Apple Autocorrects Names of Some Medications to Names of Different Medications
twitter.com
twitter.com
http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_...
https://arstechnica.com/information-technology/2013/08/confu...
It is never right that a computer should do the wrong thing.
It may be explicable, it may be avoidable, it may be correctable but if it is wrong it is wrong. We will be living in a world where computers are increasingly autonomous in their actions. Defending their behaviour now sets a dangerous precedent. Will the same people be blaming victims of self driving cars?
So whoever added drugs to the spell checker didn't think it through properly. They need to be told to take them out or put them all in and keep them updated. If we work around things it is us who will become the servants.
"Re check your work before you click submit or send. Don't place blame on Apple."
"It’s pretty easy for a sysadmin to disable autocorrect"
Educated people aren't immune from this effect either.
It was pretty surreal.
You don’t need to be working at a VC-backed company in the valley to be a tech-bro.
Also, autocorrection does not provide proper feedback to learn from mistakes. Underlining errors does.
As a thought experiment, consider voice input. Voice input is often used when a screen isn't even available for feedback, so the computer has to figure out if it thinks it transcribed you correctly. To make the usability bearable (and to keep the computational difficulty down), contemporary systems don't ask you for clarification; they just try to transcribe your sentence against a dictionary. For the voice-controlled systems of today, "autocorrect" is definitely part of the input mechanism.
That's a nice theory; the problem is that autocorrect is often stunningly bad at that, and turns slight typos into radically different words. Humans are very good at getting the correct sense from text with typos, but not so much with arbitrary replacements.
Voice control is, obviously, a whole different kettle of fish since you don't have a text input stream to start with.
It gets even worse for some non-English languages. The standard way to do autocorrect seems to be roughly "pick the the word I know that has the lowest Levenshtein distance from whatever that was typed". The problem with this is that in, for example, Finnish, every word has a few dozen conjugated forms that have a very short Levenshtein distance to the original word, each of those has a few dozen more conjugated forms that have a short distance to them, etc etc until each Finnish base noun has roughly ~2000 conjugated forms and every verb has ~12000 of them.
No autocorrect system in existence actually knows the rules on how to form those words, and none of them actually have anywhere near all of them stored, either. The result is that the minute you output anything even remotely complex it gets autoincorrected to something else. Despite this, most devices and software ships with autocorrect on. The first thing most Finns do is turn it off.
The irony of it is that even if you had perfect autocorrect, it really would not be that helpful near the endings of words, simply because the possibility space is so full of correct words that a typo has a good chance of hitting one.
I'm pretty sure that in fact it is frequency weighted. You can see that when a correct but rare word is replaced by a much more frequent word.
That assumes that you leave the typos in place rather than fixing them. If your goal is to type error-free text, then which one leads to the correct text faster? Autocorrect and then fixing the occasional word it gets wrong, or trying to edit away typos on a tiny screen?
I notice this kind of problem a lot with dictation, where it can completely transform the look of a word, and I sometimes have to say it out loud to try and figure out what I might have meant that sounds similar.
I've noticed this on Windows Phone - on WP8, the background colour of the suggested word briefly changes on an automatic correction, so it's easy to notice and double-check it's done the right thing. On WP10, it only gets slightly bolded and only until you hit space. I've had multiple emails where I've not noticed a erroneous correction until I'm later re-reading it for whatever reason.
Android's approach is to take confidence in the correction into account: if it's extremely sure it'll auto-correct, if it's less sure it'll just offer. And either way, the UI makes it clear when either of those things has happened.
"if" or "is" to I; "ran" to "Ra"; "nan" to "Na"; "drew" to "are"; "near" to "new"
I've also had a couple of occasions where trying to type a specific letter has resulted in the letter to either side (e.g. "o" or "p" when trying to type "i") and it's taken multiple tries to hit the right letter because the device thinks it knows better and has (presumably) made the touch target so tiny (and I know it's not a hardware/screen issue, because I've seen the keyboard display that you've hit a certain key then input something different)
I don't know why people hate autocorrect so much. Yes it's annoying. But trying to use a mobile "keyboard" without it is far more annoying. For every word it gets wrong, it gets 10 more right. Maybe more. And saves me a ton of time on average.
It should definitely be considered part of the input process. Certain kinds of mobile keyboads, like the swyping based ones, and the ones on tiny smart watches, would literally be unusable without autocorrect.
If you know how they work, I think they are pretty elegant. Taking a probability distribution over all the letters you could have been aiming for and missed, and all the words you could have been trying to type, and the probability of each in that context.
Touch-screen typing is an entirely different interface than a physical keyboard.
Also, the post is about a macbook, which has a keyboard.
Nowadays, macOS, Windows, and Chrome OS all come with built-in spellcheck, with various levels of default autocorrect.
The goal is to never get 'corrected' without my assent, while still going fast. This might be underexplored, because designers aim to keep it easy for everyone without training. (But this is also likely just a bad idea. I mostly just don't type on phones and curse them when I do.)
This was especially nice for me as my natural finger hits about 1/8 of an inch to the right of where I think it should. Is there any way to calibrate the touch position on android to fix this problem? Google fails me and, if this is not possible, it seems the tech industry has also. I have Android 8.0.0.
