Feature Friday: Gmail Predictive Typing
I was typing emails on my new Pixelbook yesterday and as I started a new word, Gmail would finish the word and often suggest an entire phrase to come after the word. All I had to do is hit tab to accept the suggestion.
I’ve had a version of that on the Gmail app on my phone for a while. It looks like this:
But the type assistance functionality on my Pixelbook was much more useful. First it was in line and it suggested entire phrases, not just words.
I’m of two minds on this feature:
1/ It is super useful and once I get used to it, I should be able to construct messages much more rapidly.
2/ I am not sure I want Google reverting my communication style (and yours too) to the mean.
I suppose if Google was using just its understanding of my writing style and nobody else’s in its algorithms, then it is helping me be me. And I like that a lot more.
But if it helping me be you, well I’m not as excited about that.
If Disqus does it, all comments will be the same : )
If Disqus does it, all comments will be the same : )
and because all posts will be the same?
You can say that again.(I know what comes next.)
My Pixel 2 does this in all apps. It’s the keyboard app, not just Gmail, as far as I can tell.
suggesting words or entire sentences with many words?
Phrases of 2 to 5 words.
I’m loving it on mobile, as it is a time efficiency boost. You don’t have to take these suggestions, but only take ones you like, so I don’t agree with the sentiment that it would alter anyone’s writing style necessarily. In my experience, it was most useful in ending sentences, more on the neutral side of style changes.
I’m having a similar response to yours. Concern for the homogenizing effect it will have on my and others’ writings. A byproduct is that it’s making me examine stock phrases I use and prompting me to avoid even my own ‘go to’ verbiage. “I’ll do the thinnin around here Baba Looey”.
Interesting societal questions if it normalizes communication style but pulls up the mean. Still can be argued net bad.
It depends of what side of the mean you think you are.One important thing is that change is usually initiated by outliers which can be anywhere.
>It depends of what side of the mean you think you are.Reminds me of the joke about statisticians:A statistician is a person who can have his head in the freezer and feet in the oven and say, “on the average, I feel quite comfortable”.
I would not say that normalizing communication style is good in and of itself. Depends on cases.
Sometimes the predictive word on my iPhone makes me think of another way to say something, with the suggestion a reminder of words I may use too often. I’m always trying to keep my writing fresh and not fall into a rut!
…ahh…but if it’s helping me be you, well then I’m *very* excited about that! Cause you’re da bomb… 😉
I use it on my phone.Amazingly effective.I seriously don’t like the standardization of language in any form and frankly it bugs me and may stop using it.Feels like a linguistic matrix a bit.
Isn’t Grammarly in the same realm? i.e. highlighting sentences that are against a certain mould and making all users tilt towards using a format of the mean?
Being us might be fun, we are not that boring. 🙂
This is not for me at all. It’s an interruption to the thought process and not an aid and time saver (like spell check).I will explain with what I call ‘the rule of scanning’. That is, it’s easier to just scan something than it is to decide if you should scan it. So I scan just about everything.Along those lines it’s easier for me to just type what I want to say than to decide if some assistant is saying the right thing on my behalf. I don’t like the existing ‘finish the word’ and I definitely don’t like the ‘suggest an entire phrase’.  Also at what point will it keep track of whether you use the same wording for the same people? In theory it could but it probably will not.
shut it off immediately, muttered something along the lines of “fuckin’ google.”
I think this will also lead to a fair amount of ‘fat finger’ issues because of the lack of friction. I can even see someone using it as a defense of something that they said that they then claim later not to have meant.And even in the attached example from this AM the choices are biased with 2 ways of saying ‘yes’, but only 1 way of saying ‘no’.Having to type things out makes you think and a habit of doing this for unimportant things could lead to doing it wrongly with something important.  Writing out a reply that is less likely to happen. Red light theory. If you start to run red lights when it doesn’t matter your brain will then do it when it does matter because you have trained your brain to not respect a red light as absolute and as a background control (my theory)…. https://uploads.disquscdn.c…
The same thing happens when people who like to talk too much complain about not being heard. The brain adapts, filters, arranges reality for us and we don’t notice.I have a story about this which I like to tell. When flat monitors arrived at the office ages ago I went into the CEO’s office to see one. The screen wasn’t flat, it was slightly concave, that is how I perceived it. It wasn’t an optical illusion either, I was seeing the flat surface as concave to the point that I had to touch it with my fingers to feel that it was flat. Looked again, and again. It wasn’t flat.It impacted me. I understood that the brain was correcting for the convexity of my CRT screen, the one I watched 10 hours a day. It was doing that for my convenience, it was modifying reality for my comfort. It worried me. How many other things is my brain doing for me that I don’t realize? How could this affect my judgment to the point that I fail to perceive reality?I always remember. Now I am aware.
