Rethinking Search: Making Experts out of Dilettantes
arxiv.org
arxiv.org
Nowadays, if you have a question, you may get a bunch of competing opinions and often a Wikipedia article. The Wikipedia article, if it's not a topic with lots of eyes and some consensus, may not be particularly reliable. (Of course many people don't think of this.) I think this multitude of voices is closer to how the world and knowledge actually works. There is effort and interpretation involved.
On the other had, a question answering system is more like a short paper encyclopaedia article, except probably is less reliable (no detailed expert supervision) and appears more authoritative (Google is more ingrained into most peoples' lives than encyclopaedias ever were). This may appeal to people that would like internet to always give people the one "correct" answer, but not to me.
For anything more subjective, I'd rather have the organic web, different points of view and potentially different conclusions. Serving up the best results is hard and moving the goalposts of what the actual goal is isn't beneficial.
Also questionable about crawling the entire web's information and using it in such a way that doesn't benefit the content creators. 2021 search already offers significantly less click-throughs to the content creators than 10 years ago.
This paper suggests the ‘improvement’ “webmd [1] says it may stain your teeth”.
That’s not their buggy first implementation, that’s the goal.
I feel bad.
Aren't they already getting sued and looking at regulation in EU because of this strategy?
If nothing else, have they ever used WebMD?
WebMD will say it can cause nausea, diarrhea, headache, and possibly death, because they say that about basically everything.
It's impressive research work. However, we can expect implemetation on Google will, intentionally or not, lead to more of keeping us on their property.
Google Search will become even less so a way of navigating and discovering on the open web. Is their dream, and was it always about AI? One Brain to Rule them All.
They are not even original with the idea for this type of paper.
In September last year Aran Komatsuzaki posted: "Current Limitations of Language Models: What You Need is Retrieval" https://arxiv.org/abs/2009.06857
I have also a prediction market from August last year with this question:
Will a language model with at least 100B parameters trained to do external information retrieval exist before 2023?
Seems like people gave it 61% in March this year. I was at 75%.
https://www.metaculus.com/questions/4876/will-a-language-mod...
These are often called "position papers". There's no reason why they shouldn't be written, and published.
. Only curated websites (period) and databases.
. Only the explicit search terms I give.
. Ability to block certain explicit search terms.
. Perhaps a bit of misspelling support.
Google search's uselessness (is that a word?) increases day by day as it evolves increasingly towards marketing, 'things it thinks I mean', defaulting to wikipedia.
And here I'm complaining when they stop doing that. Try a search for coronavirus and Google has hardcoded "official" results about COVID-19 (which is just one of many coronaviruses).
coronavirus site:wikipedia.com site:cdc.gov
well now, that doesn't work at all. So much for thinking you could send a thousand websites along with the search.
I wonder sometimes, and this would probably only suit me, if I could get a copy of the websites I'm actually interested in and simply have a local copy. Do local organization and search (is that FOSS or buyable?). Maybe with occasional updates.
A person could go a long way with a stack of disk drives holding libgen, sci-hub, gutenberg, magazine archives from archive.org, special interest forums, etc. Plus, it would keep you away from the less useful parts of the internet.
One thing that does make you appreciate Google is how awful Amazon search is.
coronavirus (site:wikipedia.com OR site:cdc.gov)
https://www.cambridge.org/core/journals/renaissance-quarterl...
I remembered this because I cited this essay in an undergraduate history paper I wrote. The author was my professor — which didn't hurt ;-)
https://www.gov.uk/government/consultations/online-harms-whi...
Here's the problem:
Let's say you search for "abstract art" on Etsy: [1]
This results in about 1.7 million hits.
Browsing through even just a few pages of results will make it clear that there's an enormous variety of art that passes for "abstract art" on Etsy.
I've browsed through dozens of pages of results and concluded that most of the art is (to my taste) not very good, some small amount is decent, and a tiny percentage is pretty good. But, given the sheer enormous amount of art on there, I suspect there's probably some abstract art by artists I'd never heard of on Etsy that's I'd consider to be amazing.
How do I find such art?
Browsing through all 1.7 million results would take me forever (assuming I had the patience for it and Etsy even let me), so it'd be nice if there was a faster way, but I can't think of one.
Obviously, searching for "amazing abstract art" would be absolutely useless, because everyone has different tastes, so even were the results not manipulated by sellers deliberately labeling their art as "amazing", just because someone thought a particular work of art was amazing doesn't mean I would.
Of course, you could search for various properties of the art like its color. The odds of liking a work of art might increases if it's in your favorite color(s). So that's kind of helpful, but, honestly, if the art really is amazing it could be in just about any color and still be great. Same goes for just about any other property the art might have (like size, media, etc...).
There's also the important element of surprise. Viewers aren't necessarily going to know what they like until they see it. Who before Picasso painted Guernica or before Van Gogh painted his sunflowers would have known to search for them? Often, but especially for abstract art, viewers don't know what they'll like until they see it. So even a strong AI might not help if there's nothing to ask it apart from "show me some abstract art that I'd like", the definition of which would differ for everyone.
While writing this post, I did actually come up with some ideas.
The first idea is recommendations by curators or friends with similar taste.
The second idea is giving the search service a sample of other abstract art you liked. Using image similarity this could show you other images that had a pretty good chance of being the sort of thing you like. However, this would not be helpful in finding truly novel art that you really liked for which I still don't think there's any way to search other than by simply browsing through lots, and lots, and lots of art.
You made examples of etsy being too loose with the concept and your concept of it being tighter than that.
Who is to say? I'd suppose no one, or thing.
Especially if information discovery is a proponent of the evolution of abstract art. One point of view leads you down a blind alley.
I do like your idea of crowdsourcing opinion though. That's sort of similar to having different search engines providing different answers, though having the algorithm more tightly focused on people/ideas/directions you choose as authorities.