While I do not have the numbers, there are generally three types of submissions to CS journals.
- Invited papers: these are conference papers that were fast-tracked to appear in the journal, usually via some deal between the conference organizers and the journal. The idea is to give the authors of good conference papers an incentive to publish a complete set of results to the journal, and to improve the community. These papers almost always get accepted, unless something dramatic happens (e.g. an unfixable bug in a proof).
- Full versions of accepted conference papers: these also usually get accepted, but at a lower rate. If the original paper was sufficiently extended upon, and the journal the authors aim at is not too far away from the original conference (e.g. IJCAI/AAAI submissions -> JAIR/AIJ journals is a common route) then one could reasonably expect the paper to accepted after minor/major revisions.
- Direct journal submissions: these are less common, and face a much higher rejection rate. Some editors outright discourage such submissions (this is personal knowledge, not official policy). The reason is that it is often the case that these are papers that were either rejected repeatedly from conferences, or that they are submissions from someone outside the community.
Thus, it is generally not easier to get your paper accepted to a journal before a conference. Also remember that journals' turnover rate tends to be very slow, as compared to conferences.
EDIT: I was curious and asked around. Turns out that overall acceptance rates are about 30% in AI journals. This is over a very bimodal distribution though: papers of type (1) are rare, but type (2) have a high acceptance rate. Type (3) contain a lot of random spam submissions hence the low acceptance rate. Bottom line: if you have a paper you’re proud of, get it accepted to a conference. If you have some interesting extensions or not all got into the conference version, submit to a journal.