds of people who will now question all their products, including search. I think this will have a huge fiscal hit on Google as trust in their search engine is a huge competitive advantage. They have likely damaged trust with a large portion of them customers, and this damage will impact more than AI. Do you want to use a search engine that you believe is trying to manipulate you? You might if you have no other option but my guess is many people will jump ship to something else.
Finally, another issue I have seen with DEI training, and this probably is not related to the Google AI issue, is using the training to validate people concerns or buy them off. I saw this a lot as well...where the training was built to impress a bunch of people outside the organization (i.e., it was material they were receptive too) but in doing so they made the material objectionable to those being trained. Or the training is a way to funnel resources to a group that might otherwise publicly challenge or call out the organization. These trainings almost always inflict training scars as those being trained are generally offended by the content, and even if the content is decent, they tend to pick up on the real reason for the training and then get pissed about their time being used as part of deal to buy off others.
Here's the other problem; right now, people go to AI creators to give them correct information. No one ever asked Hamilton to be historically accurate, but if LLM's and AI image generators want to gain credibility, they have to start with depicting things as they actually are, not as the creators wanted them to be.
I love this article because you are so clear about how addressing inclusivity/diversity is critical to a high-functioning society (or company) yet is also something that we have been completely unable to fix in any consistent way.
I like to describe the present actions of companies as "look and feel" diversity. The objective (normally not explicitly so but in practice) is to ensure that the company looks diverse and people feel that there is respect for diversity. This leads to many, many shortcuts - eg diversity hiring quotas, etc...
This approach also completely sidesteps/ignores the underlying issues within all companies - hiring, promotion and recognition decisions are made entirely on a discretionary basis. Which means even in all-white companies, discrimination is common.
Typically, a manager will give preference to promoting people who they "click" with and who "get-things done". In a typical organisation, most people have the base ingredients to learn and do most jobs. This often meant people who are selected looked like the manager, went to similar universities, worked in similar companies, etc... It wasn't ever fair. Companies pushing for DE&I are now simply twisting the arms of those managers to preference people who "look" a certain way. This is likely why I'm tolerant of a certain amount of "anti-discrimination, discrimination". On some level, to me, it's the same thing from the past, just packaged differently.
If an organisation / society / etc really wants to address this, they have to first figure out a better way to make hiring, selection and recognition more transparent and fair such that the people involved from all backgrounds perceive that the process was fair (This is also something a lot of companies forget. They could have perfectly fair systems but if people in the company think they aren't fair or the results of those systems look 'bad', the systems will have no credibility).
In the past, the back-room talk was about how "they got the job because they golf with the boss". Now, it's that "they look a certain way". I don't think either is healthy but addressing it isn't a quick-fix, an offsite, and a leadership commitment to diversity.
I am not a believer that liberals want to get rid of white people, but it's really, really hard to refute that argument when a company has powerful as Google releases something like this. This. I mean, if you were trying to drive people to vote for Trump, you couldn't come up with anything better than this.
Racism is primarily about fear, mental models, and yes to a smaller degree about power. The fear response for human beings is tribalism. My sense is that all of this focus on race actually makes the process of removing racism harder. Rather, the shifts in identity (who is in my tribe) must happen gradually and naturally. A far more effective method is to have a common aspirational goals: Equality of opportunity, defeating a common enemy, aiming for a more perfect union..... the common cause in these sorts of things gradually shift the picture of internal identity.
On Google Gemini, I wouldn't take any of that too seriously.... science experiments can have unintended consequences.... I don't think there is a reason to be all that judgmental ... chuckle and move on. The reaction is more telling of the times vs the actual "offence."
On the economic research front, the data on "diversity" on boards/leadership is pretty sketchy. In my observation, breakout companies and new industries are almost always NOT diverse. Especially in the startup and early growth phase, the performance of the teams is aided by their common cultural identities. Now.. as the companies become large ... moving from racing boats to large tankers.. diversity in leadership helps in avoiding obstacles.
I actually prefer the overt forms of anti-racism like affirmative action. It does what it says on the tin. And it's easy and obvious to implement. If you say half of new hires or tv ads have to be people of color, then you have a clear cut way to measure success. And people can argue if that's too high or too low. But the other stuff of dealing with miccroagressions and hidden racism is just a scam. It's fostered this ginormous DEI industry full of people feeding from the trough of guilt. And there's no way it can be imposed from the top. And the cherry on the top is that 99% of the people it's meant to help don't think asking people to show up on time or do math was racist. So DEI people try to create problems where there wasn't any.
