Post
5 Unsolved Mysteries in Synthetic Intelligence Research
Five research questions that remain open: evaluation, generalization, data rights, oversight of agents, and what the word superintelligence should mean.
Disclosure. Further reading links to a sister publication under the same ownership. Those pages are commentary, not evidence for a score or a ranking on this site.
The short answer
Five questions are still open: how to measure general capability, when a score is contaminated, who owns training data, how to oversee tool-using agents, and what superintelligence should mean in evidence. They are research problems, not product mysteries we solved.
How to read this list
This is an editorial map, not a tested ranking and not a score. BestSuperintelligent has not independently tested the products, discoveries, or headlines named here. Nothing below is a winner. Use it to decide what to verify, then check a primary source.
Research marketing likes a mystery that the product then solves. These five are not solved by a launch post.
Which questions are still open?
- What task set would show a general capability rather than a demo.
- How to detect benchmark contamination.
- Which training uses are lawful or licensed in each country.
- How a person supervises an agentic workflow that can act.
- Which definition of superintelligence can be falsified. Ours is in superintelligence and today’s tools.
Evidence in this article Editorial interpretation Our reading of what the evidence means. Judgment, clearly separated from the evidence itself.
Frequently asked questions
Has anyone solved superintelligence?
No result on this site says so. The term is not a test score.
Why call them mysteries?
Because serious papers disagree, and we will not pretend a blog list closed them.
Where is your definition of superintelligence?
In the research article on superintelligence and today’s tools, and in the glossary. A brand name is not that definition.
Sources
- How to read an evaluation BestSuperintelligent. Our own standard for what a result must show.
- Methodology BestSuperintelligent. Proposed weights and the rule that missing evidence is not a score.
Further reading
From a related publication under the same ownership. It is commentary, not evidence for anything on this site.
- What is superintelligence?, TheBestSI (sister site). Sister-site definition. Our glossary remains the definition we use.
- The uncertainties that matter most, The Best AI (sister site). Sister encyclopedia chapter. Open questions there are not closed here.