Can computers still improve at chess?

Computers have already proven to be unbeatable at chess. With classical time, no man can compete against them. So my question is, can computers still improve, or is this the best they can possibly get? Is it even possible to improve the greatest chess engines? If so, how long do you think it will take until there is a program that can beat the likes of Stockfish 12 and Alpha Zero?

there is always room for improvement, LoL

Of course they can improve... in at least three different ways: increasing search depth, improved positional evaluation and heuristics.

In Stockfish's case, anyone can suggest an improvement to the engine. The "improved Stockfish" is put to the test by playing hundreds of games against the unchanged Stockfish and statistical tests are applied to see if there is any noticeable difference. If the new version is better, that code will go through and thus Stockfish has improved.

Computers have learning algorithms. That's how they have become so advanced. It is beyond human design. the computers are improving on their own.

Alpha Zero was + 50 elo above Stockfish 8 ( Stockfish 9 strength). 

Current Stockfish 14 developemental version is estimated to be + 150 to + 250 elo stronger than Alpha Zero now. Even Leela is + 130 to + 230 elo stronger than A0 now. 

You can track and get free from development of top 2 engines here. 

https://github.com/glinscott/fishtest/wiki/Regression-Tests

https://abrok.eu/stockfish/

https://training.lczero.org/

Why is still progress? Because 3 million games of training for Leela was done yesterday (and nearly 1 million games per day approx in last year) doing everyday and hundreds of thousands of testing games for Stockfish is doing everyday on demand. 

Yes computers can still improve, and they do! You can track the development of Stockfish here.

https://github.com/glinscott/fishtest/wiki/Regression-Tests

Stockfish is open source with a massive community of people who constantly discuss new ideas and tweaks to make it stronger. If someone has an idea for improvement they can train a net and test it by having the engine play thousands of games against other versions of Stockfish.

Stockfish 13 was in development for 167 days and gained a total of 36 elo over Stockfish 12, that's +6.5 elo per month. 

yeah they could improve, maybe learn moves more human, alot of bots make very strange moves with no point to them

Computer is not over human, beating the computer is easy it has no instinct I value picking opponents apart for the thrill of the hunt challenge your minds, I play warrior mode like my ancestor warriors win or lose I love the game and no computer can match a live hunt

Just don't forget to put on your war paint before the game...

Either way viewed it is a battle of wits, challenging the others senses to be a correct move. First to make a mistake at times, people play to their comfort, have some dignity in defeat and relish in your win, enjoy people

They've been improving every year for the last 20 years... what are you even talking about?

And Alpha Zero was surpassed long ago... is this a troll topic or not?

I'm pretty sure that the chess computers are some kind of AI so it's improving. 

This is wrong in more than a few ways.

If you don't know that's fine. It doesn't make you dumb.

With regard to question, yes artifice intelligence is in its infancy stages. Bound to get better as we advance. But, it takes humans to run or program, anything man made fails.

Not true. If there is 1 million positions search computing power,

Older engines see 10^6 (6 depth ahead)

Stockfish see 2^20 ( 20 depth ahead). 

Yes, current Stockfish is extremely intelligent and search very smartly( smarter than majority of human)