Half a croissant, on a plate, with a sign in front of it saying '50c'
h a l f b a k e r y
Oh yeah? Well, eureka too.

idea: add, search, annotate, link, view, overview, recent, by name, random

meta: news, help, about, links, report a problem

account: browse anonymously, or get an account and write.



Genetic algorithm phone keyboard

At the crowsourced version, people get a phone (software) keyboard, the thing mutates and sexes until measures of error (go back a few words single letter corrections) go down a lot; being crowdsourced the data pool is big and fast. At the individual version, new whack a mole game calibrates, and some GA and NN activity occurs in response to one person's use; new mid-word spell checker is better at guessing things.
  [vote for,

So there's this glaring problem that that is using up .1% of earth's congitive bandwith and time. Phone keyboards, like the one at Android, are terrible.

No, like really terrible, at the default keyboard on my $29 walmart android phone there is a little 1234567890 press-long to shift as alternate meaning of the qwertyuiop. Press the phone lingeringly and essentials like E and I and R turn into numbers and then you have to go back to edit the typos.

It's actually noticeably annoying, and if it is true there are 6 billion phones on earth, all of which will likely eventually be touchscreen phones, then that's 6 billion people wincing at this. The solution was simple and took only 1-2 hours. I tried different keyboards from the play store, found one without 1234567890 as a shift of qwertyuiop and things are better. So, .5B is for inventions rather than suggestions, but Google, and phonemakers, why not just have an alternate keyboard people can switch to as part of Android?

Six billion people spending 1 hour a month retyping typos is 72 billion hours of "Snerk!" globally a year, or 1/730th of life spent on typos; Numerous millions of person years annually. Time for a better standard keyboard.

So if you really want to improve phone keypads:

Make some keyboards, including a few with obvious human design improvements like an absence of long-finger-dwell alternate letters, Try different keypad sizes. move the "microphone" keycharacter away from the spacebar (They did that), I'm not dyslexic, but there is a "dyslexia reduction font" they could try "normal looking" variants on.

Then improve these human designed keyboards with Genetic algorithms and neural net study of the keyboard. At a crowsourced version, a few hundred, thousand, or hundred thousand people could try out the new keyboards and the genetic algorithm would have a great time optimizing against "fewest goings back to previously written text to change 1 letter". Software would observe and fine-tune timing of any long-finger press action people, or at least some people, liked. and of course, grow and shrink the size (and font) of the letter labels on the keys. I think people hunt and peck, and study could find out if Arial hunting and pecking or Serif-hunting and pecking was .5-3% faster, with .1-1.0% less errors. They could try making the keybuttons slightly different sized, might as well test it.

A genetic algorithm might even build a new font completely around what caused the least [single letter change go backs] from typing. The crowsourced Play store version is fun, because the keyboard talks to the cloud, and passes on anonymized keystroke data to make the biggest improvements it can make rather fast, and then it updates people who have toggled they keyboard to "updateable".

At the noncrowdsourced version the improved human design is accompanied by moderate, but real genetic algorithm and neural network learning. 10,000 letters is just 2000 typed words, and perhaps enough for finger dwell time optimizing and various programmatic (rather than GA) swap outs of fonts for optimizing hunt and peck at that particular person. Size of keys, of course can change too.

New to me: the mid-word but after letter spellchecker. Typing Hrllo? perhaps you meant Hello?

Rather than waiting for a word to finish before spell checking do the spelling predictions as the word is typed to reverse near field keyboard grid anomalies. Things like "Hr are much rarer in English than He, and things like "rl" are also kind of rare; and Hrl (the word so far) are actually almost unheard of, so the near field grid autocorrector decides that amongst [rfdsw234e] It's going to put in an e, unless you keypress long, because of course you could be typing something weird like a productcode

Most people wouldn't calibrate, but some might: The near field-grid, midword spell checker could also learn what kinds of mistakes your hands most often make; undershoot or overshoot the little letter pad you press?

Have a person optionally calibrate their phone by playing a little 9 second "whack a mole game" on a grid, that changes size, until they always win, that is the keys are optimally sized, and importantly, the computer/phone now knows they almost alwys go for the upper or lower right corner of anything, of any size that looks like a key when typing, and then that informs the midword spellchecker.

beanangel, Dec 20 2020


       Well, about the Genetic Algorithms (GA) and Neural Networks: I actually do think they would improve things. The data trend from looking at publications is applying GA causes about 30% better function at things across widely different fields, although with some fields (scheduling especially) 300-400% improvements have been developed. I know people don't publish negative findings much (I have seen a 1% GA improvement published) but in some ways, to give them a little slack, genetic algorithms and neural networks are kind of basic forms. Just as you might not object to the word "into" if I said "put it into the container", saying "Use a genetic algorithm" to develop a typo-minimized font/lettersize seems basic, and inevitable. But it really isn't inevitable, so it's ok to mention it.   

