Are we getting better?

[TOTAL]

Hi All

My first post here.
I have been thinking recently about how effective the modern go learning tools and the bar set by AI such as alpha go are in increasing our skills. I can easily see how tools like KaTrain with their strong engine and the metrics help me progress very quickly, but I wonder if this value of AI support is reflected in the top limits of human capacities.

Have recent years seen any substantial leap forward of the top professionals? Whether yes or no, what measures do you find the most reliable evidence for that?

[pajaro]

I think the answer has to be yes.

Not considering AI, the global knowledge and experience is always increasing. People learn from their games, but also from others. The same person in different historical times would have different learning chances. Hopefully, the later you are born, the better. This is also true for any other field of knowledge.

With AI, the research for new moves, new analysis... is faster. Josekis take very short to be accepted, mistakes are found even in high level games, and there is more room to improve.

How would a top player today do against a top player of 20 years ago? The top player today would have the advantage. But of course, that doesn't mean that any player would win against an opponent from the past and lose against an opponent from the future. Individual talent matters.

[kvasir]

When you study game records from the past it is evident that even top players didn't always have the right understanding and when presented with new moves they often played moves that are not the best in hindsight. You can find many such examples in the opening and in joseki.

There are also examples of new (and better) strategies. In games before the 20th century it was usual to play every big point in the opening before challenging the opponent in the middle game. They followed a rough guideline of how big these moves are compered to each other, this is really useful to learn by heart, but modern players never play like this and tend to push much harder to challenge the opponent early in the game. It is largely the same knowledge that underpins their play but the strategy is different.

The general skill level and knowledge also seems to have increased steadily through the 20th century. Analysis has also become much more objective. I have observed that some "old" josekis appear to really just be what was played in some famous game, maybe there was a special reason why it is very good in that game, and sometimes they seem to be accepted for a long time without anyone questioning it. In later half of the 20th century, maybe because of more volume of published books and magazines, the analysis seems to become much more objective. Maybe a larger number of recognized "strong" players is the primary cause for better analyzes, it is easy to dismiss opinions in any human activity because the are not from the right person.

With the new AI programs it is much easier to analyze objectively. Many positions and moves that we hardly knew how to judge can now be analyzed by anyone. There are certainly many skills that we can learn using programs like KaTrain but were not accessible for us before because it was too difficult before to test our understanding and skills.

I think it must be the case that modern players are overall better than in the past but it is an unfair comparison. Still some players from the past were very good, even in hindsight and despite whatever the shortcomings of the past, would be among top tournament contenders if they arrived today in a time machine.

[jlt]

Take a player from the past and give him 3 days to study new josekis and review a few modern games. Would he play worse than modern players?

[Knotwilg]

I don't think the original question was if past players were intrinsically better or worse than modern ones. It's precisely the availability of knowledge that has increased the overall level. For me it's a yes. I'm as studious as ever before and I have trouble keeping my rank.

[TOTAL]

Naturally, there is progress in knowledge, strategies and methods of learning. Old strategies prove obsolete in the face of new ones - that is a given in all areas of human endeavor actually. Even if Einstein would shine among our contemporaries, most of them would outshine his contemporaries.

What I am after is 1) whether humans have got significantly more adept, say, from pre-AlphaGo to AlphaGo Zero. Thus, much shorter time span than a century, and human 'curve of development' in sync with that of AI.

2) Some statistical tools to make such progress measurable would be useful but would require an objective point/scale of reference, so happy to leave this part out for now.
Without it we are left to our collective subjective experience though so a more general question boils down to 3) is AI in the avant-garde yet (meaning are the best humans learning more from AI than from one another) and 4) does having such models make human players more ingenious or just more skillful.

[RobertJasiek]

kvasir: "Analysis has also become much more objective.": Yes, where application of mathematical theorems allow objectivity.