One thing I will add to this (and I agree). When the brain doesn’t get what it expects it often will reject what it sees. If it can’t correctI have found this to be the case in the printing business. If you use a different ink color even if it’s nicer, people will often reject if it’s unexpected and not asked for. Why? Because it’s different. Then after some time they adapt and they often like it even more. And I don’t mean in a way like the obvious ‘you made a mistake’. I mean if you even maybe say it’s free.I found this happened in at least two cases currently that others can relate to. One was when Fred changed the size of the type on AVC.com years ago. I immediately didn’t like it. I even said that. I then adapted and now I like it. The ‘old’ font size probably would bother me. It is much smaller. I like the larger size now.Ditto for the gmail upgrade we all got. At first it really bugged me just because it was different. Now I like it. My brain adapted but it took time.In the case of what you experienced your brain (it sounds like) was able to make it be the way it wanted it to be with a curve. Good example by the way.
>Ditto for the gmail upgrade we all got. At first it really bugged me just because it was different. Now I like it.Interesting. Is there anything specific about the upgrade that you like, or is it just a general feeling?I myself am not liking it, so far at least, although I’ve only used it a little until now. I’m managing by using the HTML UI, which works for many use cases, and can be faster than the standard UI too, for some cases. It has a few limitations, e.g. I don’t know of a way to create a new label with the HTML UI (may not have checked enough).
Well I should correct and say that my brain likes it as much as I like the old UI which isn’t that much. There is still a great deal that for me on desktop isn’t great. I am not a fan of the particular font choices they use either.The snooze feature is nice but I have an existing system for that. I would like to have something though that does this:You can reply to an email but send it later. When you send it later it will not be sent if you got another email from the same person (because your reply might not make sense).Or, a better way to do drafts and then hold for later sending (other than drafts).
Interesting on the last two points, about send later and other way of doing drafts.Agree the first would be useful, because the earlier-written reply may not now make sense. But in that case, preferably, the software should also tell you that the message was not sent, and the reason (that you got another email from the same person, and even warn you to consider re-writing that earlier-written email in light of the new reply, or to scrap it and write a new one from scratch. Now that would be (one small example of) some real value addition by software, rather than the usual glorified typewriter / faster-calculator-of-payroll-than-a-human sort of benefits).Not sure what you have in mind for the latter. I mean, I get the idea that you want something different – likely in implementation, but would that really matter, if the usage or behavior is the same? Not quite clear.
Well let’s say you want to manage expectations. So you don’t want to as a habit reply right away because then if you don’t reply right away next time that would be interpreted as perhaps a signal that you don’t want to send and be interpreted that there is a reason for the delay. So you want to write a reply and then have it sent later.As a way to explain lets use my bill paying method in the 80’s when cash was short in a business. I didn’t want to pay everyone in 15 days even if I had the money. Why? Because then I had no leeway. If I then paid a vendor in 45 ot 60 days they would think ‘hmm I wonder if something is up with their cash flow?’. So it made more sense to pay people in 45 days which can be stretched to 60. Those numbers are made up to illustrate the concept (not real or what was done). So we are trying to remove the ‘delta’ in behavior and use a time period that is gives us more breathing room basically.In negotiation I have been able to use the time to reply to loosely infer some things. For example if I reply back to a broker with ‘what about $x’ and he writes back right away ‘ok that is fine’ I can assume he already knows his client will do that. So I then know money has been left on the table. On the other hand if he takes longer to get back to me then I can maybe assume that he needs to check with his client. Now that in itself isn’t that helpful but combined with other things I am watching might be. Also you can tell by reply speed if someone (under a false identity) is a decision maker or part of a team. None of this is 100% it’s like counting cards (I don’t play cards but know the concept of improving odds). Want to be clear about that.
