0.0 Context setting

Wednesday 29 January 2020. The rain has come, eventually, to Portland and shows no sign of letting up. This is good because it feels like we had hardly any rain in the autumn, and I’d hate for this green city to become dry out and go brown.

I am, to put it mildly, somewhat freaking out and am using writing this as a bit of a self-care technique. I’ve also realized somewhat late in my life that doing this kind of writing actually helps with what my brain has partitioned into “day job” work, having figured out that thinking and writing in this way leads itself to a bunch of associations and analogies that turn out to be super helpful. So, self: remember that.

Heads up: around 3,200 words this time, so settle in.

1.0 Some things that caught my attention

Three big things and then a few smaller things this time.

1.1 The ATCR Boundary

Or: the Algorithmic Timeline and Content Recommendation Boundary

Erica Joy posted a thought experiment the other day:

a thought experiment:

imagine that the leaders at youtube, twitter, facebook, and instagram got together and decided that they would all stop suggesting/promoting content via algorithmic feeds for the next year.

what would happen? [@ericajoy]

which prompted a thought from me: if you had to draw a line separating a pre- and post-algorithmic timeline and content recommendation boundary, what would be on either side of the line?

My first thoughts for pre-algorithm mediated timelines and recommendations were LiveJournal, Blogger (but before Blogdex?), and Tumblr.

The post-boundary ones are much easier, because they’re what we have now. Facebook’s feed, for sure. Amazon’s “people who bought… also bought” recommender. Netflix’s recommender. We needn’t even mention YouTube’s.

Amy Hoy had some great additions along blog, uh, lines: (bloglines!)

trackbacks
link blogs
hand-edited aggregators like macsurfer
aggregator sites
technorati
netscape’s what’s cool 😂
yahoo directory
delicious

[@amyhoy]

One of the key features of that list is the amount of human agency. These sites acted on entirely human decisions, although maybe not Technorati?

Later on, I also remembered Flickr’s patent for serendipity and explore, which I’d mark as the first step beyond past the boundary. A good four years ago you’ve got discussion on Flickr groups about how the Explore algorithm can be understood, much like how people are now deconstructing how TikTok recommends videos in its feed.

Somewhat amusingly, while googling for Flickr’s patent, I found… myself writing about Flickr’s interestingness and serendipity back in 2014 in a much earlier episode of this very same newsletter.

I’ve also finished reading Gibson’s new book, Agency, and well, there’s this bit I also wrote about serendipity and routing engines:

In this way, asking for serendipity in terms of a “most interesting route” from Google is giving up a bit of agency, a bit of control. [Episode Ninety Eight: Engineering Serendipity; Genuinely Useful People Personalities; Hubris]

Of course it wouldn’t take someone that long to mention and mourn Google Reader.

I think the point I was trying to get at with this thought or provocation was to answer Erica’s question obliquely. My first attempt at answering her question was along the lines of “the CFO would have a fit” and “people would have to figure out where revenue was going to come from”, but then I got distracted by what I thought might be a more interesting exploration of what could take the place of recommended content in algorithmic feeds.

(Yes, I am avoiding the question of whether an environment exists that could support such an alternative arrangement in the first place).

In my vague, ill-defined category of post-boundary things, I’d even count Slashdot’s front page and anything that uses a crowdsource up/down-vote mechanic because it opens itself up to crowdsource brigading. Once you’re open to brigading, then it’s easy to reach for algorithmic mitigation methods, and at that point I’d argue you’re on your way to the content recommendation apocalyptic singularity we’re currently experiencing. (Aside, I liked the phrase runaway content recommendation event and reserve the right to use it in the-novel-that-must-not-be-named).

As I’m thinking/writing about this, I keep coming back to the idea of sufficiently-in-the-loop-humans. I don’t think in principle I have a problem with algorithmic content recommendation systems, I just think I have a problem with insufficiently managed algorithmic content recommendation systems. This, again, comes back to an issue of scaling and externalization of costs and business models. First it’s cheaper to me if I employ fewer humans — be they community moderators (not content moderators!), then even cheaper if I employ no humans, and even cheaper if I push all negative costs onto other parts of the ecosystem.