And you can't just not have those names in there by default. They have to be either blacklisted or installed at once, and that list must be updated regularly indefinitely. Adding a few choice medication names means those words become candidates for a number of other 'misspellings', so you need ALL the possible drug names to prevent correction from another (correct) drug to the newly added (incorrect) one.
Of course the dark path behind this is that a drug company can pay keyboard/spell check makers an incentive to autocorrect to their drugs. You typed "Lipitro". Did you mean "Zocor"?
So now my user experience with Excel is going to Options -> Proofing -> AutoCorrect Options -> [Find HSA on the long list] -> Delete. I took the opportunity to delete them all.
Honestly if this software was written today I don't believe Auto-Correct would be implemented the way it is, it is super lazy with no room for false positives.
EDIT: It's actually the Hang Seng Index, not HK Stock Index, but both expansions capture the meaning.
The more languages you need to type, the worse it becomes.
On my laptop though--I usually disable the autocorrect and instead limit it to underlining misspellings.
My emphasis. I dispute this. I use Vim and I have set it up to automatically correct common spelling mistakes I make. It's quite useful and I can see no downside to it.
Ironically this would often piss off the patient who was waiting for the prescription!
The pharmacist just ignored the prescription because it made no sense for the symptoms, and the dose was correct for the other medication. So it was obvious to him what she had to take.
If a pharmacy tech did this it would be scary. But an actual pharmacist?
It's no scarier (or, alternatively, as scary as) a doctor listening to your heartbeat and measure your blood pressure and choosing which medication to prescribe.
Because pharmacist are not licensed to prescribe blood pressure medication and they don't have access to your full medical history so they don't know if the medication is appropriate for you.
I'd argue that changing the drug without calling the doctor (to at least let them know there was a mistake e.g. pull-down error by the nurse assistant) isn't a good idea, but it's VERY common to have them fill with an alternative/generic when the prescribed one isn't in stock. Usually, they'll mention it.
Honestly, there's a reason that pharmacists have such strict education and licensing. They regularly have to decide exactly these things based on this limited information. It's their job (often along with mixing/compounding drugs).
Not only that but many medications are not for treating symptoms but are prophylactic. Especially post surgical ones.
I believe that short term prn medicine may contain some language like that.
For anyone else,
The VAST majority of prescription medication does not contain such language. Post surgical medication are like this: "take once a day" not something like "take this antiepileptic drug once a day to prevent seizures after brain surgery." Which is the actual purpose of the medication.
I talk from experience.
Any type of ongoing maintenance drug is going to omit any sort of symptoms, assuming any are present in the first place.
Again, experience.
Sure, this may be your experience with filling 5 or less prn prescriptions but this isn't how the majority of scripts work.
Even if the pharmacist had your FULL medical history it seems very unusual they'd be able to catch this sort of error.
I wonder how people would react to an alert box "I've noticed you have terrible spelling. I may be able to help. Do you want me to auto-correct always and assume the small but real risk of me using the wrong word changing the meaning of what you wrote?" (yes | no | cry)
I wonder how many common mistakes are due to autocorrect failing to do its (its, NOT it's - like my phone just "autocorrected" me even when I typed i-t-s) job.
It seems autocorrect developers and models fail to model anything beyond the basic "word exists or not" in the dictionary (and even then it gets some things wrong).
Clinics are more likely to enter the name of a medication in some application rather than a plain text editor.
On my 10.10.5 Macbook Air I tested quick and the issue is simply that duloxetine is not part of the dictionary and fluoxetine is. Even as I write this post I have the red line under duloxetine.
I'm not sure how the association was made on their particular computer, but it raises a good point about how the annoyances of AutoCorrect can cause some difficulty. My last name, for example, is mostly unique, but is one letter off from a type of fabric. It was mildly irksome until I had the dictionary learn my name. (At least one goof up at an airport after buying auto-corrected tickets)
In this case, there is of course a simple fix, which the author should apply, and if being deployed in a work environment where the change is important, there are steps that probably should be taken to resolve such issues, and better visible custom dictionary files. (looking on various *.exchange site when searching for "bulk add words to macOS dictionary provides many solutions)
So I agree, that it's a problem, but it's not intentional malice here on Apple, workarounds are available, and it's a really good feature request that ought be a pretty low hanging fruit.
It seems like there should also be a medical regulation that input devices used in a medical setting should have autocorrect turned off. It's useful for e-mailing your mother, but obviously not in medicine.
There are all sorts of regulations licensed medical practices need to follow, or which medical devices have to follow, which the rest of us don't have to. (Similar to aviation, for example.) Sounds like this is a good candidate for another, no?
It could autocorrect minor mistakes but not whole words
Seriously, proving loss due to something like this should be not too difficult. There is money to be made here - with the ancillary end result being Apple loses, what little position it has, in the point-of-care market.
Once I see typing correction that’s better than useless, then I’ll know to look out for driverless cars outside of the Bay Area..
