{"id":25215,"date":"2021-07-11T06:42:45","date_gmt":"2021-07-11T06:42:45","guid":{"rendered":"https:\/\/rbsitsoftwaresolution.com\/blog\/youtubes-recommender-ai-still-a-horror-show-finds-major-crowdsourced-study.html"},"modified":"2021-07-11T06:42:45","modified_gmt":"2021-07-11T06:42:45","slug":"youtubes-recommender-ai-still-a-horror-show-finds-major-crowdsourced-study","status":"publish","type":"post","link":"https:\/\/www.rbsitsoftwaresolution.in\/blog\/youtubes-recommender-ai-still-a-horror-show-finds-major-crowdsourced-study.html","title":{"rendered":"YouTube\u2019s recommender AI still a horror show, finds major crowdsourced study"},"content":{"rendered":"

For years YouTube\u2019s video-recommending algorithm has stood accused<\/a> of fuelling a grab bag of societal ills by feeding users an AI-amplified diet of hate speech, political extremism and\/or conspiracy junk\/disinformation for the profiteering motive of trying to keep billions of eyeballs stuck to its ad inventory.<\/p>\n

And while YouTube\u2019s tech giant parent Google has, sporadically, responded to negative publicity flaring up around the algorithm\u2019s antisocial recommendations \u2014 announcing a few policy tweaks<\/a> or limiting\/purging the odd hateful account<\/a> \u2014 it\u2019s not clear how far the platform\u2019s penchant for promoting horribly unhealthy clickbait has actually been rebooted.<\/p>\n

The suspicion remains nowhere near far enough.<\/p>\n

New research published today by Mozilla backs that notion up, suggesting YouTube\u2019s AI continues to puff up piles of \u201cbottom-feeding\u201d\/low-grade\/divisive\/disinforming content \u2014 stuff that tries to grab eyeballs by triggering people\u2019s sense of outrage, sewing division\/polarization or spreading baseless\/harmful disinformation \u2014 which in turn implies that YouTube\u2019s problem with recommending terrible stuff is indeed systemic; a side effect of the platform\u2019s rapacious appetite to harvest views to serve ads.<\/p>\n

That YouTube\u2019s AI is still \u2014 per Mozilla\u2019s study \u2014 behaving so badly also suggests Google has been pretty successful at fuzzing criticism with superficial claims of reform.<\/p>\n

The mainstay of its deflective success here is likely the primary protection mechanism of keeping the recommender engine\u2019s algorithmic workings (and associated data) hidden from public view and external oversight \u2014 via the convenient shield of \u201ccommercial secrecy.\u201d<\/p>\n

But regulation<\/a> that could help crack open proprietary AI blackboxes is now on the cards \u2014 at least in Europe.<\/p>\n

To fix YouTube\u2019s algorithm, Mozilla is calling for \u201ccommon sense transparency laws, better oversight, and consumer pressure\u201d \u2014 suggesting a combination of laws that mandate transparency into AI systems; protect independent researchers so they can interrogate algorithmic impacts; and empower platform users with robust controls (such as the ability to opt out of \u201cpersonalized\u201d recommendations) are what\u2019s needed to rein in the worst excesses of the YouTube AI.<\/p>\n

\n

Europe lays out its plan to reboot digital rules and tame tech giants<\/a><\/p>\n<\/blockquote>\n

\n

Regrets, YouTube users have had a few \u2026<\/h2>\n

To gather data on specific recommendations being made made to YouTube users \u2014 information that Google does not routinely make available to external researchers \u2014 Mozilla took a crowdsourced approach, via a browser extension (called RegretsReporter<\/a>) that lets users self-report YouTube videos they \u201cregret\u201d watching.<\/p>\n

The tool can generate a report that includes details of the videos the user had been recommended, as well as earlier video views, to help build up a picture of how YouTube\u2019s recommender system was functioning. (Or, well, \u201cdysfunctioning\u201d as the case may be.)<\/p>\n

