4/6/22 China Mandates Recommendation Algorithms To Be Able To Be Turned Off
View this email in your browser
Insider Digest 4/6/22: China Mandates Recommendation Algorithms To Be Able To Be Turned Off
Housekeeping / Announcements / Fun:
As I mentioned, the TBC Syndicate is syndicating our third deal! Really excited to partner with long-time friend Huashan Capital. Again, open to accredited investors only, so make sure you qualify!
What You Missed on the TBCI Discord
A translation of 2 viral articles on how the lockdown is going and insight into some of the execution difficulties on the ground
Chinese government has not increased pace of vaccines …
The below chart is as of 4/5, so already one day outdated, but the green rows are the districts that are experiencing case declines and the red ones are the ones growing in severity … unfortunately the top two districts affected, Pudong and Minhang, are also the most populous
Columns are total population in ‘0000s, total positives, positive rate, 1 of X positive, last 14 day trend, and comments on how many days it’s been dropping (green rows) or if it’s rising (red)
Some Chinese companies looking at using US auditors to get around PCAOB issue
As reported by WSJ
But also previously mentioned by hc-72 who noted that: Meihua used Houston-based Briggs and Veselka as auditors for IPO financials and their FY2020 is not showing in the PCAOB list.
I made an accidental mega viral thread on Twitter about Huawei’s massive employee dividend
Which Reuters then wrote about
But point is, Huawei is suffering but not dead. Last year profits are buoyed by one time gains from sale of Honor and X86 server division though
Either way, this led me to read up a lot more on detail on Huawei’s compensation scheme, and while it is a very, very demanding place to work, it does have some clever mechanisms in its “ESOP,” some details you might not have known are explained here
4/5/22 China Mandates Recommendation Algorithms To Be Able To Be Turned Off
As of March 1, CAC has been requiring that all recommendation algorithms must come with the ability to be turned off. While other countries have talked about it, China is the first one to do it, in line with its desire to become a regulatory leader in emerging technologies. Remember, this was a regulation that had come out in draft form last year and is one of the first steps of China’s AI framework.
I’m not here to analyze the policy, which is pretty clear-cut, but to try to make sense of what the business impact might be, and I can’t help but think it’s going to be minimal, despite what a lot of survey results say, to my surprise. But I’ll lay out the evidence and you can please prove me wrong!
First, the basics. What are the types of recommendation algorithms that fall under this guideline? Everything, actually. The exact wording from the CAC is: “the use of algorithmic technologies to provide information to users: such as generation and synthesis, personalized push, sorting and selection, retrieval and filtering, scheduling and decision-making.” Reading that, it seems like it should also include search results ... but I suppose that would be highly impractical. But anyway, everything else is included.
Next, the reality. Right now it takes 5-7 taps in each app to turn off the recommendation engine. Pinduoduo and Douyin are the only two big platforms at 7 steps, while Kuaishou, Meituan and Taobao are more brave, offering the option at just 5 steps. Some platforms will let you turn off personalized content recommendations and personalized ads separately. You can choose to keep one but not the other if you wish.
Notably, no platforms are allowing people to manage their tags individually like Taobao currently does. (Recall that the same rules already went into effect for ecommerce back in 2017 and Taobao has allowed you to turn off your 50 customer markers individually.)
At the end of the day, are the regulators trying to kill off recommendation algorithms? I don’t think so. Again, quoting official CAC statements, the intent is to *“strengthen ***the management of user models and user tags, improve the rules of interest markers recorded in user models and the management rules of user tags; strengthen the management of the algorithm recommendation service page, and establish and improve the mechanism of manual intervention and user self-selection.” Yeah, kind of a mouthful, but you get the idea — it’s about offering transparency and choice. The method chosen so far is a rather blunt instrument, since it is a “one-click” on-off button instead of one where users can “intervene and self-select” as the regulations advocate. But that is understandably a very difficult technical problem, and one that does require the platforms to give away much more information about their algorithms than they’d probably like to, even if they could offer the adjustments, so it makes sense that without further regulatory prodding, we’ll probably only have this buried one-button function for the foreseeable future.
But Chinese users say they are in favor of turning them off. One Chinese survey (of 1144 internet users) showed that 53.8% of respondents would turn off recommendation algorithms. However, their reasons for doing so may surprise you:
curb predatory over-consumption (68.2%)
prevent discriminatory pricing (61.9%)
decrease addiction (57.7%)
stop over-collection of personal data (46%)
Initially, I had thought this was just a case of users misunderstanding the value of recommendation engines, or a case of stated vs. revealed preferences. But when you look at the reasons given, it seems that it’s actually because these algorithms are working too well. Or that’s the self-reported case in China anyway.