"With the new AI programs it is much easier to analyze objectively.": No, because AI can make mistakes. Analysis with AI need not be objective. The advantage is rather that using AI allows to find more blunders, which human beings might have overlooked.

Knowledge of go theory increases and more people have access to AI as another tool; this can make everybody stronger. At the same time, thinking times decrease resulting in more blunders (as I have also observed), less tactical reading and less endgame evaluation; this can make everybody weaker.

It boils down to opinion whether the overall level has increased in the last decades. It is easier to justify improvement over centuries because opening knowledge was significantly weaker in early centuries. Circa five centuries ago, the overall knowledge of go theory was weak compared to today.

Concerning my own experience as a 5 dan since 1998, my knowledge has increased dramatically since then and the 5 dan I was in 1998 would have no chance against the 5 dan I am today. So I think European dan play must have improved significantly since then. I just cannot say exactly how many ranks in terms of 1998 ranks. Might be 0.5 or 1.5 ranks - I do not know which. When comparing the skills of 3 dan opponents then and now, I think it is closer to 1.5 ranks overall improvement. When comparing the skills of 5 dan opponents, judgement is much harder because I might be prejudiced too much. When seeing the improvement of top Europeans, 1.5 ranks is realistic but does that boil down 1:1 to 6d and 5d players, too? I am unsure. At least 0.5 ranks for sure, but it could be more.

I would not suspect the same level of improvement since 1998 of top Asian professionals for the reasons further above though.

[jlt]

Assuming 5 dan in 1998 = 4 dan in 2021, this doesn't prove that 2021 players are stronger than in 1998. Maybe there has been rank deflation. You need to examine the rank distribution before drawing conclusions.

[RobertJasiek]

My judgement is independent of rating systems and only depends on my subjective assessment of quality of play and knowledge then versus now.

[John Fairbairn]

My judgement is independent of rating systems and only depends on my subjective assessment of quality of play and knowledge then versus now.

Knowing you have greater knowledge now may affective your subjective assessment of quality (the old "if only I knew then what I know now" syndrome).

But in general, it is likely that we all underestimate how the effects of age creep up on you - tiredness, less concentration, less caring about the result, more cares in the rest of daily life, more blunders, more decrepitude. And we also tend to underestimate how early all this begins. Look at Yi Ch'ang-ho. He was not even 40 when he went off the boil. Yet I bet he knows more about go now than when he was a world-beating teenager.

But knowledge acquisition itself is a two-edged sword, anyway. All the new knowledge you soak up can knock useful old knowledge to the bottom of the pile, can play havoc with your intuition.

Remember, too, the old chestnut: an expert is someone who knows more and more about less and less. My own view is that there is a good chance that there is no modern writer demonstrably better than Shakespeare or Murasaki Shikibu, say, or no modern painter better than da Vinci or Rembrandt, or better musician than Mozart or Beethoven - but there are many modern people in those fields who are more knowledgeable ("hey, I can even tell you about literary theory, how to use acrylic paints, how to make music with a midi synthesiser" etc). The modern need to become "experts" leads people down narrow culs-de-sac. They may be Top Cat there, but will never match Renaissance Cat.

Quality tends to outdo quantity when we evaluate things,

[RobertJasiek]

I am aware of the possible impact of age but do not think that it (already) affects me significantly.

tiredness, less concentration, less caring about the result, decrepitude - no

more cares in the rest of daily life - yes but hardly while playing

more blunders - no, instead less blunders

can knock useful old knowledge to the bottom of the pile - in my case, I replace less useful by more useful knowledge

intuition - in 1998, I still played some moves under ordinary thinking time solely due to guessing; since ca. 15 years, I play all moves under sufficient thinking time due to reasoning

"an expert is someone who knows more and more about less and less": Although there may be such experts, I am not such. Instead, I know more and more about a) more generally applicable knowledge and b) more details about related or other topics. I love generalisation but if necessary I also describe the details.

[Knotwilg]

(...)

