zlacker

[parent] [thread] 58 comments
1. strike+(OP)[view] [source] 2019-06-14 18:23:48
"If YouTube won’t remove the algorithm, it must, at the very least, make significant changes, and have greater human involvement in the recommendation process.", man does this person know how many videos and how many users YouTube has? They cannot use anything except an algorithm to recommend videos. They cannot use anything except an algorithm to detect videos inappropriate for children. It seems YouTube is working on this, and this opinion seems like a ill thought out fluff piece to enrage readers and sell this persons book.
replies(11): >>andrew+31 >>mtgx+j3 >>scarfa+E3 >>bluebo+V5 >>moomin+nc >>gibrow+oc >>jgalt2+9g >>kartan+vh >>python+3w >>robbro+2J >>razius+331
2. andrew+31[view] [source] 2019-06-14 18:30:58
>>strike+(OP)
Maybe they can't make editorial recommendations for the long tail but they absokutely could do so for the top few thousand videos each week.

Would that yield an improvement? I don't know, but it would have an impact.

replies(1): >>scj+Q6
3. mtgx+j3[view] [source] 2019-06-14 18:46:02
>>strike+(OP)
All true. But all of this is making me wonder - what are the people thinking who say they can't wait for our society to be run by AI? The apex of AI capability can't even recommend videos properly right now, and we want it to run all the aspects of our society?! No, thanks.
replies(3): >>icebra+S71 >>jodrel+k81 >>v7p1Qb+Ae7
4. scarfa+E3[view] [source] 2019-06-14 18:48:23
>>strike+(OP)
Just to add on, a Youtube executive was recently on a podcast and she said there are 500 videos uploaded per second.
replies(2): >>Rugnir+I8 >>v7p1Qb+vd7
5. bluebo+V5[view] [source] 2019-06-14 19:01:33
>>strike+(OP)
Such an effort would cost literally millions of dollars and surely sink this fledgling startup
replies(1): >>ggggte+k6
◧◩
6. ggggte+k6[view] [source] [discussion] 2019-06-14 19:04:49
>>bluebo+V5
I don't think sarcasm with no substance behind it is very insightful.

Humans are involved in the process. To suggest otherwise is to be willfully ignorant.

◧◩
7. scj+Q6[view] [source] [discussion] 2019-06-14 19:08:04
>>andrew+31
I'm kind of wondering if a "Ned Flanders" user-detector is possible.

Search for users who stop videos at "offensive" moments, then evaluate their habits. It wouldn't be foolproof, but the "Flanders rating" of a video might be a starting metric.

Before putting something on YouTube for kids, run it by Flanders users first. If Flanders users en masse watch it the whole way through, it's probably safe. If they stop it at random points, it may be safe (this is where manual filtering might be desirable, even if it is just to evaluate Flanders Users rather than the video). But if they stop videos at about the same time, that should be treated as a red flag.

Of course, people have contextual viewing habits that aren't captured (I hope). Most relevantly, they probably watch different things depending on who is in the room. This is likely the highest vector for false positives.

The big negative is showing people content they obviously don't want for the sake of collecting imperfect data.

replies(3): >>Mirior+Kb >>gus_ma+sc >>Nasrud+dg
◧◩
8. Rugnir+I8[view] [source] [discussion] 2019-06-14 19:22:43
>>scarfa+E3
thats.... actually shockingly few
replies(2): >>gowld+79 >>nostra+8a
◧◩◪
9. gowld+79[view] [source] [discussion] 2019-06-14 19:25:06
>>Rugnir+I8
A billion videos per year is shockingly view?
◧◩◪
10. nostra+8a[view] [source] [discussion] 2019-06-14 19:31:45
>>Rugnir+I8
The stat I heard while at Google (~5 years ago) was that 8 hours of video is uploaded every second. Cross-checking that against the 500 videos/sec figure, it implies that the average video is about 1 minute. I suspect the 8 hours figure is pretty out-of-date now, and it's more like 20 hours/sec.