I suppose a big question is whether this (whatever this is) scales (whatever scales means). I feel like I could tell a story with explanatory power (but no predictive power, I think), where algorithmic content recommendation and content positioning was fitter in the post-capitalism environment and, well, ended up being a runaway event. How could small groups of humans, loosely joined, aggregating individual but ultimately human decisions be amplified? Is this any different than automatically identifying trending hashtags and highlighting them? Is it that I feel uncomfortable with automatically identified topics through semantic analysis or whatever deep learning technique is currently in vogue to be involved in re-ordering a feed?

Given past personal performance, more on this over the next five years, I expect.

(I mainly like the title for this section because I was trying to come up with something that sounded and worked a bit like the K-T boundary (which isn’t even the K-T boundary anymore! It’s the K-Pg boundary, which is like a double insult because it feels like Paul Graham is just getting everywhere).

1.2 It’s okay to be dumb

So the long story short is that Sonos were going to stop active development on some of their old speakers and, combined with the news that “recycling” a Sonos speaker actually meant bricking it (i.e. blowing its firmware to make it unusable by anyone).

A friend pointed out that there’s no technical reason why smart speakers like Sonos couldn’t have had a standard analog-in that would’ve dumbly-routed audio to the speaker so that the smart speaker could still be a dumb speaker. Its job, in the end, is to produce audible sound, so I expect you could make an argument that failing any internet connectivity whatsoever, you might still be able to use it as a speaker. It should be clear that it was a business decision to explicitly not include the analog-in, and that it might even have been a reasonable one at the time, because the product was — and is — a smart, connected speaker.

But then there was a panic because of the earlier terrible business decision of bricking “recycled” speakers combined with unclear communications about what would happen with older products going forward. They weren’t going to stop working at all, but they wouldn’t interoperate with new(er) speakers, and would also hold those newer speakers back in terms of implementing new features. Again, this is kind of reasonable from a software point of view but also, a cost/benefit business decision, to make a decision on a reasonable level of support after ~10 years.

Again, I think some of the dissonance is down to the fact that at the end, a Sonos Speaker is a speaker. So is there an underlying principle?

Should all smart objects have a dumb fallback, be it in the event of connectivity outage, end-of-life for software support (for example, for security reasons, some of which might even be out of the vendor’s control thanks to, say, a newly discovered underlying hardware vulnerability from a supplier) or just going out of business because it’s been a while and nothing in your company has been sunset lately, or because you’ve just been strategically acquihired?

You can guess that my answer is “of course they should”, but I have never run a smart object business and I have never been involved in putting together a BOM.

Consider the converse: is it okay for a KitchenAid mixer to stop working after 10 years? (To be clear: the old Sonos speakers would still work, in a fashion).

If your smart fridge (god forbid you have a smart fridge) loses connectivity, then you’d still expect it to fridge, right?

So now I have a train of thought about what classes and what kinds of dumb fallback smart objects might have.

Smarter people than me, like Alexandra Deschamps-Sonsino have been doing work on smart objects forever, and have built internet of things assessment tools like #betteriot.

Related: I had a problem in a cafe where my iPhone had wifi connectivity but my iPad didn’t (oh, first world problems) and I diagnosed it to the point where I realized the iPad only had IPv6 connectivity. This, I thought, was weird and I’d assumed it was a problem with iOS. The answer, of course, was more distressing: the router’s DHCP server was probably hosed, and IPv6 DHCP servers fall down… less often, maybe?

Anyway, the whole situation prompted this from Arkadiy Kukarkin:

we had this happen once with a rogue DHCP server in a smart lightbulb that sometimes was faster than the router and only gave out IPv6 leases

[@parkan]

of which, oh my god this is why we can’t have nice things and I continue to believe that horrible software/software is hard is a Great Filter. Also, that lightbulb has Dunning-Kruger and should get a hold of itself.