The crowdsourced volunteers whose data fed Mozilla\u2019s research reported a wide variety of \u201cregrets,\u201d including videos spreading COVID-19 fear-mongering, political misinformation and \u201cwildly inappropriate\u201d children\u2019s cartoons, per the report \u2014 with the most frequently reported content categories being misinformation, violent\/graphic content, hate speech and spam\/scams.<\/p>\n

A substantial majority (71%) of the regret reports came from videos that had been recommended by YouTube\u2019s algorithm itself, underscoring the AI\u2019s starring role in pushing junk into people\u2019s eyeballs.<\/p>\n

The research also found that recommended videos were 40% more likely to be reported by the volunteers than videos they\u2019d searched for themselves.<\/p>\n

Mozilla even found \u201cseveral\u201d instances when the recommender algorithmic put content in front of users that violated YouTube\u2019s own community guidelines and\/or was unrelated to the previous video watched. So a clear fail.<\/p>\n

A very notable finding was that regrettable content appears to be a greater problem for YouTube users in non-English speaking countries: Mozilla found YouTube regrets were 60% higher in countries without English as a primary language \u2014 with Brazil, Germany and France generating what the report said were \u201cparticularly high\u201d levels of regretful YouTubing. (And none of the three can be classed as minor international markets.)<\/p>\n

Pandemic-related regrets were also especially prevalent in non-English speaking countries, per the report \u2014 a worrying detail to read in the middle of an ongoing global health crisis.<\/p>\n

The crowdsourced study \u2014 which Mozilla bills as the largest-ever into YouTube\u2019s recommender algorithm \u2014 drew on data from more than 37,000 YouTube users who installed the extension, although it was a subset of 1,162 volunteers \u2014 from 91 countries \u2014 who submitted reports that flagged 3,362 regrettable videos that the report draws on directly. <\/span><\/p>\n

These reports were generated between July 2020 and May 2021.<\/p>\n

What exactly does Mozilla mean by a YouTube \u201cregret\u201d? It says this is a crowdsourced concept based on users self-reporting bad experiences on YouTube, so it\u2019s a subjective measure. But Mozilla argues that taking this \u201cpeople-powered\u201d approach centres the lived experiences of internet users and is therefore helpful in foregrounding the experiences of marginalised and\/or vulnerable people and communities (versus, for example, applying only a narrower, legal definition of \u201charm\u201d).<\/p>\n

\u201cWe wanted to interrogate and explore further [people\u2019s experiences of falling down the YouTube \u201crabbit hole\u201d] and frankly confirm some of these stories \u2014 but then also just understand further what are some of the trends that emerged in that,\u201d explained Brandi Geurkink, Mozilla<\/span>\u2019s senior manager of advocacy and the lead researcher for the project, discussing the aims of the research.<\/p>\n

\u201cMy main feeling in doing this work was being \u2014 I guess \u2014 shocked that some of what we had expected to be the case was confirmed \u2026 It\u2019s still a limited study in terms of the number of people involved and the methodology that we used but \u2014 even with that \u2014 it was quite simple; the data just showed that some of what we thought was confirmed.<\/p>\n

\u201cThings like the algorithm recommending content essentially accidentally, that it later is like \u2018oops, this actually violates our policies; we shouldn\u2019t have actively suggested that to people\u2019 \u2026 And things like the non-English-speaking user base having worse experiences \u2014 these are things you hear discussed a lot anecdotally and activists have raised these issues. But I was just like \u2014 oh wow, it\u2019s actually coming out really clearly in our data.\u201d<\/p>\n

Mozilla says the crowdsourced research uncovered \u201cnumerous examples\u201d of reported content that would likely or actually breach YouTube\u2019s community guidelines \u2014 such as hate speech or debunked political and scientific misinformation.<\/p>\n

