“A key unsolved problem for future distributed AI systems is whether knowledge gained by a spawned-off agent can be successfully reincorporated into the original agent without causing instability or corruption.”