WEBVTT - Kids Videos That Aren't for Kids

0:00:03.640 --> 0:00:06.600
<v Speaker 1>There's something very strange happening on the YouTube Kids app,

0:00:06.920 --> 0:00:09.600
<v Speaker 1>and some of it is certainly not appropriate for children.

0:00:10.440 --> 0:00:18.040
<v Speaker 1>Jonathan Strickland and this is tech stuff daily. YouTube Kids

0:00:18.160 --> 0:00:20.639
<v Speaker 1>is an app that Google markets directly to parents of

0:00:20.720 --> 0:00:23.320
<v Speaker 1>young children. In fact, on the landing page for the app,

0:00:23.360 --> 0:00:27.200
<v Speaker 1>you'll find a description that reads a safer online experience

0:00:27.240 --> 0:00:32.120
<v Speaker 1>for kids. Just below that is a caveat, however, and

0:00:32.159 --> 0:00:34.240
<v Speaker 1>one that seems to be in response to a flurry

0:00:34.240 --> 0:00:37.920
<v Speaker 1>of news stories about questionable content popping up on the platform.

0:00:37.960 --> 0:00:41.440
<v Speaker 1>The caveat says YouTube Kids is designed to filter out

0:00:41.479 --> 0:00:45.040
<v Speaker 1>inappropriate videos for kids, but no system is perfect. If

0:00:45.080 --> 0:00:47.760
<v Speaker 1>a video that's inappropriate shows up in the app, you

0:00:47.840 --> 0:00:50.120
<v Speaker 1>have the power to block it, flag it, and bring

0:00:50.159 --> 0:00:54.040
<v Speaker 1>it to our attention for fast review. That probably comes

0:00:54.080 --> 0:00:56.400
<v Speaker 1>as small comfort to parents who have seen their kids

0:00:56.440 --> 0:00:59.920
<v Speaker 1>react to some truly bizarre and in some cases out

0:01:00.080 --> 0:01:06.200
<v Speaker 1>right disturbingly violent or provocative content. Media outlets including Medium

0:01:06.319 --> 0:01:09.039
<v Speaker 1>and The New York Times have reported on the large

0:01:09.120 --> 0:01:12.399
<v Speaker 1>number of odd and sometimes upsetting videos that come up

0:01:12.440 --> 0:01:16.400
<v Speaker 1>either through searching specific terms or as suggested videos picked

0:01:16.440 --> 0:01:20.440
<v Speaker 1>by various algorithms. So, what the heck is going on there? Are,

0:01:20.640 --> 0:01:23.880
<v Speaker 1>not surprisingly, several elements to this story, and part of

0:01:23.920 --> 0:01:26.360
<v Speaker 1>it is just due to the sheer amount of video

0:01:26.440 --> 0:01:30.280
<v Speaker 1>being uploaded to YouTube. That's about four hundred hours of

0:01:30.360 --> 0:01:35.280
<v Speaker 1>video joining the platform every minute. In other words, within

0:01:35.360 --> 0:01:38.959
<v Speaker 1>ten minutes, four thousand hours of video jumps on board.

0:01:39.319 --> 0:01:42.760
<v Speaker 1>At that scale, it's literally impossible to have human beings

0:01:42.800 --> 0:01:45.440
<v Speaker 1>combed through the video content and make sure everything is

0:01:45.480 --> 0:01:48.120
<v Speaker 1>on the up and up. You either automate or you

0:01:48.280 --> 0:01:52.080
<v Speaker 1>shut down. If humans were responsible for reviewing all the footage,

0:01:52.200 --> 0:01:55.360
<v Speaker 1>they'd find themselves falling further behind with every minute of

0:01:55.480 --> 0:01:57.760
<v Speaker 1>video they reviewed. It would be as if they were

0:01:57.760 --> 0:01:59.680
<v Speaker 1>in one of those horror movies where a character runs

