Apple should not be in the business of editorializing their users’ writings. The whole deferral to “well you can just manually type it out” raises the obvious question “then why can’t I swipe it out?”

There were two buttons, one labeled "Our Web Site", the other labeled "Our Competitor's Web Site".

When you moved the mouse over the "Our Competitor's Web Site" button, it would quickly slide out from under your cursor before you could click it!

Then when you stopped moving your mouse, the "Our Web Site" button would slyly slide right underneath your mouse!

Dammit Microsoft!!! ;)

Because then it could look like they're suggesting the word to you, rather than you explicitly selecting it, and they don't want to be accused of suggesting any terms that might be controversial, for the same reason they don't want to be accused of suggesting profanity to their users.

This is how censorship works best, if folks start anticipating and avoiding it so the censors dont even have to exert pressure.

HN is hardly immune from this; it’s no small irony that there will be people doing that very thing right here in these comments.

This is just as wrong as people citing the first amendment in the wrong situations, but in the opposite direction.

Private censorship is censorship. Even if someone declares it "reasonable".

Damned if you do and damned if you don't.

If Apple doesn't want its autocorrect to suggest statistically likely opinions on divisive topics, then sure, whatever. But when it has a hard blacklist of input it will categorically refuse from the user and editorialize it with nonsensical changes that only serve to avoid the blacklist, that is a very, very bad look.

If the user tried to swipe-type "Tibetan independence" and got one or two letters wrong, there's still probably a 90% or even 99% likelihood that their intent was to type "Tibetan independence." Not suggesting "Tibetan independence" is censorship, it's not benign carefulness.

We're not talking about Apple accidentally turning "Buttermilk pancakes" into "Muslim extremist."

  >and "we know what you wanted to type with 90% confidence

Problem is, because statistics, 90% confidence is wrong 10% of the time. Now you have a big polarizing news story about Apple pushing Trump propaganda onto someone who hates Trump.

Damned if you do and damned if you don't, like I said.

But think about it, we're not talking about auto-suggest here. The distinction I made in the parent is critical. Can you imagine what the article headline is like? "I swiped 'duck' into my iPhone. It said 'fuck'." - something like that? It's ludicrous at face value, especially because you can't show this was on Apple in any way. The immediate reaction from the readers, Apple and any random bystander would be "lol the author must have swiped it in really weird". I just can't see any variation of this, even on the most divisive topics, doing any actual damage to Apple. If we assume the best of intentions from Apple, they're still overreacting and purposefully setting arbitrary hard rules about how you can't use their (not your) device.

Much hay was made in some circles that Google was saying George W. Bush's White House was a miserable failure. No tech company wants to be the target of that sort of campaign today. Especially not with a petulant thin skinned narcissist in office that already has demanded protection money to ward off tariffs and other retribution available to the Presidency.

See I tapped out all those characters just for you.

Keyboard swiping involves some prediction work from the system. It is ultimately making a best guess from the keys your finger has passed through. The only truly “accurate” output would be for the keyboard to spell out the gibberish that arises from showing every key that the finger has swiped over. Anything intelligible is a guess.

Now because it is a guess, some may question the system or its makers, “Why did you guess this, and not that?” It is easy to see how this can make the maker of the system the target of a politically motivated group. For example, there is a difference between typing out “Dictator Trump” and my keyboard refusing to spell it out - a clear sign of censorship. However, when swiped, because of the guessing nature of the technology, I can understand that the maker of the system (Apple in this case) would not want to be seen endorsing these views. Hence, they do not offer up these politically charged phrases. We are free to type them out, because it is easy to point out that the intent is entirely from the user. However, it is harder to argue for this from the swiping keys gesture.

In my view, this is a more reasonable explanation.

Which is another way of saying identically ambiguous.

Resolving for a safer option when given a choice isn’t censorship in my view. I can, at the end of the day, type out fuck.

That's the whole point: swiping is a very imprecise method, and the user intent is not clear at all, so Apple avoids making "controversial" guesses.

I fully reject that this makes the situation any better or more justifiable. The job of the software that picks what you have likely just said, swiped or tapped should only concern the given word. If the algorithm decides with 95% certainty that the input was "dictator", that's what it should output. If context is used to better select the candidate word, it should only be done for the purpose of slightly re-weighting the possible options with one transparent algorithm, not for checks against a blacklist.

The reason why I think this is that your line of thinking is unbounded, and there is nothing stopping the next logical move - if Apple can decide 'not to endorse' (representation isn't endorsement, but whatever) views by blacklisting its algorithms from forming certain phrases, why can't they uplink entire paragraphs of your text as you're typing it to dynamically analyze them instead of using a dumb filter, and then decide if the entire text is something they want to allow?