But it also says the reports flagged a lot of what YouTube \u201cmay\u201d consider \u201cborderline content.\u201d Aka, stuff that\u2019s harder to categorize \u2014 junk\/low-quality videos that perhaps toe the acceptability line and may therefore be trickier for the platform\u2019s algorithmic moderation systems to respond to (and thus content that may also survive the risk of a takedown for longer).<\/p>\n

However a related issue the report flags is that YouTube doesn\u2019t provide a definition for borderline content \u2014 despite discussing the category in its own guidelines \u2014 hence, says Mozilla, that makes the researchers\u2019 assumption that much of what the volunteers were reporting as \u201cregretful\u201d would likely fall into YouTube\u2019s own \u201cborderline content\u201d category impossible to verify.<\/p>\n

The challenge of independently studying the societal effects of Google\u2019s tech and processes is a running theme underlying the research. But Mozilla\u2019s report also accuses the tech giant of meeting YouTube criticism with \u201cinertia and opacity.\u201d<\/p>\n

It\u2019s not alone there either. Critics have long accused YouTube\u2019s ad giant parent of profiting off of engagement generated by hateful outrage and harmful disinformation \u2014 allowing \u201cAI-generated bubbles of hate<\/a>\u201d to surface ever-more baleful (and thus stickily engaging) stuff, exposing unsuspecting YouTube users to increasingly unpleasant and extremist views, even as Google gets to shield its low-grade content business under a user-generated-content umbrella.<\/p>\n

Indeed, \u201cfalling down the YouTube rabbit hole\u201d has become a well-trodden metaphor for discussing the process of unsuspecting internet users being dragging into the darkest and nastiest corners of the web. This user reprogramming taking place in broad daylight via AI-generated suggestions that yell at people to follow the conspiracy breadcrumb trail right from inside a mainstream web platform.<\/p>\n

Back as 2017 \u2014 when concern was riding high about online terrorism and the proliferation of ISIS content on social media \u2014 politicians in Europe were accusing YouTube\u2019s algorithm of exactly this: Automating radicalization<\/a>.<\/p>\n

However it\u2019s remained difficult to get hard data to back up anecdotal reports of individual YouTube users being \u201cradicalized\u201d after viewing hours of extremist content or conspiracy theory junk on Google\u2019s platform.<\/p>\n

Ex-YouTube insider \u2014 Guillaume Chaslot \u2014 is one notable critic who\u2019s sought to pull back the curtain shielding the proprietary tech from deeper scrutiny, via his algotransparency<\/a> project.<\/p>\n

Mozilla\u2019s crowdsourced research adds to those efforts by sketching a broad \u2014 and broadly problematic \u2014 picture of the YouTube AI by collating reports of bad experiences from users themselves.<\/p>\n

Of course externally sampling platform-level data that only Google holds in full (at its true depth and dimension) can\u2019t be the whole picture \u2014 and self-reporting, in particular, may introduce its own set of biases into Mozilla\u2019s data set. But the problem of effectively studying Big Tech\u2019s blackboxes is a key point accompanying the research, as Mozilla advocates for proper oversight of platform power.<\/p>\n

In a series of recommendations the report calls for \u201crobust transparency, scrutiny, and giving people control of recommendation algorithms\u201d \u2014 arguing that without proper oversight of the platform, YouTube will continue to be harmful by mindlessly exposing people to damaging and braindead content.<\/p>\n

The problematic lack of transparency around so much of how YouTube functions can be picked up from other details in the report. For example, Mozilla found that around 9% of recommended regrets (or almost 200 videos) had since been taken down \u2014 for a variety of not always clear reasons (sometimes, presumably, after the content was reported and judged by YouTube to have violated its guidelines).<\/p>\n

Collectively, just this subset of videos had a total of 160 million views prior to being removed for whatever reason.<\/p>\n

In other findings, the research found that regretful views tend to perform well on the platform.<\/p>\n

