Host competitions to make it easy to measure progress in such domains

Understanding progress in policy-relevant AI domains (e.g vulnerabilities in widely-deployed computer vision systems, or the advancing capabilities of smart industrial robots).