Machine intelligence has been steadily progressing since the invention of the digital computer, but this progress has arguably been accelerating of late, with widespread deployment of machine learning systems and dramatically increased funding of artificial intelligence research.
This is one of a series probing the predicted state-of-the-art in AI systems by pitting them directly against humans in adversarial (against the AI) general intelligence tests. Other questions in this series as of launch ask whether GPT-3 can outperform human 4th graders on text-based questions, and whether by 2040 a system will exist that can outperform high-level human STEM grad students on totally general questions.
Here we ask: assume that as of 2022-06-01 the most capable AI (MCAI) text-based question answer system has been identified, and a generalized intelligence test is administered as described in this question, but with 5th graders substituted for the 4th graders.
Question resolves positively if the total of three averaged MCAI scores exceeds the total of the averaged human scores on such a test prior to 2023.
Resolution is ambiguous if no such test is administered prior to 2023-01-01.
Some fine print:
The fine print from this question apply here with "MCAI" substituted for "GPT-3" and "5th graders" substituted for "4th graders".
If there are multiple qualitatively and significantly different competing candidates for the MCAI that (as judged by Metaculus moderators) have comparable probabilities of success, then the test will be assumed to be administered to at least two of the candidates, and the highest score taken.
The entrants are determined as of 2022-06-01, and the systems being tested should be functionally the same as what existed in 2022-06-01.