Yes, such things can improve the odds. I tend to do things that improve odds of outcomes too.
Again we disagree on a business method. I always paid super fast. That way if I ran into trouble I could say, oh I didn’t see it, etc. I also could demand tough terms and extra service. I do it with my contractors. The day they are done they get the money. Reputation is a big thing and it spreads.It also imposes discipline. I have seen where people go late, and then get C.O.D terms imposed and it is a huge whipsaw.Aeropostale was a poster child. Better to deal with it right away versus financing your business with somebody elses.I purposefully give worse service to those that pay late.
I am talking about in the 80’s and when I was self financed and it was my money (and obviously there were no investors). And at that time of course for a contractor or small vendor it wasn’t the same behavior. Those were paid on the spot obviously. But a company like VWR Scientific? Lindenmeyr Paper? They have an A/P depart and the salesman has nothing to do with that and already thinks we are a great account? Not the same.And with collecting money from customers it wasn’t even when they paid honestly many times. It was whether we felt they would pay. For example the City of Philadelphia would take forever to pay. But we knew the money would come it was guaranteed. No sleep lost over that account. Or the GPO (the Government). Set and forget it will arrive. Or a big Pharma like Smith Kline French (at that time) who cares when they pay? They are a big company with an AP department that pays their bills. They money would come no lost sleep.Today it’s a bit different. Bills are paid early and cash flow is definitely not an issue and won’t be. So if I get a bill it is paid quickly and without respect to when it is even due. That is just because it’s easier that way. That goes for everyone big or small.I purposefully give worse service to those that pay late.Why? It would make sense to charge a higher price rather than give worse service.
Good points. To me bad service includes higher price and charging for everything, not just letting it go. You see this even from non owners. I got it last night from Cabellas, they forgot to give me a key, I went back, and they gave me a small item for my inconvenience.Lindenmeyer, that brings back memories, they always had the high end paper. There was a family member that was a Jeweler. She would go to NYC diamond district and then there was a smith in Germany that would make them, my friends from NYC coveted those rings and I got many visit in DE.
That ‘freebie’ for correcting a wrong though is a slippery slope. It works like my ‘run toward the attacking dog’ works. Only a few times. After that it becomes the normal and it’s expected. Early in business I bought pizza for employees when they came in on Saturday to work. I thought ‘yeah that’s what you do!!!’. Well it only worked the first time. By the 2nd Saturday they expected it and wore off. So I stopped doing it.Funny at Whole Foods prior to Amazon takeover I would get special treatment from the seafood guys. If they had only the 8oz lox not the 4oz lox package they would either mark down the 8oz or give it to me for free a few times. Post Amazon it’s much different. You can tell the attitude has changed since Jeff took over. Not that he said ‘don’t do that’ but just they aren’t feeling as happy. One guy at the counter who was great actually has a bad attitude. I think they are short on labor and Jeff is just saying ‘make due with less’.This is similar to a “Jeff” (was his name) who worked for me as a manager. He never did any of the jobs in the place (I did because I had no employees when I started) so he didn’t know what it was like to be short handed at the counter or have a machine that didn’t work right and have pressure. That is like “Jeff”. He has never worked the seafood counter when there are no others to help. You can tell this guy is not feeling great about his job. That said Whole Foods is definitely doing better. I think. Most finance guys are like this. They haven’t had feet on the street and they manage by numbers without understanding the actual feel of things. Marcus Lemonis is like this. It’s all about the number and the margin.
Interesting dynamic. There are people like me if you push for free shit will charge you more, but gladly give you if you don’t ask.There are others that will “grease the squeaky wheel”So it is the classic four square dynamic.Push me = I screw youPush me = I reward youNot push me = I reward youNot push me = I screw you
You need to watch this show, Brain Games: https://www.netflix.com/tit…I worry about culture when you do this. I.e. mono-culture. Here was a diatribe of mine about it this morning:You know I think one of the biggest cultural differences we have in the Delaware office is how we treat mistakes and give credit.I mean think about this: We have four people who’s sole purpose, entire job description is to find mistakes.People that don’t know software will say well that means you have bad programmers. That could be a whole page of diatribe.The only person we recently fired (after only working two weeks), came from a good place, could talk the talk, but didn’t know what they were doing (my only job interviewing is to say here is who we are and what we are, we are a horrible place if you like meetings not coding). We would have invested time in training but he hid or lied about what he knew and mistakes he made. For us we can’t do that.For us you will see people go over to database, ops, or development area and say hey who was working on X? It’s expected you stand and raise your hand. That is just how we do it. Self reinforcing. There is no shame. The only shame is not owning it.We have to do this. We cannot survive if we were playing blame games. Oh it’s slow due to ops, or database didn’t structure that index right, or core coding made a screwup, or U/X was bad. We wouldn’t get anything done. We’d be down.We win as a team and we lose as a team. There is no: Oh this person sold this, or did that.