Remember, too, the old chestnut: an expert is someone who knows more and more about less and less. My own view is that there is a good chance that there is no modern writer demonstrably better than Shakespeare or Murasaki Shikibu, say, or no modern painter better than da Vinci or Rembrandt, or better musician than Mozart or Beethoven - but there are many modern people in those fields who are more knowledgeable ("hey, I can even tell you about literary theory, how to use acrylic paints, how to make music with a midi synthesiser" etc). The modern need to become "experts" leads people down narrow culs-de-sac. They may be Top Cat there, but will never match Renaissance Cat.

Quality tends to outdo quantity when we evaluate things,

I often wonder where the 21st century Leonardo da Vinci is. Or, given today's accessibility of knowledge, tools and people, how on Earth did da Vinci do it back then? There are tons of uneducated guesses I can make about it. The "perversion of specialization" which you hint at (I think) is one explanation. With total access comes abundance, is one of mine. And another is that science and arts have been democratized to a degree that it is comparatively harder for the elite to cross-pollinate.

[jlt]

But in general, it is likely that we all underestimate how the effects of age creep up on you - tiredness, less concentration, less caring about the result, more cares in the rest of daily life, more blunders, more decrepitude. And we also tend to underestimate how early all this begins. Look at Yi Ch'ang-ho. He was not even 40 when he went off the boil. Yet I bet he knows more about go now than when he was a world-beating teenager.

On the other hand, Sakata Eio, Rin Kaiho, Cho Chikun managed to win big titles after the age of 50.

[gennan]

It boils down to opinion whether the overall level has increased in the last decades. It is easier to justify improvement over centuries because opening knowledge was significantly weaker in early centuries. Circa five centuries ago, the overall knowledge of go theory was weak compared to today.

If a Japanese Edo period 7p would time-travel to today, there would be a knowledge gap. But I think strong pros of today may not be able to give more than josen handicap, maybe less if the classical pro is allowed some time to catch up with their knowledge.

Concerning my own experience as a 5 dan since 1998, my knowledge has increased dramatically since then and the 5 dan I was in 1998 would have no chance against the 5 dan I am today. So I think European dan play must have improved significantly since then. I just cannot say exactly how many ranks in terms of 1998 ranks. Might be 0.5 or 1.5 ranks - I do not know which. When comparing the skills of 3 dan opponents then and now, I think it is closer to 1.5 ranks overall improvement. When comparing the skills of 5 dan opponents, judgement is much harder because I might be prejudiced too much. When seeing the improvement of top Europeans, 1.5 ranks is realistic but does that boil down 1:1 to 6d and 5d players, too? I am unsure. At least 0.5 ranks for sure, but it could be more.

I also believe that EGF ranks have slowly grown tougher over the decades (even after the recent EGF rating system udpate that softened EGF ranks a bit).

There is a Dutch player who was a strong 5d around 1980, while having a hard time keeping an EGF 4d rating today (even though he kept studying and playing competitively during the past 40 years). He estimates that he can give a 2 stone handicap to his old self from 40 years ago.

As for myself, it was not that exceptional for me to beat 5-6d players in tournaments between 10 and 20 years ago. But I don't have a chance against today's players rated 5-6d EGF. Even though I'm pretty convinced that since then, my reading has improved significantly and my game has become more balanced.

But it's possible that I'm deluding myself, not noticing the effects of ageing and other factors that slowly reduced my competitive abilities. And maybe the variation in strength of 5-6d players has shrunk over time, because today's players are more prone to demote if they feel that their results are below par for too long.

[gowan]

I think the original post referred to the population rather than any particular individual. Some of the great old players won major titles after the age of fifty but even
then it was rare enough to be remarkable. Consider how much attention used to be paid studying games of players from the 19th century whereas now there seems to be little study of the games of twentieth century players.

I venture to say that some standard patterns understood in the 1970's (e.g. Avalanche, Taisha) would perplex twenty first century young hotshots. Of course the top programs would not be fooled.