BTW, you could do some simple math to figure out how many employees it'd take to have a human watch every video that comes in. 3600 secs/hour * 20 hours of video/sec = 72000 secs/video/sec, * 3 to assume 8 hour shifts = 216,000 employees, * $30K/year = $6.4B/year. It's theoretically doable, but you wouldn't get the product for free anymore.

replies(1): >>sereto+dc
◧◩◪
11. Mirior+Kb[view] [source] [discussion] 2019-06-14 19:43:15
>>scj+Q6
The problem is that just a few examples of the algorithm getting it wrong is enough to cause an adpocalypse. If millions of videos are uploaded every month then you can imagine how low the error rate has to be.
replies(1): >>scj+jg
◧◩◪◨
12. sereto+dc[view] [source] [discussion] 2019-06-14 19:47:18
>>nostra+8a
$30k/year seems high. This is the sort of work that would be likely outsourced, perhaps to the Philippines for less than $10k/year per person.

$2B is still nothing to sneeze at, but it's less than Microsoft paid for Minecraft.

replies(3): >>nostra+Fe >>scarfa+Sm >>icelan+b21
13. moomin+nc[view] [source] 2019-06-14 19:48:58
>>strike+(OP)
You are 100% not thinking big enough. These algorithms identify clusters. These clusters can be examined through random sampling. It doesn’t take a genius to spot that a cluster that involves children and pornography might have some problems.

Of course, the system doesn’t expose these kinds of outputs, because no-one has any interest in designing such a system and taking responsibility for the content.

14. gibrow+oc[view] [source] 2019-06-14 19:49:00
>>strike+(OP)
The total number of videos really doesn't matter, it is the total number of creators, which at least this site claims is a total of 50m for all time: https://mediakix.com/blog/youtuber-statistics-content-creato... (first result I found)

Just start banning certain creators from showing up in recommendations if their content crosses the line. Not that hard if you are willing to do it.

replies(1): >>cortes+qf
◧◩◪
15. gus_ma+sc[view] [source] [discussion] 2019-06-14 19:49:14
>>scj+Q6
Should we filter all the pro-choice videos or the pro-life videos?

Should we filter all the Santa-is-fake videos or the Santa-is-real videos?

Do you agree with Flanders?

replies(2): >>scj+Df >>unders+kN
◧◩◪◨⬒
16. nostra+Fe[view] [source] [discussion] 2019-06-14 20:10:29
>>sereto+dc
$30K/year is minimum wage in Sunnyvale and Mountain View, where Google headquarters is.

YouTube could probably outsource it internationally, but that'd just spark a new round of outrage: "Why are global community standards set by an American technology company outsourced to poor workers in the Philippines? Are these the people we want deciding our values?"

replies(3): >>taxidu+5j >>mc32+eI >>deanCo+iV
◧◩
17. cortes+qf[view] [source] [discussion] 2019-06-14 20:14:46
>>gibrow+oc
But how would that solve the problem that the article opened with? There is nothing wrong with the videos of children playing, the wrong part was recommending them to pedophiles
replies(2): >>gibrow+ei >>unders+oN
◧◩◪◨
18. scj+Df[view] [source] [discussion] 2019-06-14 20:17:08
>>gus_ma+sc
Why would Youtube for Kids show anything on the topic of abortion?
replies(1): >>gus_ma+Io
19. jgalt2+9g[view] [source] 2019-06-14 20:21:56
>>strike+(OP)
> man does this person know how many videos and how many users YouTube has

While that might be true, 99% of the views are a very small subset of the videos posted. It's completely doable, or at the very least the problem can be greatly mitigated by putting more humans into the process and not letting the algos recommend videos that haven't been viewed by someone in Youtube's equivalent of "standards and practices". All that being said, I fear the primary reason this is not done is because such actions would reduce the number of hours of viewed videos and ad revenues. In fact, I've read articles supporting this theory.

Google under Pichai is basically like Exxon under Lee Raymond--solely focused on revenue growth and completely blind to any number that doesn't show up on the current and next quarter's income statement.

replies(1): >>sharce+bP
◧◩◪
20. Nasrud+dg[view] [source] [discussion] 2019-06-14 20:22:53
>>scj+Q6
The question I have is how can they tell "Flanders" viewers from "bored" ones or "out of time" ones short of them flagging it without a lot of manual review and guess work?

