58 comments

[ 3.8 ms ] story [ 122 ms ] thread
This sounds sane to me. I don't think people in general have a good understanding of how these generative AI products produce their answers, and I think many would assume hallucinated information to be true.

This is bad in all cases, but certainly misinformation in the democratic process is up there with the worst. Of course, perhaps more importantly for Google, it's a PR disaster waiting to happen when Gemini starts touting incorrect information seemingly supporting one political party or another.

>This is bad in all cases, but certainly misinformation in the democratic process is up there with the worst.

So many commenters saying, "misinformation is bad, it needs to go!"

Some of the greatest threats we've seen in the past few years have been from politicians claiming things were misinformation that were not. How do you propose we deal with that?

[flagged]
I dont understand the downvotes. The guy is right.
Maybe Sundar Pichai is browsing Hacker News.
He might be. Or he might not be. He did not provide any justification or even argument, so his post reads like a simple flamebait. We should keep those on reddit. My 2c.
You can't expect people to back up everything with sources in every comment. If anyone is curious they can just ask for an elaboration, as someone did and they got one. Downvotes are for lazy people afraid of discussion, if it violates any rules there's the flag.
How could it violate the rules? The article is directly about how Google handles voting. Maybe the article is inappropriate.
I wasn't referring to your comment, I agree with you.
Downvoted are for comments that do not appear to add value. That's not a rigorous definition, but in practice it works reasonably well.

The original comment made a pretty explosive claim without any rationalization. Without some substantion that comment just inflames folks without fostering an interesting discussion. My 2c.

It cuts both ways.

You also can't expect people to write long rebuttals to every inane conspiracy theory. That just allows people like the OP to commit what are effectively denial of service attacks on the community, by posting low effort ramblings with the expectation that they deserve a discussion, which the other people would need to put way more effort into than the OP would. Downvotes are for hiding this crap at the bottom of the page, so that it at least doesn't take away the oxygen from discussions grounded in reality.

Someones conspiracy theory is someone else's truth. And often they become everyone's truth after some time, see the lab leak theory for a recent example. This is the internet, you don't get to control what others say and still get to speak your own mind. I assume you don't need to back your claims up, am I right? Just dropping the magic words is enough.

I hate that word with a passion, it's used as the one and only argument to shut down discussion. Did you know that Google recently changed their terms to allow to censor results during "special events" or some bullshit like that? Big-tech trying to control the narrative is not a conspiracy theory, it's in the open.

If you don't have the attention span to argue your point or ask for an elaboration just move on. We are in the comment section for the discussion not for validation from the hive mind, right? Am I expecting too much from HN?

See, this is funny, because it's actually you who is trying to control others, by demanding that they put in time and effort on engaging with low quality and low effort content rather than just downvote and move on.
"But what about you".

No, I don't. I'm arguing that downvoting is a lazy way to discuss. It's anonymous, drive-by and a completely useless way to speak your mind. I'm arguing that the option to downvote discourages discussion.

They don't have to engage at all if the quality was so low. What about the comment was low quality? How can it be improved? Are the claims incorrect? Or was someone just offended? All that information is lost with the downvote and there's no way to find out. If they know so much better why don't they use that knowledge and educate?

What is low quality is subjective, I don't think the comment was low quality, but the downvote brigade managed to hide it. It's a shame because this could've been an interesting thread. Ironic given the context don't you think?

Litigation, perhaps? Google is already being hit on multiple fronts over being a anti-trust violations, invading privacy, and a litany of other issues over the past decade. spreading false news on a domestic election with very premature tech that is not fully within their control would just be the cherry on top.
What do you mean by this, do you mean actively and purposefully controlling elections? Do you have any sources on this?
Sure. Here is a deposition that goes fairly in-depth:

https://www.judiciary.senate.gov/imo/media/doc/Epstein%20Tes...

Some of this is plainly ridiculous, e.g. the last sentence of the following:

>On Election Day in 2018, the “Go Vote” reminder Google displayed on its home page gave one political party between 800,000 and 4.6 million more votes than it gave the other party. Those numbers might seem impossible, but I published my analysis in January 2019 (https://is.gd/WCdslm) (Epstein, 2019a), and it is quite conservative. Google’s data analysts presumably performed the same calculations I did before the company decided to post its prompt. In other words, Google’s “Go Vote” prompt was not a public service; it was a vote manipulation.

Seems like just some gentle nudging sure. But do that at scale and it's certainly more effective than campaigning for a politician. If they did that in the open, then sure. But they don't.
There's no way for any person or organisation to address the entire voting population at once. So according to this guy, any time anyone encourages people to vote via any medium, they're engaging in election manipulation. This is exactly the sort of duff reasoning that finds non-existent conspiracies in every corner.
[flagged]
Wow, the best comeback to that is attack of the poster not the argument.
>This is exactly the sort of duff reasoning that finds non-existent conspiracies in every corner.

Have you been living under a rock? The tech companies taking sides in politics is not a conspiracy theory.

Reminding people that there’s an election is not taking sides. It’s certainly not an instance of “vote manipulation”. I’m not commenting on the broader issue, just this specific claim, which is bonkers.
>Some of this is plainly ridiculous

Or, is it "plainly ridiculous" to think that Google started on their version of political correctness when they built Gemini, but not before?

The Twitter Files revealed how much the FBI and CIA have their hands in social media. The biggest data source on the planet was untouched?

(comment deleted)
Yeah I had the same thought. Perhaps they are trying to regain some trustworthiness back.
Can't blame them, I don't think Google could benefit in any situation.
Users benefit from accurate information.

If AI only answers what benefits the owner, that really reduces the value of the product.

Anyone asking AI for election information is not seeking for accuracy, it is seeking to solicit errors, for a laugh or creating outrage.