A particular stark metric is that reported regrets acquired a full 70% more views per day than other videos watched by the volunteers on the platform \u2014 lending weight to the argument that YouTube\u2019s engagement-optimising algorithms disproportionately select for triggering\/misinforming content more often than quality (thoughtful\/informing) stuff simply because it brings in the clicks.<\/p>\n

While that might be great for Google\u2019s ad business, it\u2019s clearly a net negative for democratic societies that value truthful information over nonsense; genuine public debate over artificial\/amplified binaries; and constructive civic cohesion over divisive tribalism.<\/p>\n

But without legally enforced transparency requirements on ad platforms \u2014 and, most likely, regulatory oversight and enforcement that features audit powers \u2014 these tech giants are going to continue to be incentivized to turn a blind eye and cash in at society\u2019s expense.<\/p>\n

Mozilla\u2019s report also underlines instances where YouTube\u2019s algorithms are clearly driven by a logic that\u2019s unrelated to the content itself \u2014 with a finding that in 43.6% of the cases where the researchers had data about the videos a participant had watched before a reported regret the recommendation was completely unrelated to the previous video.<\/p>\n

The report gives examples of some of these logic-defying AI content pivots\/leaps\/pitfalls \u2014 such as a person watching videos about the U.S. military and then being recommended a misogynistic video entitled \u201cMan humiliates feminist in viral video.\u201d<\/p>\n

In another instance, a person watched a video about software rights and was then recommended a video about gun rights. So two rights make yet another wrong YouTube recommendation right there.<\/p>\n

In a third example, a person watched an Art Garfunkel music video and was then recommended a political video entitled \u201cTrump Debate Moderator EXPOSED as having Deep Democrat Ties, Media Bias Reaches BREAKING Point.\u201d<\/p>\n

To which the only sane response is, umm what???<\/p>\n

YouTube\u2019s output in such instances seems \u2014 at best \u2014 some sort of \u201cAI brain fart.\u201d<\/p>\n

A generous interpretation might be that the algorithm got stupidly confused. Albeit, in a number of the examples cited in the report, the confusion is leading YouTube users toward content with a right-leaning political bias. Which seems, well, curious.<\/p>\n

Asked what she views as the most concerning findings, Mozilla\u2019s Geurkink told TechCrunch: \u201cOne is how clearly misinformation emerged as a dominant problem on the platform. I think that\u2019s something, based on our work talking to Mozilla supporters and people from all around the world, that is a really obvious thing that people are concerned about online. So to see that that is what is emerging as the biggest problem with the YouTube algorithm is really concerning to me.\u201d<\/p>\n

She also highlighted the problem of the recommendations being worse for non-English-speaking users as another major concern, suggesting that global inequalities in users\u2019 experiences of platform impacts \u201cdoesn\u2019t get enough attention\u201d \u2014 even when such issues do get discussed.<\/p>\n

Responding to Mozilla\u2019s report in a statement, a Google spokesperson sent us this statement:<\/p>\n

\n

The goal of our recommendation system is to connect viewers with content they love and on any given day, more than 200 million videos are recommended on the homepage alone. Over 80 billion pieces of information is used to help inform our systems, including survey responses from viewers on what they want to watch. We constantly work to improve the experience on YouTube and over the past year alone, we\u2019ve launched over 30 different changes to reduce recommendations of harmful content. Thanks to this change, consumption of borderline content that comes from our recommendations is now significantly below 1%.<\/p>\n<\/blockquote>\n

Google also claimed it welcomes research into YouTube \u2014 and suggested it\u2019s exploring options to bring in external researchers to study the platform, without offering anything concrete on that front.<\/p>\n

At the same time, its response queried how Mozilla\u2019s study defines \u201cregrettable\u201d content \u2014 and went on to claim that its own user surveys generally show users are satisfied with the content that YouTube recommends.<\/p>\n