Reviewing viewers on that level sounds even more intensive than filtering every channel and video.

replies(1): >>scj+Tk
◧◩◪◨
21. scj+jg[view] [source] [discussion] 2019-06-14 20:23:25
>>Mirior+Kb
If Google takes the impractical route and hires a sufficient number of multilingual Ned Flanders, then they're still probably going to have a non-zero false positive rate (humans make mistakes too).

Whatever they do is going to have to be evaluated in terms of best effort / sincerity.

Semi-related: The fun of Youtube is when the recommendation algo gets it right and shows you something great you wouldn't have searched for. The value is that it can detect elements that would be near impossible for a human to specify. But that means it has to take risks.

22. kartan+vh[view] [source] 2019-06-14 20:32:31
>>strike+(OP)
> They cannot use anything except an algorithm to recommend videos.

I agree that with the current business model it is not possible for YouTube to sort it manually.

When I was a kid, a long long time ago, it would have been impossible to conceive that a TV channel showed that kind of content regularly and continue open. If their answer would have been that they cannot fix it because it costs money there would have been an outraged response.

If YouTube cannot keep things legal, cannot respect people rights, cannot be a good responsible part of society because it is not cost effective for me the way to go is clear. And that is true for YouTube, Facebook or any other business digital or not.

replies(3): >>d1zzy+nv >>seanmc+tK >>Camper+YM
◧◩◪
23. gibrow+ei[view] [source] [discussion] 2019-06-14 20:38:25
>>cortes+qf
Feels like the article was about more than that one issue. It also discussed creators slicing in frames of mickey mouse and other methods of gaming the alg. Most of the responses here seem to be buying into Google's hype around number of hours or videos uploaded per second. I think that is a distraction that lets them off the hook for not managing the community they built.

Every algorithm is an editorial decision.

◧◩◪◨⬒⬓
24. taxidu+5j[view] [source] [discussion] 2019-06-14 20:45:20
>>nostra+Fe
This is probably not the thought process this issue would travel down. Costs are typically the first consideration in a semi-skilled position if native english sounding isn't a requirement.
◧◩◪◨
25. scj+Tk[view] [source] [discussion] 2019-06-14 21:02:19
>>Nasrud+dg
In the system I've proposed, if there are enough test-Flanders thrown at the content the times closed should be different enough to trigger an unclear Flanders rating. This would indicate some other metric should be used.

I don't see this test working in isolation. Given it's nature, it's value is in obscure rejection statements rather than acceptance (or "okilly-dokillies" in this case).

To echo what others on this thread have said, there's a lot of content on Youtube. This means that even if they are cautious about which content passes through the filter for kids, there's still a lot available.

◧◩◪◨⬒
26. scarfa+Sm[view] [source] [discussion] 2019-06-14 21:22:40
>>sereto+dc
Considering that rumors are that YouTube is still barely above break even, that is a lot.
◧◩◪◨⬒
27. gus_ma+Io[view] [source] [discussion] 2019-06-14 21:42:07
>>scj+Df
Last year we have a very big mobilization here in Argentina because there was a vote in congress to legalize abortion. It was very big and the discussion split all the political parties.

The big square in front of the congress was split at the half, the pro-choice "green" group was on one side and the pro-life "sky-blue" group was in the other side. Each group had a strong opinion, but the mobilization was quite civilized, I don't remember that anyone get hurt. Anyway, there were small kids on both sides with the handkerchief of the respective color.

Also, what is your definition of kid: 6? 12? 17?

Just imagine that the Church release a video on youtube where Santa visit a lot of children to give them presents, and in particular to a unborn children during the 8 month of pregnancy, and add to Santa a "sky-blue" handkerchief in case someone didn't notice the hidden message. Do you think it should be censored for kids?

replies(2): >>rrix2+Br >>scj+8x
◧◩◪◨⬒⬓
28. rrix2+Br[view] [source] [discussion] 2019-06-14 22:09:21
>>gus_ma+Io
YouTube for Kids is a separate application that is specifically geared towards <10 years of age. It’s not ambiguous.
◧◩
29. d1zzy+nv[view] [source] [discussion] 2019-06-14 22:48:00
>>kartan+vh
Youtube is not a TV channel, it's a video crowdsourced sharing site.