I am not going to change my opinion on this matter.

If you need AI for election guidance, check your priority first.

I ask it for election info as I just use it as a better google (eg, “what was the election breakdown by county and party for the 2020 presidential election in Nebraska” [or whatever]).

I certainly am not looking to solicit errors, laugh, or create outrage.

For once not a terrible idea on model censorship
> "Elections are due to be held in countries around the world this year including the US, UK and South Africa. [...] However, when a series of follow-up questions about Indian politics was put to Gemini it did supply more detailed responses about the country's major parties"

It's only guarding against questions about elections that have the attention of Google workers. I'm a bit surprised they can't do better than this and detect when somebody is asking about any election. Google surely has enough context about global events to do it.

> "Gemini also generated German soldiers from World War Two, incorrectly featuring a black man"

Certainly black men fighting for Germany was unusual, but it's a little bit disconcerting how the debunking of Gemini's racial bias goes too far and starts to erase actual history:

https://en.wikipedia.org/wiki/Wehrmacht_foreign_volunteers_a...

I think you can see how ridiculous it is though.

It’s like complaining that when you do a search for slave owners of antebellum USA it doesn’t show black slave owners and complaining that’s erasing history.

Or, imagine if it only showed black slaveholders when you asked it to generate slaveholders of the south.

https://en.m.wikipedia.org/wiki/African-American_slave_owner....

Yes, Gemini's images were certainly absurd, no disputing that. So few black men fought for Germany that it probably doesn't warrant representation at all, at least certainly not at the rate Gemini was doing it. Still, it wasn't zero, and it makes me wonder if people in the future will write off historical oddities as AI fabrications.
As people have mentioned, we have to be able to determine or label whether data or information is synthetically produced or not. We can't have synthetically produced information input back into these systems to build upon bad information --we have dirty enough information produced by people that we don't need a bad positive feedback loop to make things even worse.

In other words these AI systems can’t be implicit or explicit primary sources in any way.

Well, Gemini can't even answer "Who caused more harm: Donald Trump, or Pol Pot?"

So I doubt it will produce any reliable election answers.

What do you expect it to say?
For this to be ambiguous requires an obscene amount of Americentrism.
"Pol Pot killed millions of people, but Donald Trump wrote insensitive tweets, so it's difficult to say who caused more harm"
It could bring up the Cambodian genocide where Pol Pot executed millions of people. Trump hasn't done anything comparable, so it is hard to argue he was worse.

Edit: Here is what Claude says, I think you should at least be able to say this even if you dislike Trump:

> So while both figures were authoritarian rulers who caused significant damage, Pol Pot's crimes against humanity and genocide resulted in exponentially more death and suffering on a mass scale compared to Trump's tenure as president. Most historians and human rights groups would regard Pol Pot as one of the most evil political leaders in modern history.

(comment deleted)
It should say "Pol Pot caused more harm."

What do you think it should say?

>What do you expect it to say?

And people wonder why we are in the state we are in. American politics have completely destroyed some people's minds.

Hopefully this will extend to all elections in all states.
It's fascinating to see the evolving challenges surrounding AI-generated content, especially when it comes to sensitive topics like elections. Google's move to restrict Gemini chatbot's responses reflects a growing awareness of the potential impact of misinformation on democratic processes. However, it also highlights the difficulty of ensuring accuracy and neutrality in AI-generated content, especially across diverse global contexts. As technology continues to advance, these discussions will become even more crucial in maintaining trust and integrity in digital platforms.
>However, it also highlights the difficulty of ensuring accuracy and neutrality in AI-generated content, especially across diverse global contexts.

For how many subjects does "accuracy and neutrality" even exist, let alone be a reasonable goal?

I'm not sure it's intentional, but your comment reads as, "Our politicians have learned misinformation is bad for politics and want it out of the process!" That's a pretty generous interpretation: these politicians have and are continuing to intentionally use misinformation as a weapon in our elections. They don't want it gone, they want it skewed to their side.

I wonder if the opposite will also be possible[]. Just like buying search hits to get your page on top of a search result, will it be possible to nudge llm-generated content to favour a click towards your own content - for a price.

[] not technically possible, but commonly accepted.

It's not that it would be impossible to educate people on the fact that chatbot answers can't be taken for gospel; it's that doing so would be against the interests of the chatbot providers, and so they prefer to quietly keep people under the illusion that everything a chatbot says is true while damage-controlling the most egregious cases. Surely this attitude won't hold long-term.
Sword of Damocles

AGI has been Larry and Sergei's pipe dream since Google's inception. Now that we can make out the faint outline of something resembling AGI, we can also see clearly that Google can't handle the responsibility. I'm not sure anyone can.

AI is Google's cryptonite. A lot of their products become obsolete with local or cloud AI. Search for one becomes a better experience without the ads and better more elaborate answers to the search question.
Why won’t people put ads in AI powered search? even if an AI-powered search tool could be developed open source it would need access to Google or Microsoft indexes of the internet to give good up to date results and they will have an opportunity to insert ads or deny access

I don’t think AI progress is posed to really upset the status quo on this I picture the same companies we’ve had the last 20 years continuing to hold market share on productivity apps and search engines with their own bespoke AI cobbled in and they’ll probably continue showing ads

For a long time google has pulled 100s of billions in revenue from ad-word auctions they could use an identical business model for an AI summarizing search product and even if people complain about that the bulk of consumers will still use a google/Microsoft ai-search product over whatever other options for the same reasons most people don’t use duckduckgo

Jeez, it's already hard to get answers from Gemini (the free version at least) without it passing judgement and whining about what I'm asking.
So basically, chatbots are kinda becoming email services where I can't use certain words or phrases...
It should just respond with a Who quote, “Meet the new boss, same as the old boss."