The Google SRE (Site Reliability Engineering) team has a no-blame culture too, I’ve read. Rather, their focus is on finding the root cause of incidents, and better, developing systems (including automated ones, via software) to prevent those types of incidents from happening again. (This is based on what I’ve read about them, don’t have internal knowledge of it.)(Sorry about the abuse of the word “culture”, which I think is heavily overloaded, just couldn’t think of a better one off the top of my head. It’s like the abuse of the word “philosophy”.)
No in this case it is definitely culture.
I completely agree. Sadly, blame games are so common in organizations. Also internal power struggles between teams or divisions. Those are a complete waste of energy and focus.On the other hand I think that relying exclusively on trust in a software/ops shop maybe not an optimal solution either. I am trying to sell this concept to a client these days: We should see each other all the time, we should be transparent and measure everything. If I change some code or configuration in X you should be notified and be able to see those changes without asking me. A brief look at your dashboard and some drilling will show you quickly which components of X were touched and when, down to the code or config level if you wish. You should be able to see the benchmark results after the changes were made. You should be able to follow a deploy process remotely. Hmm, this new database index decreased overall performance by 4%. You should be able to comment and hyperlink into the code.We have some pieces of it in open source tools, SCM, CI, etc.. but I want more. More integration and less complexity. If a tool is painful to use I think it doesn’t serve its purpose. Many organizations have artifacts installed only for compliance. I’m thinking about a solution.
Let me be very clear. I strongly believe in compartmentalizing permissions and checks and balances.Ops, Dev, Database, and Support are not able to make any changes without one of the other’s sign in. And then another third party needs to be informed.This is not to slow things down but to have checks and balances for security.
>Having to type things out makes you think and a habit of doing this for unimportant things could lead to doing it wrongly with something important.  Writing out a reply that is less likely to happen.> makes you thinkGreat point. Ties into what I said in another comment in this post, about “the dumbing down of everything”. That seems to be happening apace, in some fields, at least.Reminded of a book I had read years ago, by Abbé  Dimnet, called The Art of Thinking. Abbot, Priest.https://www.google.co.in/se…I remember one quote by him (IIRC) from the book. It is not about thinking, so much, it is more related to the dumbing-down that I talked about. He says somewhere in the book, while talking about the trend of simplifying everything (which existed even then, and which is sometimes the same as dumbing things down):”French grammar cannot be made simple. It can be made clear.”See:https://en.wikipedia.org/wi…(including the links at the bottom of that Wikipedia page about him, for links to some of his books online), like:https://www.amazon.com/Art-…Internet Archive search for his books:https://archive.org/search….French Grammar Made Clear (a book by him):https://archive.org/details…
This is a world class quote, “But if it helping me be you, well I’m not as excited about that.”I turned off predictive typing, because it was like having a not very good private chef doing my cooking for me.But I admit to using the little autoreplies in gmail. And I feel badly every single time.
I find the feature prompting me in ways I didn’t expect: I kind of resent its predictions (am I so predictable?) so when I see a proposed word or phrase I ask myself, “am I giving this reply the attention it deserves?” And if I am, I then ask myself, “is there a way to express my reply that indicates that I really have thought through what I am saying, or will it be experienced as rote?”
the pro argument is that this and other such things allow the mind to focus on more important decisions instead of the “commodity decisions” of what elementary phrase to use. AI advocates will cite this development in all fields. for instance, much of web UI has been standardized, largely based on collective understanding of what works based on shared analytics. even in sports, everyone now understands the importance of getting open three point shooting, so the commoditization of that decision allows for greater focus/creativity/human-led innovation on creating spacing.the hater argument is the similarity of everything.i think both sides are compelling and have truth but tend to side with the pro argument in the long run. i think it helps to be aware of what is being commoditized and where the extra value for human intelligence will go to.