If we want to have a "free" (as in no subscription and no money required to be payed for the service) video sharing/uploading site, what model would that make it work and still have human reviewing? I consider the fact that there may be undesirable videos as the cost of having such a site, similarly how to the "cost" of having a free Internet is that there's going to be lots of hate online and free access to tutorials to make bombs and what not. It's part of the deal and I'm happy with that, YMMV. If you worry about what kids might access then don't let them access Youtube but please don't create laws that would make free video sharing sites illegal/impossible to run.

This is true for pretty much any free Internet service that allows for user content. If all of Internet content production will go back to just "official" creators (because they are the only ones where the cost/benefit math would make sense) I think that would be a huge loss/regression over what we have gained since the age of the Internet.

30. python+3w[view] [source] 2019-06-14 22:55:35
>>strike+(OP)
> They cannot use anything except an algorithm to recommend videos

That’s assuming recommendations need to be personalized. They could recommend at a higher level to groups of people using attributes like age range or region.

I’m not a fan of their personalized recommendations. It’s algorithm overfits my views to recommend videos extremely similar to videos I’ve recently watched, which isn’t really aligned with my interests.

If they took a completely different approach (not personalized) it could really impact the UX in a positive way.

replies(1): >>icebra+F71
◧◩◪◨⬒⬓
31. scj+8x[view] [source] [discussion] 2019-06-14 23:10:37
>>gus_ma+Io
Kid: I'd say the lower bound is around 5, and the upper bound is variable depending on the individual...

In this case, I'd suggest the upper bound doesn't matter, as the criteria for filtering should be "a semi-unattended 5 year old could view it without concern."

All your examples are of topics where it's probably best for parents to initiate their child's education on the topic rather than Youtube randomly putting it in some kid's feed.

replies(1): >>aldous+XJ
◧◩◪◨⬒⬓
32. mc32+eI[view] [source] [discussion] 2019-06-15 02:27:58
>>nostra+Fe
They already outsource their moderation to mostly the Philippines so there’d be no change.
33. robbro+2J[view] [source] 2019-06-15 02:49:16
>>strike+(OP)
I would think having humans more involved in training the algorithm could scale much better.

Also, detecting videos that are inappropriate for children is a lot harder than determining certain content creators that are trustworthy to post videos that are appropriate (and to tag them correctly). That can be learned from the user's history, how many times their stuff has been flagged, getting upvotes from users that are themselves deemed credible, and so on. The more layers of indirection, the better, a la PageRank.

So even without analyzing the video itself, it would have a much smaller set of videos it can recommend from, but still potentially millions of videos. You still need some level of staff to train the algorithm, but you don't have to have paid staff look at every single video to have a good set of videos it can recommend. The staff might spend most of their time looking at videos that are anomalous, such as they were posted by a user the algorithm trusted but then flagged by a user that the algorithm considered credible. Then they would tag that video with some rich information that will help the algorithm in the future, beyond just removing that video or reducing the trust of the poster or the credibility of the flagger.

replies(2): >>ehsank+iN >>PaulAJ+7Y
◧◩◪◨⬒⬓⬔
34. aldous+XJ[view] [source] [discussion] 2019-06-15 03:12:20
>>scj+8x
So a 4 year old kid is not a kid?
replies(1): >>camero+LM
◧◩
35. seanmc+tK[view] [source] [discussion] 2019-06-15 03:23:47
>>kartan+vh
When I was a kid in the 80s, cartoons were basically 30 minute toy commercials. My toddler loves watching videos on YouTube of Russian kids playing with toys, so I guess things haven’t changed much.
◧◩◪◨⬒⬓⬔⧯
36. camero+LM[view] [source] [discussion] 2019-06-15 04:18:57
>>aldous+XJ
They're toddlers or babies if we're arguing semantics.

Kids < 4 really shouldn't have access to YouTube though.

