When someone develops a new theory on computer science, specifically in artificial intelligence domain, which is barely on schetch (so there are no measurements, nor simulations, nor finished code or ready application) with just a mathematical and conceptual description, in which scientific journals people use to publish these theories/concepts?
I can think of two cases where this (more or less) happened.
Valiant published his first paper on PAC learning in Communications of the ACM 27 (1984). Charlie Bennett got Stephen Wiesner's paper on quantum cryptography published in SIGACT News 15 (1983). I believe that at the time of publication, the editors had more latitude in accepting articles without extensive peer review than in the standard model of refereed journals.
In both cases, the idea turned out to be enormously valuable, and I am sure the prestige of Valiant and of Bennett helped the work get published. In both cases, the research was quite a bit better worked out than "barely a sketch", but it was still very sketchy, and not completely convincing to contemporary researchers.
If you are an unknown researcher, in order to get anybody to pay attention, you probably need to work out your ideas at least as well as they were worked out in these two cases, and even then I am not sure whether you could get sketchy ideas published without having the backing of a respected researcher (as Wiesner did).v
Ralph Merkle did not succeed in getting his sketchy ideas on what was to become public-key cryptography published until after Diffie, Hellman, Rivest, Shamir and Adelman had worked out the sketchy ideas in enough detail to convince everybody they were worthwhile. Fortunately, somebody put him in touch with Diffie and Hellman, who were clever enough to see the value of his ideas, talented enough to work them out in some detail, and honest enough to give him his fair share of the credit. See this article.