iPad go AI: which one to use after SmartGo Kifu?

[lebigot]

After nine great months spent fighting against the SmartGo AI on iPad (mini 2), I am outgrowing its AI. Now, I am considering moving to Little Go (Fuego), Champion Go HD (Crazy Stone) or Igowin HD (Many Faces of Go). Which one(s) do you recommend?

My needs are:

1) Get an offline AI opponent that can accompany my progress for as long as possible (without paying extra like offered in Champion Go HD, at least at first).

2) Being able to add comments to the moves while playing would be great. Little Go cannot do it. Igowin HD can, but can it do this mid-game? What about Champion Go HD?

I did not succeed in finding enough information on these three programs on the web or here, so I ended up buying both Igowin HD and Champion Go HD, and checking out Little Go (after also considering GinseiIgo, which is apparently maybe the strongest of all). The answer to my question is clear: Igowin HD allows players to add comments during a game, anywhere, and goes up to 3 kyu (calibrated against real players), which is higher than SmartGo's level (all four programs are currently stronger than SmartGo).

Igowin HD is thus more appropriate for studying, I find: I can reflect on games while I play them, and jot down some notes that help during reviews. It even allows game branches to be explored during the game, which nicely complements the note taking feature (e.g., for writing down a sequence that one has in mind). The auto-level feature and real-rank features are also convenient. I don't like the fact that it is limited to portrait mode, because the iPad can be harder to hold this way and because the keyboard is small (for writing down notes during a game), but it's a small price to pay.

Champion Go HD might be a good program to use after reaching the limits of Igowin HD: not only is it apparently stronger (and clearly stronger with the in-app purchases of games against 5–6 dan bots), but it also offers a limited-time mode (10, 20, 30 seconds), which is more geared towards human-vs-human games (servers, competitions)—I still want to spend as much time as I want on moves: qualitative progress comes before quantitative progress, I believe.

I was surprised to see how little knowledge about these programs can be found on the web. There is not enough clear information about their features and limitations to allow users to make a clear choice based on their needs. Hopefully the above details will help some.

[Boidhre]

Don't take the kyu ratings on IgoWin HD very seriously. They were way off KGS kyu levels (which are on the "weaker" side of things) when I played it last. Remember your goals here, if you just want to play bots those two are good choices and maybe look into Crazy Stone for PC, it's much stronger than the basic app on iPad. If you want to play humans some day you're not going to be very good at it if you just play bots because bots and humans play rather differently and a lot of what works against specific bots on the iPad won't work against humans.

[i3ullseye]

Not sure how helpful this comment might be to you, but I feel your pain. I get a very real sense that a lot of player discount AI completely as a solid learning method. Some seem to think other than having a teacher play a game with you, and comment as you go, everything else is inferior. Books, software, problems, etc.... only live play, and especially with a teacher, matters.

I wonder if that mind set hasn't undermined much of the great available resources by not getting effective use and reviews out there for those of us who do well with books and software and such, or may be limited to only that. I get that playing other players is awesome.... it is a 2 player game after all, and playing people is what it has been about for thousands of years. But, when you are a weaker player (I hate the term beginner... you can play for years and still suck, and I am not ashamed to say I suck) sometimes people online can see you as ignorant or rude. You find great people online, but you also run into a lot of people who aren't so nice. Against an AI, I can play out every last move in 'known' dead territory to see what happens since I didn't realize it was dead. In live games, this tends to make people often not want to play you again.

[lebigot]

If you want to play humans some day you're not going to be very good at it if you just play bots because bots and humans play rather differently and a lot of what works against specific bots on the iPad won't work against humans.

If I understand correctly, the problem is that every AI/player is more or less strong in each facet of the game: some programs or humans might simply never punish some mistakes, or might never attack in some way, etc.—this goes beyond consciously exploiting the weaknesses of a specific AI or player, it's just that some things never happen in their games. So, if I understand correctly, the problem is not so much an AI/human dichotomy than the necessity of playing against many different players so that fewer and fewer facets of the game are left uncovered. Is this correct?

[Boidhre]

If you want to play humans some day you're not going to be very good at it if you just play bots because bots and humans play rather differently and a lot of what works against specific bots on the iPad won't work against humans.

If I understand correctly, the problem is that every AI/player is more or less strong in each facet of the game: some programs or humans might simply never punish some mistakes, or might never attack in some way, etc.—this goes beyond consciously exploiting the weaknesses of a specific AI or player, it's just that some things never happen in their games. So, if I understand correctly, the problem is not so much an AI/human dichotomy than the necessity of playing against many different players so that fewer and fewer facets of the game are left uncovered. Is this correct?

That's one important aspect of it, though the weaker the hardware the bigger the problem is with playing just AIs. It's not the only issue but essentially yeah, there isn't a problem playing AIs if you're exposed to a lot of human opponents regularly as well.

[Mike Novack]

That's one important aspect of it, though the weaker the hardware the bigger the problem is with playing just AIs. It's not the only issue but essentially yeah, there isn't a problem playing AIs if you're exposed to a lot of human opponents regularly as well.

The strength of the hardware matters only when the programs are making heavy demands on "crunch power". When these programs are using just their "go knowledge based AI" and not a "MCTS evaluator" a reasonably powerful ipad probably plenty of hardware power.

[Boidhre]

That's one important aspect of it, though the weaker the hardware the bigger the problem is with playing just AIs. It's not the only issue but essentially yeah, there isn't a problem playing AIs if you're exposed to a lot of human opponents regularly as well.

The strength of the hardware matters only when the programs are making heavy demands on "crunch power". When these programs are using just their "go knowledge based AI" and not a "MCTS evaluator" a reasonably powerful ipad probably plenty of hardware power.

`
I found the fuseki play of Champion Go to be much stranger on phones than tablets than on PC.

Edit: Actually, I'll take that back. Rechecked both apps. Both seemed to play similarly when I tested there.

[Mike Novack]

As long as these programs are still "playing by the book" there isn't even a difference between the "go knowledge based" AI's and the MCTS programs. All that matters is the size of the "book" and randomization strategies for selecting which of the book options chosen (I'd guess something like the same percentage of time chosen by the book games).

In both cases, divergence from the book will take place when a non-book move is made. But there may be some tricks involved (that allow for transpositions to be recognized).