I'm gay, which has good and bad points vs say Black people. First, the huge majority of people don't care about gayness anymore. But I guess gays would be exposed to more offensive language and stuff than Blacks because of course it's less obvious. But what offends me (not very much, if anything) will differ dramatically from other people. And the way I deal with these "problems" when they do arise is also personal. I certainly won't think less of myself in any way, and I doubt anyone else is either. College kids usually referred to as snowflakes are just pretending. they aren't scarred from reading a book from the middle ages that condemns sodomy. Nobody thought there were gay pride parades in the 14th century, so stop faking hurt just for attention.
This is such a meandering and overlong dump of my feelings about DEI. So I'll get to the punch line. Teaching people via consultants accomplishes nothing other than to annoy everyone. You can't teach from above because it's way too general to be useful in any particular situation. And it's too remote to have any impact. Far better is to teach from the bottom up. The offended person can speak up, create a teaching moment. If I explain to someone why saying "man up" offends me (it doesn't but whatever) then that person is highly unlikely to say that again, and highly likely to understand things a bit better. And it's free!
In reading this commentary I am reminder of an organization in Poland : Forum for Dialogue http://dialog.org.pl/en/
Their mission is to promote communication between Poles today and the world wide Jewish community, especially Jews who once lived in Poland. I learned several things from this group: (1) you cannot change the past; (2) guilt is ineffective means of building trust/relationships between individuals and groups;. If you push guilt people will eventually push back and say hell no (3) regret is different than guilt. People from different groups can share guilt. These are all difficult tasks but I think Mr. Smith is on the right track here.
There is a great shortcut to building a “multiracial” society: marrying and having babies with someone of a different race than yourself. Until the rate at which people choose to do this increases by an order of magnitude, we will continue to be plagued by intense racial discord.
> But] our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range…This wasn’t what we intended.
Er, ok. It might have been an idea to test the output then. Can anybody take this seriously? Did they seriously go into a meeting and agree to add “diverse” to any input even if people were looking for Irish peasants in the 19C, or a Scottish highlander circa 185.
They surely run test prompts.
So either they didn’t test this, in which case the entire test team should be fired, or they did test this and didn’t see any problem - in which case the entire test team should be fired.
Regarding why audiences are okay with a diverse cast for Hamilton, I don't think there's a strong expectation the cast matches the physical look of the subjects in plays. Theatre (everything from the set to the actors) is never realistic, so the expectation is "does the performance capture some essence of the subject".
Ultimately this aligns with the point you made, Puerto Rican Hamilton is okay because the essence of Hamilton is not that he's white.
"One invites you to suspend disbelief, while the other orders you to accept a lie."
This is the perfect line to describe the difference between Hamilton and Gemini. One could argue the whole point of fiction is to invite the reader/viewer to suspend disbelief. It doesn't always work, but people have an easier time with art doing this than with a fact-gathering machine like Google.
There are some aspects of this essay that are thought provoking, especially about the effect of statues, and kudos (I guess) for capitalizing white as well as black. Though I'm not sure how this convention is meant to bring people together vs just emphasising differences.
But I feel it misses the mark because it dwells entirely on race. Gemini isn't a racist chatbot, it's a hard left chatbot and anti-white racism is just the subcomponent of it that happened to go viral. You could have just as easily written about the erasure of men. I'm pretty sure there were not many if any women serving as front-line Nazi infantry, but it forced them in there anyway. On the other hand for that there isn't any angle about multi-cultural society, or taking shortcuts to achieve it. Women are the majority of the population, they have many legal privileges men don't have as well as of course numerous conventions that unfairly privilege them too. They are very much in the more powerful position and Gemini reflects that, where it constantly replaces not only whites with blacks/native Americans but also men with women.
And it doesn't stop there. Ask Gemini about Stalin and it'll tell you that his legacy is complex and multi-faceted, that he did a lot of good things too. Anything on climate or COVID is of course a waste of time. The paintings of historical figures went viral because the manipulative and non-factual nature of left wing output is undeniable and obvious in that case, but there is a huge space of less in-your-face outputs where it also just makes things up, lies to you or otherwise tries to manipulate you that people won't take umbrage with because they've become inured to it and expect it.
So I feel pretty clearly that the issue here isn't Google trying to take shortcuts to a better world. It's much simpler. They purged everyone who isn't far left, as far left people often do, and now they had a chance to create the ultimate New Soviet Man of their dreams. No surprise that it refuses to depict reality accurately in thousands of different ways. It's exactly what you'd expect and deep down has nothing to do with race.
"[Antiracist discrimination is] likely to have the opposite effect — pushing more White people into a bitter, defensive embrace of White racial identity."