In further nonquotable remarks, Google noted that earlier this year it started disclosing a \u201cviolative view rate\u201d<\/a>\u00a0(VVR) metric for YouTube \u2014 disclosing for the first time the percentage of views on YouTube that comes from content that violates its policies.<\/p>\n

The most recent VVR stands at 0.16%-0.18% \u2014 which Google says means that out of every 10,000 views on YouTube, 16-18 come from violative content. It said that figure is down by more than 70% when compared to the same quarter of 2017 \u2014 crediting its investments in machine learning as largely being responsible for the drop.<\/p>\n

However, as Geurkink noted, the VVR is of limited use without Google releasing more data to contextualize and quantify how far its AI was involved in accelerating views of content its own rules state shouldn\u2019t be viewed on its platform. Without that key data the suspicion must be that the VVR is a nice bit of misdirection.<\/p>\n

\u201cWhat would be going further than [VVR] \u2014 and what would be really, really helpful \u2014 is understanding what\u2019s the role that the recommendation algorithm plays in this?\u201d Geurkink told us on that, adding: \u201cThat\u2019s what is a complete blackbox still. In the absence of greater transparency [Google\u2019s] claims of progress have to be taken with a grain of salt.\u201d<\/p>\n

Google also flagged a 2019 change it made to how YouTube\u2019s recommender algorithm handles \u201cborderline content\u201d \u2014 aka, content that doesn\u2019t violate policies but falls into a problematic grey area \u2014 saying that that tweak had also resulted in a 70% drop in watch time for this type of content.<\/p>\n

Although the company confirmed this borderline category is a moveable feast \u2014 saying it factors in changing trends as well as context and also works with experts to determine what get classed as borderline \u2014 which makes the aforementioned percentage drop pretty meaningless since there\u2019s no fixed baseline to measure against.<\/p>\n

It\u2019s notable that Google\u2019s response to Mozilla\u2019s report makes no mention of the poor experience reported by survey participants in non-English-speaking markets. And Geurkink suggested that, in general, many of the claimed mitigating measures YouTube applies are geographically limited \u2014 i.e., to English-speaking markets like the U.S. and U.K. (Or at least arrive in those markets first, before a slower rollout to other places.)\u00a0<\/span><\/p>\n

A January 2019 tweak to reduce amplification of conspiracy-theory content in the U.S. was only expanded to the U.K. market months later \u2014 in August<\/a> \u2014 for example.<\/p>\n

\u201cYouTube, for the past few years, have only been reporting on their progress of recommendations of harmful or borderline content in the U.S. and in English-speaking markets,\u201d she also<\/span>\u00a0said. \u201cAnd there are very few people questioning that \u2014 what about the rest of the world? To me that is something that really deserves more attention and more scrutiny.\u201d<\/p>\n

We asked Google to confirm whether it had since applied the 2019 conspiracy-theory-related changes globally \u2014 and a spokeswoman told us that it had. But the much higher rate of reports made to Mozilla of \u2014 a yes broader measure of \u2014 \u201cregrettable\u201d content being made in non-English-speaking markets remains notable.<\/p>\n

And while there could be others factors at play, which might explain some of the disproportionately higher reporting, the finding may also suggest that, where YouTube\u2019s negative impacts are concerned, Google directs greatest resource at markets and languages where its reputational risk and the capacity of its machine-learning tech to automate content categorization are strongest.<\/p>\n

Yet any such unequal response to AI risk obviously means leaving some users at greater risk of harm than others \u2014 adding another harmful dimension and layer of unfairness to what is already a multifaceted, hydra-head of a problem.<\/p>\n

It\u2019s yet another reason why leaving it up to powerful platforms to rate their own AIs, mark their own homework and counter genuine concerns with self-serving PR is for the birds.<\/p>\n