       Thanks for WKTE on about mid-word predictive text. I did not know it already existed. The main remaining novelty is the little # grid map of side keys the algorithm uses to guess what letter to change, and that may already exist, it is just I've never seen/heard/read about sidekey region processing applied to a touchscreen keyboard. That does bring up a slightly new possibility, POS (Point of Sale) touchscreens for clerks and auto-checkout folks could have # grid based "best guess" action, 1-2% fewer errors at checkout would be welcome, and at busy places, could make a difference.   

       Scrounging for excuses: I may be looking at kind of old technology. If they do not already, at writing touchscreen keyboards/POS they could use an invisible predictive grid instead of a button(!), that surrounds the button. If you have used an ATM where you had to hit the button just right and it often took more than one try (my credit union's ATM) the programmers could have actually placed a # grid of invisible buttons around the main software defined object button and done interpolation/interpretation. Perhaps there are simply better development environments (tools) and practices now where instead of a small area "sensing button", the option at the tool is easily there to define a much bigger, transparent "sensing field" overlay on the screen. At the ATM this would make touching anywhere on the top of the screen be "yes", and the bottom, "no". It might be sort of "defensive UI development" if you think your outdooor weather ATM is going to degrade in performance, even if they say it'll be fine. New ATMs and POS machines are likely much better than the ones I see around the small (30K) town/city where I live.   

       Improving Walmarts self-checkout scanner POS, Add an LED line of light where the card slides in. Right now its Who can tell? Slot, ledge, or ziggurat styling line? In real life of course it just works, but there is, almost everytime for me a moment of wondering which /ledge/aperture/thing is actually the slot. Maybe like hundreds of millions of other people have this experience that could be solved with one LED. Also, change all the red (non green or blue) LEDs to be IR/invisible. It looks like it's telling you its busy (red), from a partial view at a distance. If it's only emitting green its more obvious its available.   

       The Walmart touchscreen seems to work much better than my phone. I do not recall ever having made an error with a walmart touchscreen. Ok, once. But it was a human error. I did not want to wait for a swipecard bearing authorization person to appear, so I switched machines early on. My switch worked fine for me. It's likely I pressed a "cancel button" but I might have just walked away.   

       So what do you think of the 6 billion phone users, 1 hour a month fixing typos (an underestimate) occupying 1/730th of human existence? Globally that is. Perhaps someone could think of a mental way to make fixing typos pleasant.
beanangel, Dec 21 2020

       Like sitting in the passenger seat while the car is driving itself, computational machines that finish/repair your sentences, are going to be spine chilling. The progression looks to be there but still a bit off.   

       Though, if a critical mass of people sign up to teach the circuitry, this shock will come a lot faster.
wjt, Dec 21 2020

       The Words:Idea ratio here seems to be very high. As [kdf] says, predictive text systems that do mid-word correction and which learn from your typing (like SwiftKey) are WKTE. I can’t see what’s new in this idea.
hippo, Dec 21 2020

       Would someone kindly write a three sentence summary?
Voice, Dec 21 2020

       "Genetic algorithm improves"
pocmloc, Dec 21 2020

       On rereading it, (oh my), I'm just exhorting change, and scrounging up some ways to get it done.   

       A deeper thinker might have looked to "How is that even though everybody knows better there can be stuff that doesn't work"?   

       I have thought a little about that, and pondered why there is negative heterogenity at the physical culture of people. I think people that know more math than me might actually be able to do something about negative heteogenity, which is a great thing to believe.   

       [Voice] wants a three sentence summary...   

       You are typing "Hrllo". You mean "Hello". Mid word it guesses that the second letter is an e, not from sentence continuity or context, but from your own personal most frequent finger-miss map (#).   

       Phone software makers, crowdsourced or individually make finger miss maps so that when there's a typo the phone is better at spotting and fixing it. I was unaware of Swiftkey [link] when I wrote the idea. So I thought 2d keygrid correction (#) was new. I just read about Swiftkey, apparently keygrid correction is actually new here.   

       Some of the the word:idea ratio came from using the .5b window as the composition environment. I feel hints of contrition, but there's an an obvious way to improve. Just compose in a different window.   

       Also, [hippo] thank you for the Swiftkey mention, it looks like something I should get and swiftkey went through most of the things I thought were fresh.
beanangel, Dec 21 2020


back: main index

business  computer  culture  fashion  food  halfbakery  home  other  product  public  science  sport  vehicle