1.3 The internet is a real place with people in it doing people things to each other, why is this so hard to understand

No, really, why is this so hard to understand?

On the day Kobe Bryant died, a Washington Post reporter was dogpiled on Twitter for linking to a piece at a different publication commenting on Kobe Bryant’s rape case. That reporter was suspended by the Washington Post, since reinstated.

Stacy-Marie Ishmael, who you should follow, pointed out:

A basic responsibility of newsrooms right now is to protect their reporters from malicious dogpiling, harassment, and threats of violence — whatever the source.

[@s_m_i]

This collided in my head with the news that YouTube moderators are being forced to sign a statement acknowledging the job can give them PTSD.

This prompted the thought that occupational health and safety standards and regulations (in the U.S., more commonly known as OSHA) should obviously apply to online environments:

I’m not being facetious here; OSHA for hazardous online work environments - from content moderators to journalists.

[@hondanhon]

It feels weird to me (but, I suppose, understandable, for those who do not have to deal with it and are privileged enough to not have to deal with it) that there are people who don’t think online environments are, or can be, harmful, or that protections only apply to some weird distinction of the “real world”, like making sure a fire extinguisher exists or that the exit is clearly marked.

This isn’t as simple as some sort of net-native thing or a “young people these days” thing. There are people, like who Gretchen McCulloch talks about, who have a weird internet career, and the fact that they’re even best-described as weird internet careers is an example of this liminal, in-between state that we’re in. (Thank you to Kelsey Merkley for introducing me to weird internet career, something I’ve been trying to figure out for myself for… well, ever since I’ve been on the internet).

My point is that the internet has been so othered that failing to acknowledge it as a place where people are, doing people things, is causing increasing amounts of harm, and not just because there are more people there doing more people things.

At a high level, internet people bear some blame for this. The rhetoric from John Perry Barlow in the Declaration of Independence of Cyberspace is separatist, casting off whatever it is that constraints us in the physical world. I’m not seeking to minimize the differences (and there are good ones!) afforded by online media and spaces, but the sticking point is this, which I’ve already apparently facetiously repeated above, but actually mean sincerely and want people to think about and take seriously:

The internet is full of people doing people things.

While things may happen differently on the internet and while certain assumptions that we’ve grown up with or evolved with may not apply in the same way on the internet that are definitely very true and interesting (e.g. the speed of communication, the potential for perfect replication with near zero cost), ultimately, the internet is a place populated and created by people.

The intersection between the online world and what some people still refer to as real life is people. People are where those worlds touch and come together. And for increasingly large and unignorable numbers of people, those places touch all the time now, and we have the capacity to bring them with us, wherever we are.

I have Will Perrin to thank for pointing me to the equivalent U.K. legislation, from the Health and Safety Executive, which already recognises that telephone verbal abuse is a form of work-related violence. But, look! The guidance specifically covers telephone verbal abuse, even though the underlying legislation doesn’t mention a mode of abuse, merely the context, that it happen at work. Now, it’s reasonable to say that the guidance is couched as telephone verbal abuse because of the number of call-center jobs. But again, this is reactionary and betrays a regulatory misunderstanding of what the internet is and what it’s doing.

There’s a connection here, too, with changing public attitudes to mental health and disability, and the idea that you can suffer non-physical, debilitating harm, but it looks like you’re fine on the outside. Turns out, not true.

This sense that online is a sufficiently different place applies to people you’d otherwise think are net-native, too: “Facebook” talks about how it is different and a different space from physical spaces and while they’re not wrong, they’re also missing some fundamentals, I think, about what people do and how people do things.

