You mean besides the fact there is apparently no AI implemented?..
Yes, it would work, but as I said above, IMHO it would definitely be overkill for something as basic as an environment-to-screen brightness equivalence. Itās just a simple ratio, corrected for individual preferences (thatās is the part lacking here).
My point is that unless somebody has some strange eye condition, what is too bright/dark one day will be the just as dark/bright the next one, so there is no need for āintelligenceā, itās a fixed, set once for all times ratio. A simple slider would do.
(And, we have more chances to get a simple slider, than an all-singing, all-dancing AI⦠)
It will be the same outcome, only is the AI method more user friendly and precise for different lighting conditions. Using AI for this is not overkill. But if FP decides to go for a manual setting, then thatās fine by me as well. In the end we need something more personalized and not a one size fits all solution regarding the auto brightness.
Sorry, but which part of āSensor value X ā Screen Setting Yā would need intelligence? Itās a simple table of correspondence!
Itās not like brightness comes in different flavors and contexts, itās a simple linear scale from 0 to 100k Lux, to which should correspond a given screen brightness setting (from 0 to 7 I guess).
āIntelligenceā would only serve to raise the price tag, clog the CPU, drain the battery, and introduce an inexhaustible source of errors/unwarranted stupid AI initiatives (āOh, subdued lighting! Iāll play some soft, romantic music and order champagne on Amazonā⦠).
Oh well, I guess we must agree to differ on that one.
One is an extra settings section where the user has to define brightness levels for certain situations. I think thatās cumbersome and user unfriendly.
The other one is that these settings are transparent for the user. The user notices itās too bright or too dim, then adjusts it. Thatās then registered. When you do it again, itās registered and fine tuned. With a few manual adjustments the auto brightness is personalized and setting that has been transparent for the user. Thatās very user friendly. And no, this simple AI is not resource intensive at all.
We might be talking about totally different things. There are no ācertain situationsā in my proposal, there is only a lower and an upper screen brightness limit to set, period. And if you find it is overall too bright, use the new āextra dimā setting.
Indeed, thatās your personal preference. You have to take into account other peopleās preferences. For me the current settings are pretty much how I would personalize it. So from that point of view I could argue no change is needed.
For you itās the lows you want to increase.
There are also people who would prefer lows overall unless there is direct sunlight.
The AI approach is the most user friendly and accurate personalization way of doing it. I hope FP will do that. But yeah, I think thatās enough about this subject
Back to you⦠The fact the status quo suits you, is not a reason to not introduce means for other people to be able to change it on their phonesā¦
Sure, but setting lower/upper limits wouldnāt prevent that, would it?
Of course Fairphone could also implement a delta slider, one which introduces a correction factor (ideally positive and negative), which would allow to add a +something or -something correction to all the settings. Here again, very simple system, cheap as dirt, any intern could implement it.
Designer 101.
You are still thinking in upper and lower preferences Because you want to change that.
What I stated was an example to let you see your own point of view from a different angle. I agree personalization is needed. And creating an AI machanism isnāt that difficult as you think. The training model is what you describe as manual settings. Which is more cumbersome, less user friendly and less precise.
If someone prefers low brightness indoors, but not outdoors, then your suggestion to only change the lows and highs wonāt work in those different environments. Itās very limited then.
I think you have this point of view because you never had a phone with adaptive brightness like this. It really is making life easier for everyone.
Sure, but keep in mind that ābetter is the enemy of goodā.
I donāt think in absolute terms, but inside the limits of what I think Fairphone could achieve. We know they, err, āarenāt too good with softwareā, they donāt have enough human resources to throw at it, and they are already swamped by numerous, important issues to fix, some as serious as not being able to make calls ā on a phone!..
While I agree that having shiny sophisticated new things would be swell, I also think in the current situation itās totally unrealistic: Fairphone clearly can barely fix the known serious problems, so asking for additional bells & whistles is pointless.
On the other hand adding the simple slider solution is so simple and cheap they might be able to pull it off, eventually.