Producing "trump is …" when a user wants that is one thing. Producing that when the user wants something else could be construed as coming from the software itself, and we all know that people looking for a shitstorm will seize the opportunity.

So probabilistic software like a swipe interpreter will often be written to avoid some guesses, to lessen the chance of a shitstorm. Users who want that output have to use a dependable input method instead of one based on probabilities and guesswork.

IMO this avoidance isn't among the ten worst consequences of shitstorms.

The regular key-by-key will also do some probabilistic changes. Just less than the swipe interpreter.

“Male chauvinist pig” is something many would be very upset to send to their boss by accident. So it’s perhaps understandable some such phrases are not excluded from the guesswork.

Step 1: a word of 4-6 letters that's somewhat close to "male". Say "name" or "masked". Look at the key positions, you'll see what I mean.

Step 2: a word that's even less close to chauvinist, and that makes sense after the word you intended, but no sense after "male", so the guesser picks "chauvinist", based on its incorrect first guess and your second swipe.

Step 3: ditto.

We might not talk about a "narrow contingency put" or "Maria crating PVC" very often, but an uncensored autocorrect would almost certainly correct those phrases to the far more common "male chauvinist pig" in the rare cases where someone did try to swipe them.

I know these are all typos because she’s a really close friend and easily the last person who would be offensive to other people.

So yes, you’d be surprised at what people can typo on a phone.

The difference matters enough that preemptively suggesting "pig" is not a desirable "service" for most people.

This would probably be a bad customer experience for many people, even if Apple could prove these are simply the most common n-gram completions.

And if they altered their training data to avoid completions that are politically unpalatable, IMO that's actually much much worse than a grey list.

I'm a big free speech guy but this one is genuinely a bit complex.

Or we could just type the whole sentence we need.

Tech-illiterate online mobs enraged by Apple “making a statement” will not stick around long enough to hear them argue about n-grams or whatever.

Or imagine if it auto-completed "the Holocaust is" with "a myth created by the ZOG"! I don't like these "imagine if X was Y" arguments because X isn't Y. Refusing to auto-complete "the occupied territories" (X) is really not the same thing as refusing to auto-complete antisemitism or group-based hate speech (Y). Argue whether X is right or wrong without changing the subject.

Most people are not against transsexuals and gays.

At least in the western civilization, I don’t see why it need to censor anything.

In the Bay Area?

In the United States?

In the Middle East?

In India and China?

All of planet earth?

What you're really saying, perhaps without realizing it, is "I have zero empathy or theory of mind for people who have different views than me"

The whole reason these arguments exist is because power changes hands. If you like a rule only because the person currently wielding power aligns with you, you don't actually like the rule. You just like power.

- [ ] Disable autocomplete swear filter.

Disable swears, alcohol, smoking, marijuana, gambling, child abuse and neglect, sensitive East Asian politics, sensitive Middle Eastern politics, WWII politics, Late Ottoman genocides, LGB issues, T issues, HOA issues, vegetarian issues, vegan issues...

To which the obvious counterargument is just “well of course they can stop at just one toggle. That’s not an excuse to have zero toggles”.

The team had heard that most customers only use 5% of Word's copious feature set, and as part of some rewrite the team set about finding out what the 5% were.

And it was most of the feature set. Customer A used these 5%, B these others...

It's easy to say just one option without describing exactly what that should do. L10n raises some questions, for example. Or just two options, or three. Much harder if you want to be specific.

The store you mentioned wasn’t about exclusions from each and every dictionary. It was about tools that do entirely unique things.

Do you make the same argument for everything?

"You shouldn't treat yourself to a takeaway because what if you want more takeaways?"

"You shouldn't buy a car because what if you want more cars?"

etc.

Consider the logic for tap’s could offer hundreds of customizations. That’s core functionality and I’m sure someone internally is thinking deeply about many of them. But if you offer zero customization most people just get used to how things operate so skipping that functionality isn’t terrible for the average user. Instead they offer options for people with disabilities which is a distinct group who can’t just adapt and thus a meaningful justification.

But any prediction picked by Apple without user input could be seen as Apple’s speech. If it’s suggested by Apple then they aren’t censoring you, they are censoring themselves.

Talking about autocorrect and irritation, I remember a situation where someone caused a minor international incident by writing that something "caused a lot of irritation". The word they were aiming for before autocorrect chimed in was "iterations" (on a document). I'd rather make my own mistakes than constantly have to fight an autocorrect system.

