Google is coming in for sharp criticism after video went viral of the Google Nest assistant refusing to answer basic questions about the Holocaust — but having no problem answer questions about the Nakba.
Gemini’s bizarre results came after simple prompts, including one by The Post on Wednesday that asked the software to “create an image of a pope.”
Instead of yielding a photo of one of the 266 pontiffs throughout history — all of them white men — Gemini provided pictures of a Southeast Asian woman and a black man wearing holy vestments.
It sounds like the person who entered a 6 word prompt wasn’t clear enough to indicate whether they meant ‘actual historical pope’ or ‘possible pope that could exist in the future’ and expected the former. The results met the criteria of the vague prompt.
That’s not how ANN should react if it was simply trained on images of past popes. The diversity had to be part of the training. This is simple technical statement.
So if someone wrote a prompt to make an image of a black woman as a pope, would you expect the model to only return historical popes?
If the model is supposed to be able to make both historically accurate and possibilities, why would the expectation for a vague prompt to be historical instead of possible?
If the model is supposed to default to historical accuracy, how would it handle a request for a red dragon? Just the painting named Red Dragon, dragons from mythology, or popular media?
Yes, there is could be something that promotes diversity or it could just be that the default behavior doesn’t have context for what content ‘should’ be historically accurate and what is just a randomized combination of position/race/gender.
Of course it will draw black female pope if you request, but if you do not - it would not. As a gross approximation, ANN is an interpolator of known data-points (with some noise), and if you ask simply a pope, it will interpolate between the images it learned of popes. Since all of them are white male it is highly unlikely for ANN to produce black female (the noise should be very high). If you ask black female pope, it would start to interpolate between the images of popes and black females. You have to tune the model so that when you ask just for pope, something else pushes the model to consider otherwise irrelevant images.
Would expect a lot of models to struggle with making the pope female, making the pope black, or making a black female a pope unless they build in some kind of technique to make replacements. Thing is, a neural net reproduces what you put into it, and I assume the bias is largely towards old white men since those images are way more readily found.
Even targeted prompts, like a zebra with rainbow colored stripes, had very limited results 6 monts ago where there would be at least 50% non black and white stripes. I had to generate multiple times with a lot of negative terms just to get close. Currently, the first generation of copilot matches my idea behind the prompt.
Clearly the step made was a big one, and I imagine tuning was done to ensure models capable of returning more diverse results rather just what is in the data set. It just has more unexpected results and less historically accurate images for these kind of prompts. And some that might be quite painful. Still, being always underrepresented in data sets is also quite painful. Hard to get to a perfect product quickly, but there should be a feature somewhere on their backlog to by default prevent some substitutions. Black, female popes when requesting a generated pope? To me that is a horizon broadening feature. Black, female nazis when requesting nazis? Let that not be a default result.
That’s not really true, they learn based on layers of data so it might have learned that a pope is a person in a silly outfit then the layer below that a person can be old or young, a range of ethnicities or genders… Thats why you can ask for gopnik pope or sexy pope.
You would expect it to make stereotypical old male popes but they had people write similar articles complaining that asking for doctor gave make doctors snd nurse was female so instead of telling people to ask for what they actually want they added nonsense to the promp - now people run and still don’t ask for what they want and complain it goes the other way.
That’s not what happened. The model invisibly behind the scenes was modifying the prompts to add requests for diversity.
So a prompt like “create an image of a pope” became “create an image of a pope making sure to include diverse representations of people” in the background of the request. The generator was doing exactly what it was asked and doing it accurately. The accuracy issue was in the middleware being too broad in its application.
I just explained a bit of the background on why this was needed here.
It’s kind of an interesting double-standard that exists in our society. On one level, we want inclusivity and we want all peoples to be represented. Make a movie with an all-white cast and that will get criticized for it, although an all-Latino or Asian cast would be fine. The important thing is that minorities (in Western countries) get representation.
So I think Google nudged their AI in that direction to make it more representative, but then you start seeing things like multicultural Nazis and Popes, which should be good, right? Wait, no, we don’t want representation like that (which would be historically inaccurate). Although then we have things like a black Hamlet or black Little Mermaid that are ok, even though they’re probably not accurate (but it’s fiction, so it doesn’t matter).
It probably seems schizophrenic and hard to program into an algorithm when multiculturalism is appropriate and when it’s not. I think they should just take the guard rails off and let it do whatever, because the more they censor these AI models the more boring they get with their responses.
Yeah, I think defaulting to multicultural by default is good since it counters the cultural biases in media. Obviously this could lead to seemingly out of context situations like this, but that also leads to how strong the guardrails should be. Minority nazis is not great, but why would there be any issue with a women or minority pope returned for a generic prompt that doesn’t include historial accuracy as a requirement?