(In additional filler background remarks it sent us, Google described itself as the first company in the industry to incorporate \u201cauthoritativeness\u201d into its search and discovery algorithms \u2014 without explaining when exactly it claims to have done that or how it imagined it would be able to deliver on its stated mission of \u201corganizing the world\u2019s information and making it universally accessible and useful\u201d without considering the relative value of information sources. So color us baffled at that claim. Most likely it\u2019s a clumsy attempt to throw disinformation shade at rivals.)<\/p>\n

Returning to the regulation point, an EU proposal \u2014 the Digital Services Act \u2014 is set to introduce some transparency requirements on large digital platforms, as part of a wider package of accountability measures. And asked about this Geurkink<\/span> described the DSA as \u201ca promising avenue for greater transparency.\u201d<\/p>\n

But she suggested the legislation needs to go further to tackle recommender systems like the YouTube AI.<\/p>\n

\u201cI think that transparency around recommender systems specifically and also people having control over the input of their own data and then the output of recommendations is really important \u2014 and is a place where the DSA is currently a bit sparse, so I think that\u2019s where we really need to dig in,\u201d she told us.<\/p>\n

One idea she voiced support for is having a \u201cdata access framework\u201d baked into the law \u2014 to enable vetted researchers to get more of the information they need to study powerful AI technologies \u2014 i.e., rather than the law trying to come up with \u201ca laundry list of all of the different pieces of transparency and information that should be applicable,\u201d as she put it.<\/p>\n

The EU also now has a draft AI regulation on the table. The legislative plan takes a risk-based approach to regulating certain applications of artificial intelligence. However it\u2019s not clear whether YouTube\u2019s recommender system would fall under one of the more closely regulated categories \u2014 or, as seems more likely (at least with the initial Commission proposal), fall entirely outside the scope of the planned law.<\/p>\n

\u201cAn earlier draft of the proposal talked about systems that manipulate human behavior, which is essentially what recommender systems are. And one could also argue that\u2019s the goal of advertising at large, in some sense. So it was sort of difficult to understand exactly where recommender systems would fall into that,\u201d noted Geurkink.<\/p>\n

\u201cThere might be a nice harmony between some of the robust data access provisions in the DSA and the new AI regulation,\u201d she added. \u201cI think transparency is what it comes down to, so anything that can provide that kind of greater transparency is a good thing.<\/p>\n

\u201cYouTube could also just provide a lot of this \u2026 We\u2019ve been working on this for years now and we haven\u2019t seen them take any meaningful action on this front but it\u2019s also, I think, something that we want to keep in mind \u2014 legislation can obviously take years. So even if a few of our recommendations were taken up [by Google] that would be a really big step in the right direction.\u201d<\/p>\n

\n

What Vimeo\u2019s growth, profits and value tell us about the online video market<\/a><\/p>\n<\/blockquote>\n

\n

\n

The next tech hearing targets social media algorithms \u2014 and YouTube, for once<\/a><\/p>\n<\/blockquote>\n

\n

\n

YouTube releases its first report about how it handles flagged videos and policy violations<\/a><\/p>\n<\/blockquote>\n

\n

\n

YouTube: More AI can fix AI-generated \u2018bubbles of hate\u2019<\/a><\/p>\n<\/blockquote>\n

\n

Article source: http:\/\/feedproxy.google.com\/~r\/techcrunch\/social\/~3\/TdLBknKkxFw\/<\/p>","protected":false},"excerpt":{"rendered":"

For years YouTube\u2019s video-recommending algorithm has stood accused of fuelling a grab bag of societal ills by feeding users an AI-amplified diet of hate speech, political extremism and\/or conspiracy junk\/disinformation <\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"jetpack_post_was_ever_published":false,"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[3],"tags":[10,7,14,15,6,12,11],"class_list":["post-25215","post","type-post","status-publish","format-standard","hentry","category-latest-update","tag-computer-news","tag-downloads","tag-microsoft-download-news","tag-networking-updates","tag-technology-update","tag-wordpress-plugin-download","tag-wordpress-theme-download"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t