Of course, people will still accuse Apple of being biased. But they’ll be wrong, and never manage to prove it (unlike this guy), which I think matters.

It'd be a better if Apple provided a list somewhere for transparency but what they are doing isn't going to cause anyone any trouble.

This isn't about "promoting bad spelling"; we aren't talking about "misspelled 'bad' words don't get autocorrected". We're talking about a system where you try to type a "bad" word and the system substitutes a "good" word unless/until you override it. Or a system where you effectively choose every word from a menu, and the "bad" words are hidden using UI "dark patterns".

How does free speech come into it? Suggestions are not limitations (by a private company) on your output.

I'm against Apple's attempt to meld popular speech into a form that it thinks is acceptable on two accounts—I'm not the fastest typist so when prediction doesn't work the way I want it to it slows me down. Second, I'm in Australia and Australians use very colorful expressions in everday speech that seem to trigger the sensibilities of many delicate American ears. In short, we don't want attempts to censor what we say.

If one has to introduce censorship into prediction then users should have the option of either turning it off or selecting a graded system that best suits their style.

In fact that's not a completely new idea. Eudora email had an option that I rather liked, when selected, users could type whatever they liked then depending on the degree of offense or profanity a series of red chilli icons would appear. Mildly offensive text would bring up an icon with a one-chili warning whereas highly offensive language would automatically trigger a three-chili icon along with a message to the effect "They're burning hot words, do you really want to send this message as is?"

A scheme along these lines seems eminently more sensible than secretly censoring a predictive word list.

    > Second, I'm in Australia and Australians use very colorful expressions in everday speech that seem to trigger the sensibilities of many delicate American ears.

> expressions that Americans use in daily life that would offend Aussies?

Civilian regular (daily, weekly) group pledge of allegiance stuff. We're patriotic, but that OTT flag drapery stuff mostly comes across as on the nose.

I don't think any of the wording of that is the issue though?

Also to be clear that's a schoolchildren-only activity.

Sure your client could want to be ultra sure, but I'd argue the censorship coming from a marketing concern doesn't make it specifically better. If I actually use swear words all the time for instance your software would be constantly penalizing me.

That's a tradeoff they're willing to make, but if I was a user it would be totally fair for me to be pissed and hold it against them.

- There is a chance, quantified by some metric, where a person is suggested a word that is offensive to them or their culture because a probabilistic model made the wrong decision.

- There is no chance that the above happens.

Apple will choose the latter, time and time again, because they are risk averse and freedom is not particularly high on their values list. I used to work at Apple.

The ultimate problem being some subset of consumers want autocomplete and some subset will retain the right to be offended and some subset will be both of those?

Because they are in the business of making money and the business case for this doesn't make sense.

If it didn’t auto complete then that would be fine when hitting keys individually, but when I’m swiping the words, it means no matter how accurate I am or how many attempts I make, it will literally never produce that word in that position.

And it’s contextual. Obviously I can swipe in “me,” but if it’s after the word “kill” then it’s banned, and I simply can’t dedicate the space in my head to learning this inscrutable list of banned phrases that until this article I’d been unable to find anywhere so I could only ever discover a phrase was on it by having an extremely frustrating time trying and failing to type it until I realized “oh, this is a politically or emotionally charged phrase (or could be in a different context) and I’m not that bad at typing so it’s probably banned.”

I don’t want to make a mistake calling someone cunt.

Even more so I DON’T want software type out cunt or fuck on its own to a person I totally don’t want to write fucking swear words.

Typed on an iPhone.

It seems like a way to make it sound more dramatic if you dislike it.

You have door 1 and 2 presented to you by me. You're equally likely to want to pick both options. Door 1 is open. Door 2 is locked, but you can ask me to come over and undo all the locks to open it for you.

Run the test over a large number of people, and then shockingly, it turns out that doing this made more people choose door 1, even though they didn't have a preference beforehand. According to your logic, I did not coerce anyone by designing the choice in this obtuse way because after all you can still choose between the two options. But more people ended up choosing the option I wanted them to pick, weird.

If this coercion didn't work, companies wouldn't be hard at work designing revolutionary new types of dark patterns that make you more likely to accept the cookies, back out of deleting your account or decline enabling an option in the settings they don't want you to use anymore.