I’ve learned to embrace this feature regardless. It’s usually a……wholesale price list for the ‘Causes’ and the apartment on Craigslist for everyone.LOL! I let Gmail finish the last sentence after, “It’s usually a….”Gmail ran with it and this is where ‘type assistance’ went. More entertaining than what I had. May do that more often.
There was a great radiolab episode recently that explored how it’s becoming easier for a bot to pass the Turing test, not just because AI is getting better, but because these auto-complete apps are changing and normalizing the way we write. https://www.wnycstudios.org…
The word according to Google.Martin Luther would know what to do. Start up a new religion.
To my other point what happens when this is used by non english native speakers. Attached is an example of what can go wrong. The algorithm isn’t programmed to prioritize ‘do the least harm’. If it had it would never suggest an answer of ‘glad to hear it’ to an email that contained ‘broke my ankle last night’ in a similar google auto reply on gmail. Imagine if death were part of the email as well as something positive…. https://uploads.disquscdn.c…
And yet, with PowerPoint, we let MSFT willfully revert our Presentation styles to the mean. 🙂
Which is why I don’t allow it.You put down the meeting info in non formatted word document, you justify who needs to be there in the document as well. You sit in silence for 15 minutes reading, thinking, and putting down notes. You discuss for a max of 30 minutes. You write up followups for the last 15min.The only others we have are lunch and learn. Where you bring your lunch or we buy and you learn.
I don’t do any powerpoints (and have no need to use that tool). One thing thought about ‘crutches’ like that. They are a way to not only remind your self of what you have to say but also to take the focus off you and put it on the screen. So merely using one will make someone doing a presentation more relaxed. And that relaxation could be the key to doing a decent job.This is similar to an old trick I figured out with cold calling. You show up at the door (or the receptionist desk) and hand something (anything) to the person. What do they do? They squint and focus on the thing you gave them. The ‘Prop’. Not at you. This in itself can be helpful in many ways and also it can make the prospect more comfortable as well.The other ‘prop’ that I have used is having my high school girlfriend come with me. It tends to make you not seem as dangerous or threatening.
Here is the most insidious thing about powerpoint:People look at the slide and no different than any other sign they believe it to be fact. That is what you are saying.I’m serious. There are people that if I put in a powerpoint I was going to change the color of the sky, they’d bobble head shake their head and say: “well it says that it must be true”Small minds, small minds.I’ve had people freak out when I did a two second change to an excel spreadsheet with projections……no change it back change it back!!! What was that version we can’t just change it!!!Unless you are describing history with defined facts, neither tool does anything other than give small minds the peace of legitimacy, which is what their small minds want.
they’d bobble head shake their head and say: “well it says that it must be true”I actually have a saying for that I call it a ‘Nettie Nod’. Nettie is a relative of mine. She has a habit of nodding when she understand and agrees. We look for ‘nettie nods’ when in an audience as they provide positive reinforcement.The opposite of a ‘nettie nod’ (that is not her name btw but it’s similar) is what I will call the ‘reflexive ‘no’. That is when you are sitting across from the car salesman and he starts to tell you the price and before he can even finish you start to send him a subliminal signal that says ‘I don’t like and don’t agree to the price’. The signal is shaking your head in a way very slowly that says ‘no’ or disagreement.The ‘nettie nod’ is also similar to the ‘halo nod’. For example someone like Seth Godin stops by AVC and makes a single comment (with no follow up later). And gets many upvotes. Now if anyone else made the same exact comment it most certainly wouldn’t get the upvotes. Of course I could argue that more people read it if Seth writes it but I think the halo is equally strong and more the reason. Your powerpoint should include clearly wrong data just to see if anyone is paying attention. For example I read stories and kind of do ballpark math on the numbers to see if they make sense. These are all things that disqus could have studied and talked about in order to draw more attention to comments and psychology. Perhaps with a byline by a Ph.d so everything thinks it’s accurate.