One tiny thought I have is that legislative or government attention to this might change as the number and type of constituents affected start changing too. In the U.S., the electoral college means less-dense, commonly rural areas have disproportionate legislative representation, so I can spin a story about new internet-mediated jobs like customer service providing new employment opportunities in those areas. If those working conditions don’t turn out to be so great, maybe there’ll be some pressure.

1.4 Some smaller (ha!) things

• Under Armour discontinues support for its connected health hardware, in the latest episode of companies that don’t make hardware and don’t treat it as a core part of their business find it really hard and invariably give up, of which see also: Nike and the Fuelband, the latter case somewhat complicated by their relationship with Apple. Related to this, is…

• … Buzzfeed’s editor Ben Smith moving over to the New York Times. The reason why they’re related? Because they are more data in my constant “is it faster to start a new thing than to change an existing thing” question, a sibling to “can you start a new culture faster than changing an existing institution’s culture”, which many of you will instantly recognize as “oh god, you’re talking about digital transformation” or “my company doesn’t get the internet”. The galaxy brain answer to this question of course is: that’s not right question, or you’re asking the wrong question.

• Here is a reasonable history of Friend-of-a-Friend, the social graph markup standard which scratches many itches, in the context of waves hands whatever it is that’s going on with the internet and people these days.

• Reinventing and renaming a giant expensive logistics system might not be the best way to get a logistics system that works, but sure, give it a shot. It does not help, in my mind, that ALIS sounds a bit like VALIS.

• If you are of a certain internet age, you may remember SubEthaEdit, used by some people to mind-blowingly take collaborative realtime notes at conferences like O’Reilly ETech, which means you might also remember that the team behind it also developed the iOS version of Carcasonne (now EOLed), and for an even deeper cut, you may remember that it coincided with an interest in Erlang (for which I have a Mr. Biddulph to thank, I think.

• This 2016 essay that won a prize for practical ethics arguing that we should take moral advice from computer systems on my first read fails to address a number of practical issues and, at an extreme, starts to assume things like strong A.I., or requires what might be access to a societally unreasonable amount of personal information. There’s the mathematical proof that you could compute a moral recommendation based on sufficient axioms, but then what I see happening is that the implementation of those sufficient axioms is left as an exercise to the reader, and, well, that’s cheating. I started re-reading Daniel Dennett, and was reminded of his argument that Searle’s Chinese Room completely misses the point: a Chinese Room that could do what Searle argues is essentially the stupendously complex system that is a human being. Practically speaking, imagine how big that room must be, how complex that room must be.

• Like everyone else you know, I am writing a book. I’m actually writing two books (I think I’m at ~20k words on the novel, and about 1/3 the way through the non-fiction proposal) and trying very hard to ignore all the people who are happy the book-writing, because as soon as I think about it, all the words stop coming out. I am reminded of Kate McKean telling me something along the lines of “the worst thing you can tell someone is that they should write a book,” which instantly makes her the best agent I know. She also thinks you don’t need to write every day, which makes her even better than the best agent, etc.

• I did a dumb thing on Twitter that some of you might find funny.

Last time, I wrote about Flint and Kindling and it was incredibly nice to hear from people for whom this struck a chord, especially people who understood what it meant to feel about “not finishing things”.

I have to admit that I don’t have that whole thing figured out, only that I finally had the realization that perhaps this is an unhealthy way to think about yourself. I want to make clear that once you have that realization, nothing is fixed. But having the realization at all is probably a big first step to having a more compassionate attitude to yourself. I know it’s hard.

This was a paid-subscriber only episode, the first one of the year. Feel free to forward it to someone who might appreciate it or get something out of it.

A reminder that I’ll be in Washington, D.C. on March 11-13, co-chairing the Code for America Summit and that if you ping me, I probably have a discount code. You should come, if you can, and even if you’re not that into “government” or “civics”. I promise you it’ll be interesting. Also, because I managed to get the fricking Librarian of Congress to be a keynote speaker.

Again, I love getting notes from you. You’re all real people.

Cheers,

Dan