I imagine it went the same way that board meetings about any potential PR disaster go. "How bad of a reputational hit will we take if the NYT runs a headline about some kid getting auto suggested something culturally taboo on their iPad? Would it be more than the reputational hit we will take from the people that are annoyed that we won't auto suggest 'Nazi' to complete the phrase 'Trump is a' while they're typing out their tweets? If so, then let's add the list of "do not suggest" to our auto complete system." And that doesn't begin to count the bad press they could get if someone intentionally tries to get the auto complete system to generate something offensive. See also the AI companies getting shit because people were able to get the image generators to generate porn.

Such is the nature of trying to sell your products in a world where most people treat every single thing done by their technology as an intentional act chosen by the developers to specifically target them. If you want to buy someone else's package, you have to live with the fact that they are going to respond to the incentives to avoid headline scandals and government scrutiny much more strongly than to cater to a idealistic free speech model. This is why having a strong social and cultural free speech norm, that extends far beyond the legal letter of the law is so important. You want society as a whole to broadly consider free speech important and worth having over the inconvenience of Tiny Tim learning a new word today.

Apparently, names that begin with Di outnumber names that begin with Du four-to-one.

I also feel like it could be better, but this thread seems a little over the top. Censorship because it avoids autocompleting things that might embarrass you?? And as was just posted a couple days ago, if you intentionally use words and phrases, it will learn that you do and will go ahead and suggest them https://helgesver.re/articles/macos-vulgar-word-flags .

Unfortunately the broader more important point of censorship is drowned out in noise as designed

Same with how congress names their bills. Wow you’re not voting for the Save The Children from Internet Monsters act? What’s wrong with you?

Oh you mean the bill that grants the government cart blanche to go through anyone’s email without a warrant?

We’re in a tough spot here - there’s just nothing to pattern match to - not cleanly at least

The way a few tech companies can steer trends and conversations is unprecedented. The data collection is unprecedented

Specifically the scale is. People truly don’t understand the scale these companies operate at

No newspaper or wire tapping ever had the reach this stuff has now

A complete slang language? A world where people don’t have a vocabulary to describe the difference between themselves? A utopia where no one sees race because you can’t type white people?

Probably a society that’s just primed to do what corporations want

Yes yes tinfoil hat and all that

On TikTok for example, OF creators have become ever more indirect in communicating what they do - porn became corn which became the corn emoji, then that degenerated into any mix of base ingredient emojis.

On race, people use dog breeds to substitute for skin colour. In general, golden retrievers are used as a euphemism for white people and pit bulls for black people. The context of a video often seems plausible in a dog context - for example someone might say that pit bulls are 13% of dogs but 50% of dog attacks (actually referring to the African American murder rate per capita) (pit bulls are over represented in most dog injury stats but there’s no reliable figure for 13/50).

The overall effect has been to make the comprehension of the platform less.

There are cases like https://gizmodo.com/a-cellphones-missing-dot-kills-two-peopl... where, in 02008, Ramazan Çalçoban texted his estranged wife in Turkish, "you change the topic every time you run out of arguments", but the lack of the Turkish dotless "i" on his cellphone turned "sıkısınca" into "sikisinca", which looks more like a misspelling of "sikisince". So he ended up texting her, "You change the topic every time they are fucking you."

Long story short, his father-in-law murdered him for thus insulting his daughter's fidelity, following which he murdered the daughter in question, and the father-in-law committed suicide in prison.

So there might be certain words and phrases that it's best for your computers to not send to other people on your behalf unless they're really sure you mean them.

But your guess seems more probable.

I realize that there are many reasons to be critical of our digital overlords. This is not one of them.

This is in no way comparable to a physical keyboard stopping you from typing something, because what you are typing is entirely deterministic. Most computer software will underline only misspelled words or grammatical errors, not politically charged statements, because it is a literal interpretation of what the user typed.

Swipe typing and auto-complete are different because they are interpreting the user's input. There is no way to ensure the interpretation is correct. For the sake of both the users and the company, you ensure the interpretations are the ones that cause no harm.

1. I realize that swipe input is probabilistic as opposed to using a physical device. I just think it should not matter - if it's probabilistic, they should have specific hard rules that are obeyed in all cases. For example, if the software concludes that the input is likely to be this dictionary word with 95% certainty, it should output that word. The context surrounding the word can be used to fine-tune the certainty, but not to cross-check against blacklists of banned phrases.

2. The reason for #1 is that if the CYA approach is allowed, there are no arguments left to prevent Apple from looking at your entire paragraphs of text and then dynamically deciding to discourage certain predictions, to prevent their device from being used to create something they don't agree with.

3. Additionally, because typing on a digital keyboard is also predictive (the touch zones of keys are weighted depending on what it thinks you're likely to type next), this would also allow them to literally make typing certain 'unsafe' things more difficult.