Some are finding they regret turning the algorithms off though. In one hilarious anecdote, one user who turned off his short video recommendation algorithm turned it back on after 5 minutes ... “arghh! it only seems to be suggesting COVID content to me, because I’m located in a hard-hit province.” Adult users find themselves frustrated that they are now seeing content that would entertain much younger audiences instead. Furthermore, it’s not like they are being served less ads ... the ads are simply less relevant. The user experience is by no means better. (I’ve tried this myself inadvertently with TikTok on my kitchen Alexa. If you don’t log in, it simply serves you what it thinks the “modal user” wants to see, I suppose, and it just had no overlap with my interests whatsoever.)
Nonetheless, I was very surprised by the high numbers of folks wanting to turn off the algorithms and so did my own unscientific Twitter poll. A surprising number of people responded, and what do you know? Of those who answered, 47% said they would turn it off for all or most of the apps they used! Another 36% said they would do it for some apps, and only 17% said none or barely any, which would have been my own answer.
However, most of these respondents seemed to be focused on the poor user experience the algorithms provided, specifically highlighting that they didn’t feel like they were being served sufficiently relevant content, or just wished that they had more freedom to see what was out there. This seems almost to be the opposite of the biggest concerns of Chinese netizens. That the algorithms are not working as well as intended, versus too well / too exploitative. Of course, I must emphasize again that my survey in particular was very unscientific, so you should take these observations with a lump of salt.
But besides the fact that maybe Chinese recommendation algorithms are working “too well” for netizens’ tastes, are they actually make-or-break for businesses?
The answer? It varies. The most oft-cited stats of Netflix saying 75% of what people watch are from some kind of recommendation, 60% of YouTube clicks are from recommendations, and 35% of Amazon’s sales are from cross-sales (this data point is from 2006 though!!), sound impressive but don’t capture the full business impact. According to this 2019 analysis, ecommerce direct revenue increases are often just between 1-5% (an HBR article I can’t locate but I see quoted apparently claimed 10%+, but still, that isn’t as tremendous as you’d think), although for other businesses that rely on increasing clickthrough rates, those can sometimes be boosted substantially (+30%, even 300% maybe). So it depends on your business model and what you’re using the algorithm for, though it does seem that it’s almost always a positive impact, unless you’ve launched it in such a way that it detracts from the user experience. And 1% increase in sales is still 1%, and can be a lot if you’re talking about a massive base. So while it is not make-or-break for most businesses — unless it is the core of your product like the short video products out there — it is still one of those things that platforms are not going to give up willingly on their own, because they do produce positive business outcomes.
Finally, as I’ve said before when this rule came out, we can look at the impact that it had on the ecommerce sector, which has had to live with a version of this rule since 2019. The law in question is the Ecommerce Law passed in September 2018 by the CAC, and the specific statute I’m talking about is #18, which states that “if an e-commerce operator provides consumers with search results for goods or services based on their hobbies, consumption habits, and other characteristics, they shall also provide the consumers with options that are not specific to their personal characteristics ...” This resulted in Alibaba having had the ability to let you manage your platform-assigned personalization tags for the last three years. And I think the business results speak for themselves ... it didn’t seem to have much impact. How much of that is just the fact that users couldn’t figure out how to work it vs. they genuinely didn’t want to edit their tags, I unfortunately do not know, but it’s pretty obvious that very few people ended up using it, which is the same fate I’d predict for this regulation as well, especially if the platforms continue hiding it like they’re doing.
So while this is kind of a big deal for companies to give up a bit more control to consumers, and is likely the start of much more impactful regulations, it does not by itself seem to have much business impact. Primarily because the majority of users will never figure out how to turn it off, and secondarily because the platforms — digital entertainment / content platforms — are such crappy user experiences without algorithms.
What do you think?
Reply here on the Circle forum.
Have any comments or questions? See you on the Discord server!
Copyright (C) *|CURRENT_YEAR|* *|LIST:COMPANY|*. All rights reserved.
*|IFNOT:ARCHIVE_PAGE|**|LIST:DESCRIPTION|**|END:IF|*
*|IFNOT:ARCHIVE_PAGE|**|HTML:LIST_ADDRESS_HTML|**|END:IF|*
Update Preferences | Unsubscribe
*|IF:REWARDS|* *|HTML:REWARDS|* *|END:IF|*