@EvilJS you mean right now? today?

@vzn since the start up to SSRI, then break, and then commercialize, all the time with short break?

@vzn ok, summing up: I like to have quite full tree of possible moves (no prunning, I will need those results), show possible moves to the human, read the reaction (emotions, amuzement), read which path was to be taken by human (to estimate strategy), adjust the move to the most not expected (not the one that was calculated as best), iterate next move.

Now if this is failing (amuzement was there but the outcome is not winning for computer) then increase level, if it broke the players path, adjust to decrease (to avoid making human playing at stressed condition), to employ strategy, not defensive playing.

So I am trying to give steering of ai to EEG read of player, making feedback. The goal is not to win (this would be also nice), but to let human feedback moves. This will allow to prepare player for unexpected moves, increasing the level.

To adjust I need clean reads, cannot bore or stress too much the player. So lower branching factor of the game seems good choice. My problem is in presenting outcomes (if there are too many of them) and the second - while adjustung ai level, I do not have good measure of "hardness".

@EvilJS hard to follow. are you saying you want to measure human reactions to game choices presented to them? and they make choices from the list? what do you mean "increase level"? "broke the players path"? "decrease" what?

"increase level" = better play?

@vzn yes, exactly, measure human reaction to choices presented - imagine that you have the 5 very good moves in chess, all from the top of evaluation tree. I show them as animation, that such move might occur, all five of them.

And the choice is made based on feedback, not evaluation score (but they were evaluated, those are top possible moves)

By increase level of AI I mean better playing computer player. By level up for human - increasing the amount of moves to think about it - also meaning better playing.

Decrease in AI - make the AI playing worse moves, decrease hardness, make the game easier for human.

@EvilJS improve computer algorithms based on human expert analysis of moves?

@vzn the main goal is operationalization (finding indicators, making it measureable) of deep vs shallow learning theory, which states the correlation of emotions with learning speed and type of learning.

The outcomes are: improved learning by checking human state, adjusted game for player (more personal, not purely random or the same every time), so reinforcement for human (the task does not matter, but psychology loves games, very good at measuring progress).

The best game for learning is poker, but this is hard to use.

Poker for real money, making real game is optimal in learning for human.

But any kind of token based poker is harmful for poker playing, and does not have this properties.

So making more human-like player, improving the AI or "trying to follow the human strategy" (I do not count on that, but we shall see) are things to cover in meanwhile.

Several years ago I made modification to learning curve, memory storage, short-term memory speed up for long-term memory. So instead of learning like 20h before the exam and than loosing everything one might spend 4 hours and remember it afterwards. This game X emotions is improving the learning pattern further.

Making it short: time pressured learning with improper repetition cycles (there are none) ends up in shallow learning, connected with danger, so avoidance mechanism kicks in - remember the situation to avoid danger in future, hence the memory storage is not decreased in ine month but several days - pass and forget. But even when people are systematic and make good, clean learning, there are obstacles in the process, I want to get rid of them.

Before eeg and emotions I was doing research in long-term memory, recall, memory recovery techniques and the quite recent results of "forgetting", which are proven (I also checked it) that we in fact not forget, just decrease in recall speed, which is degrading very fast, but memories are preserved.

@EvilJS all impressive. have you published some?

← worked 4yr in neurobio lab ages ago

@vzn some of this yes, collaborative effort, or to be tested more and then published, or part of the thesis.

@vzn didn't knew. Anyway I really hate results that show marginal correlation or are using not conservative tests to make it count

I prefer results that are repetable, without prepare they just work. And the tests I do right now are screening.

I am not against publishing, it just works badly. There are dozens of very poor things published - working but not functional, making very strong statements and hidding the only important parts.

If I want to publish something I have to reveal the metodology, wait several days and others will start using that. And I do not need any publications, do not want to bother explaining everything...

No offence to others in the field, but when they publish "awesome results, with correlation ~0.15 on huge sample - 20 people" I cannot stop laughing.

And this funny differentiation by gender, age, right-handedness... I respect working results without methodological failures.

So learning theory improvements are in fact freely available and not even attributed the most of the time. I do not care. But when I try to build something practical, I prefer not to publish it, also issued patent would fail by publication... But someday all of the work, commercial or purely scientific will be freely available. For now I would like to afford new hardware for the lab...

@vzn why are you so interested in publishing everything?

If someone works at university, pursue higher degree etc. is (at least here) obliged to publish. Which is nice - the more is available, and stupid - pushing people to do obligatory articles causes decrease in publication rate (why publish now, if in the next semester I am obliged to get points for article). Also pushing it causes decrease in level (not in all publications, but the time pressure causes harm).

Expeeiment on 25 people is minimum, 10 if this was physiological, more than 40 per group (of not rejected results) to make it unbiased with additional control group. 2 groups + 1 control = 120. But some are rejected, so... ~135 people to conduct experiment. It takes time, but when there is rush, why bother? And the results published reflect this...

Technicians to conduct experiments (not skewing results), making groups random, controlling homogenity of group (not because one have to but because it is easier), everything goes away mass-publishing...

And for the top rated places you have to pay for review, wait, then it is published, of course not freely, or you can write a book, but it is better to be highly published - it is cyclic commercial nonsense, where knowledge is not improving the science...

keeping it simple <- this is how really progress looks.

@EvilJS lol you just "published" ~50 lines :P

once met a phd grad student (neurobio lab) who had worked ~7 years without publishing anything himself... o_O

@vzn so maybe asking him to publish anything would not be bad idea, but pushing people to make several papers per half year is really bad.

@EvilJS this seems to be sensitive topic for you. think this is all valid insight. routlinely encourage many on here to publish stuff (most who seem to have done/ accomplished far less than you). and again, myself consider blogs a form of publishing, and even SE chat rooms have permanent transcripts aka "(e)publishing".

@vzn cannot deny, it makes me sick, how it really looks...

@EvilJS ?

I mean the publishing industry, knowledge sharing etc.