MIT Technology Review reports on new research finding that AI agents currently cannot conduct open-ended AI research — free-form investigation without predetermined answers that requires judgment and creativity. The piece explains that this limitation suggests recursive self-improvement, in which AI systems upgrade themselves with minimal human oversight, will likely take much longer than the AI industry’s most optimistic timelines suggest. It notes the open question going forward is whether open-ended research is truly necessary for recursive self-improvement, or whether narrower incremental gains could achieve similar ends.