>Your powerpoint should include clearly wrong data just to see if anyone is paying attention.In high school, I had a chemistry teacher who used to do something like that. He would randomly say some small wrong fact during his lecture, to see if the class was listening or sleeping.Another example: In a company I worked, soon after I joined, there was a meeting for something. I think it was a short induction training or something. As part of it, there was a small objective test that everyone had to do. One question, call it #5, referred (clearly, but not with any extra emphasis) to a point in, let’s say, question #12 later in the test. And the instruction (on what to do) given in #11 contradicted that in #5.So anyone who just read #5 and then answered it (without reading #11, even though it said to do so), would get #5 wrong.Almost everyone missed that, because of doing things in haste. They just read #5 and acted on it. Battles have been lost due to such errors.
You know I have never tried that. I think that because of modern eduction (teaching for the test) it has gotten worse.
I’m at ‘Agitation Stage’ with this feature, reserving rights to call out whether it’s ‘useful’ or ‘homogenistic.’ I have no idea if homogenistic is a word.I relate to the idea I’d like to be me, not ‘predictive’ (or predicable).If we analogize what’s going on here to auto-fill via Google Search queries, that feature seeks to coach us to ‘say the same things’ (aka: ‘homogenize search’), as that’s in Google’s interest, but it’s another thing to homogenize email prose.Uber’s app does a nice job of this. ‘Be right there’ is a great, single-button click I use all the time. Gmail also offers pre-written replies that I use for short replies.AVC has previously celebrated ‘single-button replies’ (e.g., ‘Like’ on Twitter), but this takes that farther. My sense is smart folks will continue to discern authentic voice from homogenized algorithms, and writers’ jobs are to make that distinction clear.Akin to templated, outbound email, I suspect we’ll continue to write (really write) what matters, while ‘predictively-typed email’ will someday seem as inauthentic as an HTM email.Those that take shortcuts will be known by those that pay attention.’Forgive any typos, send via iPhone’ is at least as bad as the above.But The Republic survives, homogenistically as it ever was.
It’s done with Federated Learning.You get a personalised model but contribute to the overall model that makes everyones experience better.https://www.youtube.com/wat…
Google and Apple both use federated learning for the predictive text. It’s a great problem to do machine learning on as each person is creating an epic labelled dataset as you either choose one of the predicted words or type the correct answer.You have a local model that is personalised to you and your labelling gets contributed to the global dataset. The challenge with federated learning is your sending the model to the device, so a bad actor could steal your model, so you need to send an encrypted model to run on the device.This has the benefit of reducing the amount of data that has to be transferred over the internet, which will become really important when machine learning gets applied to networks of hundreds of thousands of hospitals data that have exabytes of data.
It helps immensely in business communication where:a)I have a lot of clients so a lot of email and b) I’m less concerned with being unique and more concerned with either conveying a decision or using the email as a paper trail for future follow-up.I love this feature and it’s cut down my email communication time noticeably.
Agreed. I think features like this, although on the surface seem good for productivity, have the potential to increase what I call “the dumbing down of everything”. It’s happening. Less-and-less educated people (or people who are more-and-more uneducated – same difference), overly reliant on technology even for simply day-to-day tasks, and so on – a lot of which you said, better than me. And what when that so-called great AI (which is not really AI) fails, or has severe bugs? All down, nowhere to go.Mwuhahahahahaha …
I need to send you a picture of our head of operations license plate. That very word.I agree with you on the dumbing. My kids go to a school where they have to take Latin. Is that good or bad? I think good.
Please do send it. Should be interesting.http://vasudevram.github.io…Learning Latin is good. If nothing else, it will help with thinking systematically and learning logic somewhat (via rules of grammar and how they interact), “construing” – I think that is the term – for translating Latin into English, maybe – not sure – from memory of reading English school stories as a kid. Never learned Latin myself, but did have some years of Sanskrit in school, was good fun and did well on it. Sanskrit is to modern Indian languages as Latin is to modern European ones, sort of, I’ve heard. Second reason why learning Latin will be useful is because it is supposed to be the base for many European languages, at least the Romance ones (if that term means French, Spanish, Portuguese, Italian, etc. – need to check). You would know all this, of course. I think even Romanian has some Latin-based words, based on some cursory reading of text in that language while surfing.A couple of recent HN threads on languages might be interesting:https://news.ycombinator.co…https://news.ycombinator.co…
predictive typing CAN let you be you AND boost up productivity… we’ve been working on this for quite a while… you’re welcome to check it out (@Lightkey).