This is a decent article that summarizes Peter Norvig's 2007 Singularity Summit talk, which made it sound like Google is definitely not working on AGI. http://news.cnet.com/8301-10784_3-9774501-7.html
This interview implies to me that at some point in the future he will assign Google researchers to work on the broad problem of AGI, even though at present Google has no employees dedicated to AGI.
Pretty much, with the addition that they are working on the prerequisites for perhaps other reasons, but I suspect with the AGI potential in mind. Do not remember which one of the 3 CEOs said this, but they know that the ultimate search output is NOT pages of ranked links; it is the answer to your search query. This smacks of an AGI-hard problem.
(Sorry this is mostly a link instead of a post, but I think it will interesting to the FAI folks here)
I helped arrange this interview with Peter Norvig:
http://www.reddit.com/r/blog/comments/b8aln/peter_norvig_answers_your_questions_ask_me/
I think the answer to the AGI question 4 is telling, but judge for yourself. (BTW, the 'components' Peter referred to are probabilistic relational learning and hierarchical modeling. He singled these two in his singularity summit talk)