Thatās why I advocate it. āHalf a loaf is better than no breadā and all that.
No serious commercial company will ever accept such a thing, because they are legally responsible for the product they sell to their clients.
So, when (not if!) eventually Fairphone needs to fix that black-box āAIā software blob, that intern will be gone, and nobody will know why it does not work as intended, or what to do about it.
They might not have paid it very much, but the time wasted because of it will be expensive, not to mention they really donāt need yet another support case.
Machine learning is already fiendishly hard to get right, control and debug (even companies the size of Microsoft struggle getting it right-ish), so if you donāt have the know-how at home it is unmanageable.
This is really a dead simple form of AI, really. And from experience I can tell you that interns get to develop serious improvements in an infrastructure, even for banks. It doesnāt get more commercial than that
Please donāt call this āAIā. This has nothing to do at all with any kind of āAIā. Itās just adjusting the control curve in a more comfortable way. Instead of manually changing the mapping from 5-100% for 0-100lx you tell the device to use for example 20% at 2lx and maybe 80% at 70lx and it will remember this.
I had once a color graphics display from Eizo which work exactly that way. There was a light sensor built in and when ever you changed the brightness manually it stored the combination of environment brightness and the selected brightness setting and over the time the automatic brightness control worked exactly how I preferred it. However there was also an option to reset this.
This is AI, at least it should be. I donāt understand why FP doesnāt have it configured like that. Weāll see how they will approach it in the next few updates.
AI is composed out of several fields, and in this case itās machine learning, which is also divided into several fields. Probably supervised learning is used here since the user provides input for the desired brightness levels, which is not a linear preference, the user may prefer different levels of brightness in different lighting conditions.
Now, this discussion will not be used by FP to base their decisions on, and I think enough has been said about it. Iāve asked a friend of mine which is the head of AI for Dutch education and research development. If he says Iām wrong, then Iāll come back and correct myself. Otherwise, I think enough time was spend on this subject
His response:
yeah I mean by now, there are several generations of AI in production and since this is such an easy task I imagine they would have done this in some of the earlier stages, so most likely supervised (actually active learning)
Maybe the situation is also caused by the fact that FP does not develop the stuff on their own but mostly rely on the partner who created the phone according to their specifications.
And also if it is called āmachine learningā it is just a simple algorithm to adjust the automatic brightness adjustment. There are only two parameters: current environment brightness and display brightness. When you change the display brightness you tell the system how bright it should be for the current environment brightness.
There is not much to ālearnā - its just a number which gets adjusted and the sytem will remember that number. So when you tell it to go to 20% at 3lx it should always be that way as long as you donāt change it to another value. Thatās the idea of it: remember how bright a user wants the display at a specific environment brightness.
It is called machine learning, and thatās a field in AI. Letās not diverge the discussion towards semantics like that. Those are clearly defined and if Android Police is correct it should behave in how Iāve been trying to explain it in this thread several times.
Guys please⦠itās obvious that with the update a feature was added that wasnāt there before.
Thatās a fact.
Nobody minds the new feature coming in, the problem is here and all alone. That many just find crap and the others just great. Those who think it sucks were neither asked (how is that supposed to work) nor was a good controllable function built in, by switch or something else. Discussing that it is a matter of personal taste is totally unnecessary. Itās just shit and annoying, because there was simply no consideration for the other group. The learning function does not work reliably! Thatās a fact! Iāve readjusted the controller myself at least 100 times. Maybe then the ambient light will be measured differently again in certain values, I donāt know. Itās definitely annoying and I donāt feel like having to constantly regulate it manually because of people who have to wear sunglasses when itās raining.
Just put in a switch for extended darkness or whatever. I have to admit, since the colors of the Quickistep Launcher are not even being adjusted in a meaningful way, or the Bluetooth AdaptX HD Audio Bug is being fixed, I think that something like this will never happen⦠What Fairphone delivers here in this price range is just sad.