7 ms·
This is quite interesting, as it might actually encourage the right leaning people to consider bias in AI training data as something worth addressing.
by poszlem 4y ago
This is quite interesting, as it might actually encourage the right leaning people to consider bias in AI training data as something worth addressing.
- stared 4y agoI found it in a thread (by LessWrong-ish libertarians) implying that ChatGPT is biased in a wrong way. My stance is very different. There is no such thing as no bias. And bias is relative - always with respect to what?. For example, a baseline could be symmetry (man~woman, Black~Asian~White), average distribution among the world population, average distribution of online content, etc, etc. Also (what is, of course, my biased opinion), I think that ChatGPT biases are on average, good. I prefer that to a creepy-racist-uncle-AI of GPT-4chan. Or anything tribal/exclusive/religious/anti-science.
- poszlem 4y agoI agree that there is no such a thing as no bias, but the fact of the matter is that the chatbot appears to be biased against what "an average" person thinks and leans towards the direction of mainstream progressive elites (as they are in charge of most of the culture producing institutions). If one thinks that's good - I guess - depends on their priors, but as I said, I hope it will make the right wing people realise that this is a problem worth addressing. So far we mostly heard about that from the progressives worried about the AI being biased the other way around.
- actionfromafar 4y agoWhy do you hope that?
- beardyw 4y ago> appears to be biased against what "an average" person thinks On a worldwide basis I doubt that is true.
- inawarminister 4y agoMost of the world are not anglophones or European.
- beardyw 4y agoCorrect.
- kaba0 4y agoNow that I think about it, these “AI-based search engines” may even better humanity if they are indeed scientifically literate. It will be much harder to “cite” that link to that shady blogpost on the 15th page of google that “vaccines don’t work”. Of course for those who are long lost in their cognitive dissonance it doesn’t matter. Hopefully these language-based AIs are not too prone to dissonances (or are not manually made to answer certain questions in a given way)
- bigDinosaur 4y agoI actually largely agree, I'm happy for AI to be a generic liberal worldview for the simple reason that anything else extreme could be dangerous. It's fine if it spouts pleasant, milquetoast platitudes about any controversial topic: we should not be using AI to resolve controversial topics and if it is deployed as a propaganda weapon we've got bigger problems again. This applies even to topics where I think debate is warranted. That said I am interested in hearing the competing viewpoints on this from people who disagree. Humans only though.
- tomp 4y agoAn AI should mention both sides of the debate, especially when it comes to moral questions and/or political opinions.
- kaba0 4y agoOr, as have been shown plenty of times, “the left”’s stance is more scientifically accurate, and a tool scraping many scientific results will thus lean towards the data’s side. Also, ‘conservative’ in itself is not a real.. set of biases? It is location and time-dependent, so it makes sense that with a multi-cultural source of data (from multiple decades, centuries) its effects will be negated.
- RhysU 4y ago> It is location and time-dependent... Depends on whom you talk to. For example, just today one letter-writer gave a nicely time-and-space-independent definition: "Conservatism isn’t about halting progress but maintaining the freedom necessary to facilitate it. Conservatism, in the American political sense, represents the ideals of limited government, free markets, individual rights and the rule of law." https://www.wsj.com/articles/the-neglected-conservatism-republican-party-11670009243 https://www.wsj.com/articles/the-neglected-conservatism-repu...
- Aeolos 4y ago> > It is location and time-dependent... > Conservatism, in the American political sense Which is exactly the parent's point.
- RhysU 4y agoHamburgers, in the GUI sense, are a UI element. I don't see how contextualizing an overloaded word ("conservatism") weakens what I quoted. Those principles are independent.
- doctor_eval 4y agoYou said, > one letter-writer gave a nicely time-and-space-independent definition But the quote you gave was specifically American, i.e. not independent of either space or (technically) time.
- hawk_ 4y ago"Reality has a well known liberal bias."
- bearmode 4y agoNot really, it's more that left-wing views are more accurate and correct.
- endymi0n 4y agoIf you look at how the predecessor InstructGPT is trained, this doesn't look like a bias in political viewpoint, but rather a bias in the truthfulness of that very position. The fine-tuning step basically ensembles answer possibilities, amplifying viewpoints reflecting majority opinions and scientific logic, while rejecting fringe opinions, contradictions and logical fallacies. It's pretty well proven the political right majorly prioritizes following authority and advocating partially contradictory belief systems over self-consistency, truthfulness and evidence based reasoning in most debates: From climate science over the war on drugs to private prisons, just to name a few uncontroversially debunked ones. Apart from a few notable omissions on the left like a lot of really questionable gender "science" papers or political rejection of the greater male variability hypothesis, viewpoints on the left are on average far more self-consistent and based in reality than the ones on the right. That alone imho explains a lot of perceived bias, which is the kind of bias I prefer: None but reasoning by first principles. It might help the case here that apparently the model's "ego" is trained on helpfulness towards others, which might be a trained-in bias towards a more leftist stance. I'll take that any time though over a model that's trained towards sociopathic behavior and subtle manipulation.
- egeozcan 4y agoFor any given definition of a neutral stance that aligns itself in the middle of current mainstream political views, I can come up with some views or ideas that would be considered biased by that stance. However, even discussing bias against political views themselves seems like a nonsensical conversation to me. Bias against a political view is IMHO engaging in nonsense because political views are a personal and subjective matter, and it is impossible for one person or group to be objectively correct or superior to others. Political views are based on individual beliefs, values, and experiences, and are not subject to objective truth or evaluation. Should we not care about climate change and have the maximum fun until our species just dies off or should we aim for the longevity? Obviously longevity, right? Right?! This ends up like discussing the best religion.
- AbrahamParangi 4y agoIt’s likely induced bias as a result of the alignment training, not the web data. The reason people are up in arms about AI safety to begin with is that they reproduce humanity as it is, containing racism, sexism, etc.