A very large percentage of driver have no business driving, they will give a driver's license to anyone with a pulse and even people who have no driver's licenses are all over the road. They don't care, drive drunk, tired, half blind, with medical conditions, or otherwise seriously impaired. People would routinely come through my drive thru drunk and stoned, slurring, and barely comprehensible when I worked at BK. You haven't seen dangerous until you've seen someone with bipolar disorder drive while manic.
Youre talking about a trained professional, the majority of randos aren't randomly spell checking documents.
So self driving cars have to be better than your average driver not a professional driver. Which, frankly, is a pretty low bar.
The names only sound confusingly similar to people unfamiliar with them - for physicians and pharmacists, it's rare to confuse them.
The similarity in the names is actually intentional and a feature. For example, erythromycin and azithromycin have the same suffix and similar prefixes, which is an indicator that, not only do they belong to the same class, but one is a derivative of the other.
It's been a while since I studied this, but based on the names alone, I would be willing to bet that Escitalopram is an S-enantiomer of Citalopram - ie, a mirror image of the drug (or an isolation of one of the two mirror images present in the latter). If you say it out loud, it's almost a mnemonic pun.
I could also guess that's an SSRI based on the "pram" suffix, though admittedly you made that easier by mentioning it was an antidepressant in the first place.
INN names (like citalopram) are actually incredibly well-designed for the intended user (a pharmacist or physician, not a patient). I wish engineering terms had the same level of consistency.
Edit: Turns out I was right. Both are SSRIs, and the former is the S-enantiomer of the latter.
The racemic mix usually has a slightly higher level of side effects.
Possibly: the process to manufacture the molecules is also patented, and you must make the racemic mix to extract the active enantiomer. Not sure though.
Not exactly. You're right that escitalopram is the S-enantiomer; citalopram is the racemic mixture of R- and S-enantiomers (of which only S- has the antidepressant effect).
Of course, not everyone actually wants autocorrect. For me it's about 50% helpful and 50% a hindrance.
What could also work is having multiple hooks (different "niche" dictionaries) which could be enabled or disabled.
In SwiftKey you can select multiple languages (dictionaries) which should be supported (and corrected, if you opt for that option).
But what I mean is that say a user could enable a medication dictionary, or a UNIX manual dictionary, or a law dictionary, or...
https://druginfo.nlm.nih.gov/drugportal/, for example.
To rely on your phone for autocorrecting difficult medication brands and then blame the phone manufacturer when it doesn't work is really a 1st world problem.
I'm not saying that Apple is unable to remedy this situation. I'm just saying the users are unrealistic in belieiving that Apple has thought of this situation beforehand and made any changes to aide in the autocorrecting of medicine brands.
Because they are brand names. And there are thousands of them.
Apple has been in the business of automating daily tasks on the iphone for a decade now, holding them to account when it comes to when autocorrect could lead to a person's death, and when it is trivially innocent, is at this point perfectly fair.
There are thousands of brands, so having a cherry-picked list of just a few of them preinstalled in the dictionary is something anyone who specialises in autocorrect (say: someone who's worked on it for 10 years and has run into virtually every edge case loads of times by now) knows you should not be doing.
Instead of having autocorrect kick in at all, it should simply go "I do not know this word", so that the blame does lie with the user when they tell the phone to add their medication to the dictionary.
"Correcting" a correct word to a different one can _solely_ be blamed on Apple - the user might not even know that the OS does autocorrections, so can hardly be blamed for the system silently changing something they typed correctly (at most you can say they should proof-read the statement). It's an understandable issue, but an issue nonetheless. Especially on desktop, it seems like an easy problem to avoid - just prompt the user that there may be a mistake rather than changing it without their knowledge.
In other words, this is 100% a fault in the hands of the users in the supposed problem case.
Now, all that said, I'd not be sad at all if this resulted in Apple disabling this feature by default. I always have to disable it on every mac as it drives me nuts. I type quickly so I never see the autocorrection popup.
How is the specific case of autocorrecting "duloxetine" to "fluoxetine" the system "working as designed"?
> this is 100% a fault in the hands of the users in the supposed problem case.
Why is a case like autocorrecting "duloxetine" to "fluoxetine" absolutely not in any way a fault of the developers?
Would an autocorrect system with the smarts to avoid that problem not be better?
.
You seem to be thinking of this in terms of absolutes. One point is that autocorrect, like everything else, has costs and benefits. Just because it has flaws doesn't mean it hasn't got benefits. For me the benefits outweigh the costs. Another is that when flaws are found they can be fixed
I didn't know auto-correct was enabled by default on macOS, and would never have noticed that the OS was changing what I was typing without notifying me.
This was an annoyance, and for a long time I was just super annoyed with it until I decided "there has to be a way to fix this", and there was. It was a simple double-tap on the trackpad "Learn Spelling", and it was done. I type pretty quickly too, but I'm also pretty persistent in watching what I'm doing and catch the lines as they appear.
I the case of the article, there's an association now on the Mac between the two words, which I'm pretty sure that "Learn spelling" will override. But by default if it hits a word it doesn't know and doesn't know what it could possibly be, it just leaves it as a red line.
I always disable autocorrect on all my devices.