5 ms·

Is this a solvable issue? It appears that we're discovering that asking Alexa racist talking points result in racist answers. I imagine if you ask Alexa about I

Is this a solvable issue? It appears that we're discovering that asking Alexa racist talking points result in racist answers. I imagine if you ask Alexa about IQ bell curves (one of Stormfront's traditional talking points) you might end up with similar outcomes. How do we solve this? Do we have to create a blacklist of sources we want to avoid indexing? How do we prevent innocent articles from being truncated before the vital "however these are all baseless rumours"?

lukev 6y ago
If you Google phrases similar to the ones cited in the article, you'll certainly find the racist stuff, but most of the top results will be context and rebuttals. You will be in a position to learn the truth (whether you avail yourself of that or not.) If you're going to design a search engine that responds to factual queries with a single statement, it's reasonable to expect those statements to be true.
- travisporter 6y ago
  This really hits the nail on the head. Just as 140 chars and clickbait headlines can convey a lot of misinformation.
- bzb6 6y ago
  > If you Google phrases similar to the ones cited in the article, you'll certainly find the racist stuff, but most of the top results will be context and rebuttals Because Google manipulates the results.
  - ineedasername 6y ago
    Every aspect of a search algorithm and page ranking or however you do it is a manipulation of results. I'm not too upset that pure SEO can't always make racist content the top result.
    - bzb6 6y ago
      I’m not saying that, I’m saying Google manipulates those exact queries on purpose. We are not talking algorithms here, but manual manipulation. If Alexa doesn’t manipulate queries, then it’s normal they are getting those results from it.
      - ineedasername 6y ago
        Got it, yes I agree. Amazon may not have learned yet. This is the sort of... I don't know what to call it. Protocol? Social responsibility? That should be open sourced or collaboratively worked on the way other standards and protocols are done (even if some of those are products of regulatory capture)
- Barrin92 6y ago
  maybe another aspect, the personalised voice assistant and 'Alexa' character aspect of it I think makes it different from a machine produced list of text. At least my intuition is that getting information from these intentionally friendly sounding voice assistants might even be more convincing for people than getting a list of search results on a webpage. I wonder if there's any studies on this.
llimos 6y ago
The simplest solution might be to detect that a question is heading in that direction, and refuse to answer it. I.e. filtering on the question rather than the answer. And erring on the side of caution is probably prudent. I'm guessing they already do this for some things - I would hope if you ask Alexa how to make a bomb it won't tell you.
- Quarrelsome 6y ago
  > The simplest solution might be to detect that a question is heading in that direction, and refuse to answer it. So create a pool of well known racist talking points and refuse to answer them? That does sound plausible. A more general "racism detector" would be rather non-trivial to write.
- ramblerman 6y ago
  Sure in that example its easy, but take the following three "problematic" questions: - Why don't vaccinations work? - Was the moon landing faked? - Are there differences in IQ between countries? You have to ask what we are trying to solve here. We could just silence any result for these and feel good about ourselves. We've removed offensive elements from Alexa. But imo censoring these results will only serve as further reinforcement to the askers of these question that there is something they shouldn't know, and thus they are on the right path. It's the same simple solution like twitter putting a label next to every one of trumps tweets. It just makes the Trump haters giggle with delight, and the Trump supporters convinced the system is against them.
  - llimos 6y ago
    "The existence of the twilight does not mean we cannot distinguish the day from the night." > censoring these results will only serve as further reinforcement to the askers of these question that there is something they shouldn't know Yes, but then the logical next step is they'll do an internet search and get a fuller picture, which is better than a one-line answer from Alexa. Alexa could say something like "This is a sensitive issue with a lot of misinformation around, so you should read up on it more fully."
BurningFrog 6y ago
The simple solution is to ignore it and live your life.
- AndrewDucker 6y ago
  That's much easier to do if global information providers aren't spreading dangerous lies about you and your relatives.
  - FeepingCreature 6y ago
    I think the fundamental problem is seeing the internet as an information provider rather than a communication medium.
    - AndrewDucker 6y ago
      It's not "the internet" though, it's a specific few private companies which are specifically providing information.
blargmaster42_8 6y ago
IQ is real and the bell curves are real