0:01:59.680 --> 0:02:01.600
<v Speaker 1>down a hall only to see the door at the

0:02:01.720 --> 0:02:04.640
<v Speaker 1>end get further away. Another issue is that we've gotten

0:02:04.680 --> 0:02:08.240
<v Speaker 1>pretty good at automation in general, not great at it,

0:02:08.560 --> 0:02:11.399
<v Speaker 1>but good enough for it to be a problem. For example,

0:02:11.639 --> 0:02:13.839
<v Speaker 1>you might create an algorithm that looks for the most

0:02:13.880 --> 0:02:18.560
<v Speaker 1>popular search terms used on platform like YouTube kids. You

0:02:18.600 --> 0:02:21.240
<v Speaker 1>get this list of terms, which likely have little to

0:02:21.280 --> 0:02:24.080
<v Speaker 1>do with one another, and you use a different algorithm

0:02:24.120 --> 0:02:27.280
<v Speaker 1>to piece together a nonsensical video that takes those different

0:02:27.280 --> 0:02:30.480
<v Speaker 1>components and creates a sort of mash up. The resulting

0:02:30.560 --> 0:02:33.560
<v Speaker 1>video would likely be pretty awful, but still pop up

0:02:33.600 --> 0:02:36.080
<v Speaker 1>in search or along a list of related videos in

0:02:36.080 --> 0:02:39.600
<v Speaker 1>a sidebar. There are also automated programs that drive up

0:02:39.680 --> 0:02:42.760
<v Speaker 1>video viewing numbers. In other words, some of these videos

0:02:42.760 --> 0:02:46.320
<v Speaker 1>may have been designed and even assembled by bots and

0:02:46.520 --> 0:02:50.200
<v Speaker 1>shown to other bots. Based on the quality of some

0:02:50.240 --> 0:02:52.000
<v Speaker 1>of these videos, I think it's safe to say that

0:02:52.120 --> 0:02:55.720
<v Speaker 1>robots have exceedingly low standards when it comes to entertainment.

0:02:56.480 --> 0:02:58.600
<v Speaker 1>If that were all there were to this story, it

0:02:58.639 --> 0:03:02.239
<v Speaker 1>would be curious but not terribly important. But those videos

0:03:02.240 --> 0:03:06.280
<v Speaker 1>are also seen by children, not just brainless algorithms, and

0:03:06.360 --> 0:03:09.480
<v Speaker 1>some of the videos can be quite upsetting. It's likely

0:03:09.600 --> 0:03:13.320
<v Speaker 1>that so called bad actors made some of those videos,

0:03:13.360 --> 0:03:18.040
<v Speaker 1>particularly the more egregiously violent or inappropriate ones, whether they

0:03:18.040 --> 0:03:20.560
<v Speaker 1>were doing it as an attempt at satire or just

0:03:20.639 --> 0:03:22.960
<v Speaker 1>to take figures that appeal to kids and put them

0:03:22.960 --> 0:03:26.079
<v Speaker 1>in inappropriate situations. The problem is that some of these

0:03:26.160 --> 0:03:28.920
<v Speaker 1>videos are slipping through the filters on YouTube kids and

0:03:28.960 --> 0:03:32.280
<v Speaker 1>are being seen by actual children. So why would anyone

0:03:32.320 --> 0:03:35.400
<v Speaker 1>bother to do this? If we set aside those who

0:03:35.480 --> 0:03:39.240
<v Speaker 1>just enjoy stirring up trouble and causing distress. It likely

0:03:39.280 --> 0:03:43.280
<v Speaker 1>comes down to money. Specifically, we're talking ad revenue. If

0:03:43.320 --> 0:03:44.960
<v Speaker 1>you want to make a lot of money off of

0:03:45.000 --> 0:03:47.680
<v Speaker 1>YouTube videos, you need to get as many eyeballs on

0:03:47.760 --> 0:03:50.400
<v Speaker 1>those videos as possible while bringing down the cost of

0:03:50.400 --> 0:03:53.120
<v Speaker 1>production as much as you can. That's where all that

0:03:53.240 --> 0:03:57.000
<v Speaker 1>automation comes in. The videos don't have to have good content,

0:03:57.040 --> 0:03:59.560
<v Speaker 1>they just have to be good at being discovered. A

0:03:59.640 --> 0:04:02.880
<v Speaker 1>quick search through the kid videos will uncover plenty of

0:04:03.000 --> 0:04:07.360
<v Speaker 1>ones that have nonsensical titles consisting solely of popular keywords.