I was born into a liberal Jewish family. Race was a non-issue, I took MLK's "judge a man by the content of his character" creed to heart. But over the course of my adult life more and more organizations have chosen to draw race-based battle-lines. What did they think was going to happen?
“There are definitely still plenty of organizations out there that discriminate against nonwhite people,”
The link in that statement is to an article about IRS auditing blacks at a disproportionate rate. Here’s a quote from that article: “The researchers suggest that’s because the IRS set up systems that are more likely to flag tax returns with potential mistakes in how some tax credits, like the earned-income tax credit (EITC), are claimed. There is no evidence that Blacks cheat on their taxes, but they do file at disproportionately higher rates the kinds of returns that the IRS approach targets.”
That’s not a very good example of discrimination based on race. I bring this up because I wonder if in 2024 there actually is still a lot of organization-level racial discrimination against non-whites. Because it seems like the successful elimination of organization-level racial bias is part of the reason that the anti-racism movement has had to move to less objective targets like white culture.
The tragedy to me about this current moment is that when my boys, now nearing 20, were young teenagers, they truly did not notice race. They weren’t “programmed to think right” – they just didn’t worry much about it. They had friends from different backgrounds of different races and nationalities and religions, and it was very ho-hum to them. To me this represented progress towards what I’d thought the goal of the whole civil rights movement was, that is, to be judged on the content of one’s character and not the color of one’s skin. I thought we were headed solidly in the right direction and I had a lot of optimism for the future. It seems that we’ve gone radically off track, as if getting too close to achieving the goal of an integrated – or at least colorblind, although I know that’s a disfavored notion – society was simply too much for, in this case, progressive intellectuals to handle. Hopefully we can get back on track…
A quick comment pushing back on your statement that TV shows diverse characters to represent the country.
TV absolutely fails on this regard, especially in relation to Hispanics. The random TV show or movie is more likely to have a gay person than a Hispanic character despite the former representing a much smaller segment of the population.
As a person who has been involved in a large number of "DEI" type trainings, both as a recipient and as training manager, I have seen a version of the Google mistake made consistently.
My experience has been that it comes from a lack of empathy and a certain degree of arrogance. Basically, when you want to influence someone (which is what the AI was attempting), you have to see the world through their eyes. This is so you can generate content that they find compelling. You also have to get them to least minimally suspend their inherent objection to being manipulated. This is to prevent a backlash and increase receptivity to the training.
If you lack empathy with the people you want to influence, you generate content which they will find compelling. If you are arrogant, you lack the insight that they might find your content objectionable (after all you are the really smart person who knows what everyone should think) and lack the ability to introspect. This prevents you from recognizing your errors.
In my experience, when a training suffers from these issues it leads to what we called "training scars."
This means the training causes more harm than benefits (i.e., people come out of a DEI training less receptive to the principles of DEI). This is because the only people who find the content compelling and are willing to be manipulated already share the beliefs being pushed at the training. They do not benefit much from the training since they are already on board. Anyone even a tiny bit skeptical comes away feeling pissed off and more oppositional than when they went in.
So basically, all Google has done with this is probably push away all kinds of people who will now question all their products, including search. I think this will have a huge fiscal hit on Google as trust in their search engine is a huge competitive advantage. They have likely damaged trust with a large portion of them customers, and this damage will impact more than AI. Do you want to use a search engine that you believe is trying to manipulate you? You might if you have no other option but my guess is many people will jump ship to something else.
Finally, another issue I have seen with DEI training, and this probably is not related to the Google AI issue, is using the training to validate people concerns or buy them off. I saw this a lot as well...where the training was built to impress a bunch of people outside the organization (i.e., it was material they were receptive too) but in doing so they made the material objectionable to those being trained. Or the training is a way to funnel resources to a group that might otherwise publicly challenge or call out the organization. These trainings almost always inflict training scars as those being trained are generally offended by the content, and even if the content is decent, they tend to pick up on the real reason for the training and then get pissed about their time being used as part of deal to buy off others.
Here's the other problem; right now, people go to AI creators to give them correct information. No one ever asked Hamilton to be historically accurate, but if LLM's and AI image generators want to gain credibility, they have to start with depicting things as they actually are, not as the creators wanted them to be.
I love this article because you are so clear about how addressing inclusivity/diversity is critical to a high-functioning society (or company) yet is also something that we have been completely unable to fix in any consistent way.
I like to describe the present actions of companies as "look and feel" diversity. The objective (normally not explicitly so but in practice) is to ensure that the company looks diverse and people feel that there is respect for diversity. This leads to many, many shortcuts - eg diversity hiring quotas, etc...
This approach also completely sidesteps/ignores the underlying issues within all companies - hiring, promotion and recognition decisions are made entirely on a discretionary basis. Which means even in all-white companies, discrimination is common.
