Formerly, a 3-3 invasion inside a naked hoshi would have undoubtedly been ridiculed. The thickness White gets after the classic joseki sequence was, and still is, regarded as far superior to Black's corner.
3-3 was only an option when neither the kakari at A or B is good, e.g. when White has extensions on both sides.
(Apart from beginners of go,) AlphaGo was the first to invade a naked hoshi, playing as Master (P) on Tygem and Fox Go Server, notably in game 23 vs. Kim Jiseok, at move 19, and in game 36 vs. Gu Zihao, at move 29. You can replay the games here.
AlphaGo's innovation was to omit the last two exchanges of Black's hane-connection (moves 9 and 11) with White's block and tigermouth (10 and 12) that unnecessarily strengthen White. This has been a blind spot in the go theory we used to have. Everyone who learned the joseki would have played these exchanges automatically.
Ever since, 3-3 has established itself as another option to deal with a hoshi. Numerous pros have been experimenting with 3-3, with the general idea to give the opponent a wall while ending in sente to play an extension that restricts the wall. The position below is one such example.
This changed when Li Zhe 6p famously proclaimed that the 3-3 invasion's function was actually to rob the hoshi of its base, and that a simple kakari instead would be too easy on the hoshi.
Li Zhe demonstrated this impressively by invading at move 7 vs. Li Xuanhao 6p.
Li Zhe's subsequent split at 15 bears a hint of a pincer that is aiming at attacking White's „wall“.
White approached from the left and Black continued on the right. Up to 25, it is indeed difficult to tell who is on the attacking side. Nobody could confidently claim that White's wall is superior in strength to Black's splitting stones. And thus, the game is now favouring Black, as White's position on the lower hemisphere does not hold the needed advantage to offset Black's advantage on the upper half of the board.
After Li Zhe so impressively took apart his opponent's star point, world champion Ke Jie 9p was the fastest to adapt. He has used the 3-3 invasion in almost every official game since and built up a winning streak. (Coincidentally, his winning streak was stopped by Tan Xiao 8p, in Ke Jie's last match before facing AlphaGo next week. In that game, Ke Jie stopped invading 3-3 for unknown reasons.)
Let us have a look at the various countermeasures Ke Jie's opponents attempted.
Lee Donghoon 8p extended at 19 and followed up with a double hane in order to take sente for a move on the right. However, the kosumi at 29 was perhaps questionable as White got to split (or „pincer“) at 30 anyway.
Eventually, Black felt the need to come back at 45 to strengthen his „wall“. It was gote however and White could play another move somewhere else on the board. (Black later added one more move in upper right that was also ignored.)
Next, Ke Jie faced the very Li Zhe, proponent of the 3-3 invasion's effectiveness .
White again took sente, this time without a double-hane, to close the lower side at 52, lest Black play here and render White's wall useless. Ke Jie decided to settle the shape in the following manner:
In the end the corner proved to be too big. White's centre got destroyed later and Black won this game fairly easily.
In the next game, Ke Jie faced Dang Yifei 9p. Basically the same thing happened as in Li Zhe's first game.
White then added a move on the lower side and strengthened his wall, and Black got to invade at A, ultimately tilting the territorial balance in Black's favour.
Finally, Ke Jie broke Li Zhe's record for the earliest 3-3 invasion in official games when he invaded against Li Qincheng 9p at move 6.
In this game, Black decided to approach the splitting stone from the side of his wall at 15, perhaps acknowledging the weakness of that group. The result has a counterintuitive vibe of trying to surround territory with thickness.
Eventually it became Black's only option to continue surrounding territory and White won the game without any problems after reducing it at 36.
(We can note that both players missed the timing to play at 30. Black should have blocked there, exchanging it with White's hane-connect, since Black was going to play 29 anyway. And White should have crawled one move earlier as to not give Black this opportunity. A result of the players' inexperience in this shape.)
It is too early to discard the star point as a bad move. Some claim that the bad move was actually the usual kakari to a 4-4 point, and that the 3-3 invasion should be the first choice. Whether this is true or not, the 3-3 has at least become another option next to good old approach moves.
(One way to handle the invasion that is being experimented with is tenuki, not discussed in this article. As absurd as it sounds, at this point we can only say „Why not?“)
In conclusion, go theory is undergoing a potential paradigm shift. Joseki books will need to be rewritten once a conclusion is reached regarding this invasion, and we will have to update our approach to even the most basic concepts in go.
This makes us wonder which other blind spots in our go theory we have yet to discover. Hopefully the upcoming match with AlphaGo will give us even more fresh ideas.
In this instance, AlphaGo's idea was further developed by a human. Indeed, some pros believe that a combination of AI and humans would produce the best results, superior to AI-only or human-only. Even though humans are widely considered to be no match to AlphaGo, this makes sense in a way. We can find an analogy in computer-assisted translation that is both more accurate than computer-only translation and more efficient than human-only translation.
Regrettably, AlphaGo's only way so far to communicate with us is by playing actual moves. Surely go theory would take another leap if AlphaGo was given the ability to explain their ideas.
(Thanks to go4go for the sgf files.)