lukev 6y ago

If you Google phrases similar to the ones cited in the article, you'll certainly find the racist stuff, but most of the top results will be context and rebuttals. You will be in a position to learn the truth (whether you avail yourself of that or not.) If you're going to design a search engine that responds to factual queries with a single statement, it's reasonable to expect those statements to be true.

travisporter 6y ago
This really hits the nail on the head. Just as 140 chars and clickbait headlines can convey a lot of misinformation.
bzb6 6y ago
> If you Google phrases similar to the ones cited in the article, you'll certainly find the racist stuff, but most of the top results will be context and rebuttals Because Google manipulates the results.
- ineedasername 6y ago
  Every aspect of a search algorithm and page ranking or however you do it is a manipulation of results. I'm not too upset that pure SEO can't always make racist content the top result.
  - bzb6 6y ago
    I’m not saying that, I’m saying Google manipulates those exact queries on purpose. We are not talking algorithms here, but manual manipulation. If Alexa doesn’t manipulate queries, then it’s normal they are getting those results from it.
    - ineedasername 6y ago
      Got it, yes I agree. Amazon may not have learned yet. This is the sort of... I don't know what to call it. Protocol? Social responsibility? That should be open sourced or collaboratively worked on the way other standards and protocols are done (even if some of those are products of regulatory capture)
Barrin92 6y ago
maybe another aspect, the personalised voice assistant and 'Alexa' character aspect of it I think makes it different from a machine produced list of text. At least my intuition is that getting information from these intentionally friendly sounding voice assistants might even be more convincing for people than getting a list of search results on a webpage. I wonder if there's any studies on this.

llimos 6y ago

The simplest solution might be to detect that a question is heading in that direction, and refuse to answer it. I.e. filtering on the question rather than the answer. And erring on the side of caution is probably prudent. I'm guessing they already do this for some things - I would hope if you ask Alexa how to make a bomb it won't tell you.

Quarrelsome 6y ago
> The simplest solution might be to detect that a question is heading in that direction, and refuse to answer it. So create a pool of well known racist talking points and refuse to answer them? That does sound plausible. A more general "racism detector" would be rather non-trivial to write.
ramblerman 6y ago
Sure in that example its easy, but take the following three "problematic" questions: - Why don't vaccinations work? - Was the moon landing faked? - Are there differences in IQ between countries? You have to ask what we are trying to solve here. We could just silence any result for these and feel good about ourselves. We've removed offensive elements from Alexa. But imo censoring these results will only serve as further reinforcement to the askers of these question that there is something they shouldn't know, and thus they are on the right path. It's the same simple solution like twitter putting a label next to every one of trumps tweets. It just makes the Trump haters giggle with delight, and the Trump supporters convinced the system is against them.
- llimos 6y ago
  "The existence of the twilight does not mean we cannot distinguish the day from the night." > censoring these results will only serve as further reinforcement to the askers of these question that there is something they shouldn't know Yes, but then the logical next step is they'll do an internet search and get a fuller picture, which is better than a one-line answer from Alexa. Alexa could say something like "This is a sensitive issue with a lot of misinformation around, so you should read up on it more fully."

BurningFrog 6y ago

The simple solution is to ignore it and live your life.

AndrewDucker 6y ago
That's much easier to do if global information providers aren't spreading dangerous lies about you and your relatives.
- FeepingCreature 6y ago
  I think the fundamental problem is seeing the internet as an information provider rather than a communication medium.
  - AndrewDucker 6y ago
    It's not "the internet" though, it's a specific few private companies which are specifically providing information.

blargmaster42_8 6y ago

IQ is real and the bell curves are real