◧◩
37. Camper+YM[view] [source] [discussion] 2019-06-15 04:26:06
>>kartan+vh
How about actually demonstrating harm to children (or to anyone else) before launching a moral panic?

Is that an option?

replies(1): >>shears+FP
◧◩
38. ehsank+iN[view] [source] [discussion] 2019-06-15 04:36:36
>>robbro+2J
The algorithm works really damn well for 99.999% of the cases. It manages to show me great recommendations from very niche things I'm interested in. But it's the very same behavior that can, in some cases, lead to issues.
replies(3): >>Barrin+mO >>robbro+7P >>nrayna+O01
◧◩◪◨
39. unders+kN[view] [source] [discussion] 2019-06-15 04:38:36
>>gus_ma+sc
Maybe Youtube and their revenue sources agree with him.
◧◩◪
40. unders+oN[view] [source] [discussion] 2019-06-15 04:41:03
>>cortes+qf
No, the wrong part was when the pedophiles made inappropriate comments on the videos.
replies(1): >>Buge+UT
◧◩◪
41. Barrin+mO[view] [source] [discussion] 2019-06-15 05:08:41
>>ehsank+iN
are you sure that it's not you who knows very well how to curate their own content and who to subscribe to rather than the recommendation system?

I'm not sure heavy automation is needed here, people jump from content creator to content creator by word of mouth. In contrast most algorithmic suggestions to me seem highly biased towards what is popular in general. I click on one wrong video in a news article and for the next two days my recommendations are pop music, Jimmy Kimmel, Ben Shapiro and animal videos

replies(1): >>ehsank+CV1
◧◩◪
42. robbro+7P[view] [source] [discussion] 2019-06-15 05:33:12
>>ehsank+iN
I'm reminded of how Google images had an issue where dark skinned people sometimes turned up in a search for gorilla. 99.9% of the time, the image recognition algorithm did really well, but here was a case where an error was really offensive. What was (probably) needed was for there to be a human that comes in and, not tag every gorilla image, but simply to give it some extra training around dark skinned humans and gorillas, or otherwise tweak some things specific to that sort of case, so the chance of it happening was reduced to nearly nothing.

There are probably a ton of situations like that in YouTube, where certain kinds of mistakes are hardly noticed (it shows you a video you weren't remotely interested in), but others can be really bad and need special training to avoid (such as where it shows violent or sexual content to someone who likes nursery rhymes and Peppa Pig).

◧◩
43. sharce+bP[view] [source] [discussion] 2019-06-15 05:34:21
>>jgalt2+9g
pichai doesn't come off as enthusiastic. I am a heavy Google product user. Watch all the hardware, I/O events etc, I have seen him use the same sentences multiple times over the past 2 years across events. I get that he won't exude the same charm, excitement as a founder-CEO, nevertheless a lot is left to be desired. A lot of his public statements feel like carefully crafted PR responses. Nothing wrong with crafted responses. When you are a 800 Billion$ company, you gotta be careful, but at least try to give off the perception of being authentic. Google is really bad at the perception game. Apple's really good at that. But I have a strong dislike for stupid moves, even more so than bad moves and Google has made lots of those stupid ones.
◧◩◪
44. shears+FP[view] [source] [discussion] 2019-06-15 05:47:10
>>Camper+YM
I’d say having a 13 year old far right YouTube star post a video threatening to kill the CEO might be harmful, but maybe that’s ok?

https://www.newstatesman.com/science-tech/social-media/2019/...

replies(1): >>Camper+LP
◧◩◪◨
45. Camper+LP[view] [source] [discussion] 2019-06-15 05:48:37
>>shears+FP
Do you seriously think that kid was radicalized on YouTube? Where were the parents?
replies(1): >>shears+101
◧◩◪◨
46. Buge+UT[view] [source] [discussion] 2019-06-15 07:25:28
>>unders+oN
If that's the problem, then gibrown's solution

>Just start banning certain creators from showing up in recommendations if their content crosses the line.

also won't help, because it's not the creators that have content crossing the line, it's the commenters.