0:04:07.720 --> 0:04:10.280
<v Speaker 1>If people were involved in any step in that process,

0:04:10.600 --> 0:04:13.800
<v Speaker 1>it was with a light touch. Children also tend to

0:04:13.840 --> 0:04:17.880
<v Speaker 1>fixate uncertain characters, songs, nursery rhymes, or other elements that

0:04:18.000 --> 0:04:21.760
<v Speaker 1>video producers can exploit. The result is an onslaught of

0:04:21.839 --> 0:04:26.320
<v Speaker 1>low quality and sometimes wildly inappropriate content. The money side

0:04:26.320 --> 0:04:29.640
<v Speaker 1>of this issue is a huge problem. The video publishers

0:04:29.640 --> 0:04:32.359
<v Speaker 1>are making money by serving up ads against these videos.

0:04:32.640 --> 0:04:35.000
<v Speaker 1>YouTube takes a cut of that money as well, so

0:04:35.080 --> 0:04:37.680
<v Speaker 1>there's a lot of incentive to make tons of content

0:04:37.800 --> 0:04:40.960
<v Speaker 1>optimized to get lots of views, and little incentive on

0:04:41.000 --> 0:04:44.200
<v Speaker 1>YouTube's part to crack down on it. In addition, short

0:04:44.200 --> 0:04:48.480
<v Speaker 1>of developing incredibly effective algorithms that can somehow judge quality.

0:04:48.520 --> 0:04:51.560
<v Speaker 1>There are few logical options open to the company to

0:04:51.600 --> 0:04:55.440
<v Speaker 1>combat the trend. At the moment, the onus seems to

0:04:55.480 --> 0:04:58.640
<v Speaker 1>fall on parents. They are meant to report troublesome videos,

0:04:58.640 --> 0:05:00.920
<v Speaker 1>which can be removed by you Tube, but by then

0:05:00.960 --> 0:05:03.560
<v Speaker 1>the damage may be done personally. I do think parents

0:05:03.560 --> 0:05:06.400
<v Speaker 1>have a responsibility to make certain their children aren't being

0:05:06.440 --> 0:05:10.320
<v Speaker 1>exposed to inappropriate material, but I also think YouTube should

0:05:10.320 --> 0:05:13.800
<v Speaker 1>hold itself accountable for marketing and app to kids without

0:05:13.839 --> 0:05:16.160
<v Speaker 1>being able to make certain the content on that app

0:05:16.240 --> 0:05:20.039
<v Speaker 1>is actually appropriate. If the company cannot guarantee that, I

0:05:20.040 --> 0:05:22.760
<v Speaker 1>think it should probably not offer the app, But there's

0:05:22.760 --> 0:05:25.120
<v Speaker 1>a lot of money to be made, so I suspect

0:05:25.120 --> 0:05:28.320
<v Speaker 1>it will stick around. That's all for today. To learn

0:05:28.320 --> 0:05:32.200
<v Speaker 1>more about streaming media, online communities, and other tricky subjects

0:05:32.200 --> 0:05:34.279
<v Speaker 1>in the age of the Internet, subscribe to the Tech

0:05:34.320 --> 0:05:37.280
<v Speaker 1>Stuff podcast we published on Wednesdays and Fridays, and we

0:05:37.320 --> 0:05:40.880
<v Speaker 1>explore all things tech that includes the good, the bad,

0:05:41.240 --> 0:05:43.120
<v Speaker 1>and the ugly. I'll see you again, s