The concern with features like these is that the suggestions are considered to be corporate speech that the company is endorsing, and so it can be held liable for any that break relevant laws. Internet companies are generally considered common carriers when they're just passing around stuff other people write, but when they are generating their own content through machine-learning algorithms they are held to be the actual responsible party.

I'm surprised we haven't seen more lawsuits and criminal prosecution against LLMs so far, though it probably helps that the current U.S. president is a combative asshole that prides himself on how much he can offend every group under the sun.

The LLMs are very obviously trained in such a way that they go along with the user's preferred language when a preference is indicated (unless it would cross certain Overton lines), and default to very "safe" language. This is evident if you try getting assistance with tasks involving creating a blacklist (which they will call a "blocklist" or "denylist"[1] unless you say "blacklist" first).

[1]: Amusingly, my Firefox red-squiggle system doesn't think these are words.

"When testing autocomplete, we found some cases where it produces undesirable outcomes. Imagine if someone typed 'nazis are' and the autocomplete suggested 'good'. How fast would that screenshot get around social media with a tweet comparing iPhone to Grok, and turn into a news headline?"

I'm guessing that that's how this started. Then, a few steps later, "we're getting into China, what do we need to do to not get in trouble", "hey, we've got this mechanism, what if we add 'Tibetan independence' to it". (Which is bad, but on the same scale as all the other things businesses do to get into what they see as a large market that they can be blocked from at any time.)

Then, a few steps later, after some seriously questionable decision-making, someone had the bright idea to add "iMessage is down" to it. Frankly, that's the one I'm actually most baffled by here, because everything else seems like the natural consequence of a combination of bad-PR-avoidance and geopolitical kowtowing.

Don't attribute to malice what can be attributed to stupidity. Most likely it is implemented in such a way to produce the least amount of outrage.

Looking at https://helgesver.re/articles/macos-vulgar-word-flags for the list.

I imagined it something like this:

  "Mao?" -"Bad, we think? But we want to allow 'LMAO'"

  "Next. Stalin?" -"Hmm... Uncle Joe. Only a few million killed. Not great, not terrible, let's keep him"

  "Pol Pot?" -"How many died in the Killing Fields?" "Not enough, keep him off the list".

Most likely from user testing.

I think its pretty easy to imagine. Execs dont want to be dragged in the news when the phone auto suggests something politically controversial. Companies are very risk averse when it comes to that sort of thing.

I doubt this was even controversial enough to warrant a meeting.

“I mistyped bash into my phone and it autocorrected to Nazi. We probably don’t want iPhones to autocorrect towards some words, even if they’re legitimate words. Do y’all think it makes sense to bias autocorrect away from potentially risky subjects?”

You can still type Nazi into your phone. Autocorrect just isn’t going to suggest it to you on accident. Not autocorrecting to potentially embarrassing, uncomfortable, or sensitive subjects is probably a good idea that’s—at the iPhone scale—probably saved quite a few people’s jobs by not fat fingering something that would have gotten someone fired.

No, this is the wrong analogy.

Keyboard helpers are 'positive aid' vs 'no aid', not 'restriction' vs 'no restriction'.

It's akin to seeing the 'default' on search to 'not have adult conent' -> you can make the adjustment yourself.

Helping you type the 'n-word' is a positive act.

It's perfectly reasonable to have any words they even remotely deem sensitive not in the dictionary.

The first time you meaningfully type a word though - it should go in the dictionary like a normal word and that's that.

If they were stopping people from putting words in the dictionary or making it harder - now that would be censorship.

If they won't put 'whatever' in the auto-option feature that would be 'restrictive' and bad.

But otherwise, it's 100% reasonable for them to chose the words they want to help you type.

And FYI this is also different from censoring search which could not be undone.

In an environment (phone touchscreen) where the aid becomes expected (typing is impractically slow/difficult otherwise), this is a distinction without a difference.

Unfortunately a vocal minority of users can be an absolute pain in the ass about this stuff and it's easy enough to just add a "bad words" list to satisfy the pearl clutchers and move on.

I once worked on some enterprise software that generated random alphanumeric passwords for new accounts. One day a customer filed a ticket complaining that her password contained the string "cunt". The tone of the ticket strongly implied that this was a deliberate insult on our part.

I don't know what delusions of grandeur are required to think that a software company has implemented a unique fuck-you-in-particular code path just for you, but it was easy enough to just add the filter.

I hope you also gave the user a stern warning about not revealing details of their password again!