An oracle (Wikipedia, Lesswrongwiki) is a theoretical capability-constrained artificial intelligence (AI) limited to answering questions. An oracle would plausibly be easier to safely implement than a general AI, and it may help to solve the safety issues associated with general AI. Therefore, a reasonable case can be made for developing an oracle first, even if a safe general AI seems feasible. However, an oracle would have considerably less (though still large) upside than a general AI, so it may be less appealing to investors.
Will an oracle superintelligence be developed before a general superintelligence?
This question will immediately resolve positively if an AI limited to answering questions achieves reliably superhuman performance across virtually all questions of interest before an AI achieves generally superhuman performance across virtually all human activities of interest. If a generally superhuman AI is developed first, the question will immediately resolve negatively. By default, the question will resolve ambiguously on the resolve date, 2099-12-31.
Successful creation of either technology would presumably be extremely obvious and uncontroversial, with a great amount of media coverage and scientific attention. However, if there is significant disagreement over whether a given apparent achievement resolves the question, it will be determined by Metaculus moderators.