[Computer-go] Semeai and ko

Robert Jasiek jasiek at snafu.de
Tue Apr 26 08:37:27 PDT 2011

On 26.04.2011 16:52, Brian Sheppard wrote:
> I have always found kos in semeais to be bewildering.
 > I haven't come across any rules for how to count liberties when a ko is
> involved.

You have identified the complexity correctly. I started to study semeais 
with kos but after one month I postponed further study because I 
realized that there are probably already hundreds of basic(!) cases. My 
estimated research time for 2 group semeai without aji defects and 
without non-trivial scoring prediction (but arbitrarily many dame, basic 
nakade eyes, basic kos, approach moves) is between 3 and 24 months full 
time work. It might be faster to write a program that does all the maths 
proofs by itself... but writing what it shall find is already non-trivial.

Some of the major distinctions I have noticed so far are:
- topology of the semeai
- approach defects
- aji defects
- kos external for a particular group versus internal
- kos initially open or closed
- ordinary kos or flower-viewing kos, either with up to n approach moves

In other words, currently you do not have a realistic chance for a 
reasonably complete expert system. Try to use other approaches to the 
semeai with kos problem. E.g., identify equivalent intersections and 
replace them by one virtual, multiple intersection. Then apply MC or 

If your interest is all ko shapes rather than a strong program, read my 


And please do not forget to share your findings! :)

robert jasiek

More information about the Computer-go mailing list