Typically, a manager will give preference to promoting people who they "click" with and who "get-things done". In a typical organisation, most people have the base ingredients to learn and do most jobs. This often meant people who are selected looked like the manager, went to similar universities, worked in similar companies, etc... It wasn't ever fair. Companies pushing for DE&I are now simply twisting the arms of those managers to preference people who "look" a certain way. This is likely why I'm tolerant of a certain amount of "anti-discrimination, discrimination". On some level, to me, it's the same thing from the past, just packaged differently.
If an organisation / society / etc really wants to address this, they have to first figure out a better way to make hiring, selection and recognition more transparent and fair such that the people involved from all backgrounds perceive that the process was fair (This is also something a lot of companies forget. They could have perfectly fair systems but if people in the company think they aren't fair or the results of those systems look 'bad', the systems will have no credibility).
In the past, the back-room talk was about how "they got the job because they golf with the boss". Now, it's that "they look a certain way". I don't think either is healthy but addressing it isn't a quick-fix, an offsite, and a leadership commitment to diversity.
I am not a believer that liberals want to get rid of white people, but it's really, really hard to refute that argument when a company has powerful as Google releases something like this. This. I mean, if you were trying to drive people to vote for Trump, you couldn't come up with anything better than this.
Racism is primarily about fear, mental models, and yes to a smaller degree about power. The fear response for human beings is tribalism. My sense is that all of this focus on race actually makes the process of removing racism harder. Rather, the shifts in identity (who is in my tribe) must happen gradually and naturally. A far more effective method is to have a common aspirational goals: Equality of opportunity, defeating a common enemy, aiming for a more perfect union..... the common cause in these sorts of things gradually shift the picture of internal identity.
On Google Gemini, I wouldn't take any of that too seriously.... science experiments can have unintended consequences.... I don't think there is a reason to be all that judgmental ... chuckle and move on. The reaction is more telling of the times vs the actual "offence."
On the economic research front, the data on "diversity" on boards/leadership is pretty sketchy. In my observation, breakout companies and new industries are almost always NOT diverse. Especially in the startup and early growth phase, the performance of the teams is aided by their common cultural identities. Now.. as the companies become large ... moving from racing boats to large tankers.. diversity in leadership helps in avoiding obstacles.
I actually prefer the overt forms of anti-racism like affirmative action. It does what it says on the tin. And it's easy and obvious to implement. If you say half of new hires or tv ads have to be people of color, then you have a clear cut way to measure success. And people can argue if that's too high or too low. But the other stuff of dealing with miccroagressions and hidden racism is just a scam. It's fostered this ginormous DEI industry full of people feeding from the trough of guilt. And there's no way it can be imposed from the top. And the cherry on the top is that 99% of the people it's meant to help don't think asking people to show up on time or do math was racist. So DEI people try to create problems where there wasn't any.
I'm gay, which has good and bad points vs say Black people. First, the huge majority of people don't care about gayness anymore. But I guess gays would be exposed to more offensive language and stuff than Blacks because of course it's less obvious. But what offends me (not very much, if anything) will differ dramatically from other people. And the way I deal with these "problems" when they do arise is also personal. I certainly won't think less of myself in any way, and I doubt anyone else is either. College kids usually referred to as snowflakes are just pretending. they aren't scarred from reading a book from the middle ages that condemns sodomy. Nobody thought there were gay pride parades in the 14th century, so stop faking hurt just for attention.
This is such a meandering and overlong dump of my feelings about DEI. So I'll get to the punch line. Teaching people via consultants accomplishes nothing other than to annoy everyone. You can't teach from above because it's way too general to be useful in any particular situation. And it's too remote to have any impact. Far better is to teach from the bottom up. The offended person can speak up, create a teaching moment. If I explain to someone why saying "man up" offends me (it doesn't but whatever) then that person is highly unlikely to say that again, and highly likely to understand things a bit better. And it's free!
In reading this commentary I am reminder of an organization in Poland : Forum for Dialogue http://dialog.org.pl/en/
Their mission is to promote communication between Poles today and the world wide Jewish community, especially Jews who once lived in Poland. I learned several things from this group: (1) you cannot change the past; (2) guilt is ineffective means of building trust/relationships between individuals and groups;. If you push guilt people will eventually push back and say hell no (3) regret is different than guilt. People from different groups can share guilt. These are all difficult tasks but I think Mr. Smith is on the right track here.
There is a great shortcut to building a “multiracial” society: marrying and having babies with someone of a different race than yourself. Until the rate at which people choose to do this increases by an order of magnitude, we will continue to be plagued by intense racial discord.