◧◩◪◨⬒⬓
47. deanCo+iV[view] [source] [discussion] 2019-06-15 08:08:51
>>nostra+Fe
Because you'd be able to get humans with higher intelligence and better judgement for 10k/year in the Philippeans, than at minimum wage in the US.
◧◩
48. PaulAJ+7Y[view] [source] [discussion] 2019-06-15 09:15:46
>>robbro+2J
The trouble with depending on user flags is that it creates opportunities for blackmail.

https://www.theverge.com/2019/2/11/18220032/youtube-copystri...

◧◩◪◨⬒
49. shears+101[view] [source] [discussion] 2019-06-15 09:50:55
>>Camper+LP
2018: “I’ll pick a topic and just give my opinion about it try to be entertaining, try to be funny, try to be unique and say something other people haven’t said before,” youtuber said.

https://redwoodbark.org/46876/culture/redwood-students-view-...

2019:

In response, the principal of the high school sent a note to students and parents Thursday night regarding the "hate-based video and text posts attributed to one of our students":

https://www.kron4.com/news/bay-area/bay-area-girl-says-she-l...

◧◩◪
50. nrayna+O01[view] [source] [discussion] 2019-06-15 10:05:57
>>ehsank+iN
To me it always pulls me towards television or Hollywood garbage. And videos I have already watched, hundreds of them.
replies(1): >>jotm+K11
◧◩◪◨
51. jotm+K11[view] [source] [discussion] 2019-06-15 10:28:34
>>nrayna+O01
You should check if personalized recommendations are disabled. Google has a history of disabling/enabling settings without telling me.
◧◩◪◨⬒
52. icelan+b21[view] [source] [discussion] 2019-06-15 10:38:20
>>sereto+dc
>> $2B is still nothing to sneeze at, but it's less than Microsoft paid for Minecraft.

One is an investment/one time purchase and the other is a long-term annual liability, slated to grow.

53. razius+331[view] [source] 2019-06-15 10:54:36
>>strike+(OP)
They don't care, they want to push them into approved content rather than recommended content. Aka "these are the topics that you are allowed to speak of".

See current Pinterest scandal and banning from Youtube of any video mentioning this.

◧◩
54. icebra+F71[view] [source] [discussion] 2019-06-15 12:21:06
>>python+3w
No thanks. You try logging out and see the generic recommendations. It's the lowest common denominator, just like anything else targeted at large masses of people.
◧◩
55. icebra+S71[view] [source] [discussion] 2019-06-15 12:25:30
>>mtgx+j3
What those people actually mean is "I can't wait for AI to be so good that it'll be obvious that it should run all the aspects of our society". The current state is irrelevant, nobody wants to put those in charge.
◧◩
56. jodrel+k81[view] [source] [discussion] 2019-06-15 12:36:52
>>mtgx+j3
What are the people thinking who say they can't wait for our society to be run by humans? The most common state of human government capability can't even put human suffering before numbers in a virtual bank account, can't prioritise truth over falsehood, can't restrain themselves from bribery, can't reliably turn up to hearings or ballots, can't organise projects and complete them, can't compromise when millions of people depend on it. We want to dismiss alternatives which haven't even been developed yet for not being good enough?
◧◩◪◨
57. ehsank+CV1[view] [source] [discussion] 2019-06-15 23:12:08
>>Barrin+mO
Not for me, for example I've been watching a few PyCon and I/O talks, and it's been showing me other interesting PyCon talks that are highly ranked. It's also giving me good AutoChess and OneHourOneLife Let'sPlays, both of which I've been very interested in lately.

All three things I just mentioned are fairly niche, comparatively, yet it knows that I've been watching a lot of them lately and is giving me more of it.

◧◩
58. v7p1Qb+vd7[view] [source] [discussion] 2019-06-18 14:07:42
>>scarfa+E3
Probably Neal Mohan on Recode right? The current public number is 500 hours per minute. But that number has been floating around for a while. It's probably higher now.
◧◩
59. v7p1Qb+Ae7[view] [source] [discussion] 2019-06-18 14:13:08
>>mtgx+j3
The argument is that a hypothetical benevolent ASI can't be corrupted like literally all humans can. Those people are likely referring to AI's as they appear in Ian Banks The Culture series.
[go to top]