The AI Economist: Taxation policy design via two-level deep multiagent reinforcement learning.

Artificial intelligence (AI) and reinforcement learning (RL) have improved many areas but are not yet widely adopted in economic policy design, mechanism design, or economics at large. The AI Economist is a two-level, deep RL framework for policy design in which agents and a social planner coadapt. In particular, the AI Economist uses structured curriculum learning to stabilize the challenging two-level, coadaptive learning problem. We validate this framework in the domain of taxation. In one-step economies, the AI Economist recovers the optimal tax policy of economic theory. In spatiotemporal economies, the AI Economist substantially improves both utilitarian social welfare and the trade-off between equality and productivity over baselines. It does so despite emergent tax-gaming strategies while accounting for emergent labor specialization, agent interactions, and behavioral change. These results demonstrate that two-level, deep RL complements economic theory and unlocks an AI-based approach to designing and understanding economic policy.

Effect of an emergency department HIV testing program on the proportion of emergency department patients who have been tested.

OBJECTIVE: The lack of well-described population-level outcome measures for emergency department (ED) HIV testing is one barrier to translation of screening into practice. We demonstrate the impact of an ED diagnostic testing and targeted screening program on the proportion of ED patients ever tested for HIV and explore cumulative effects on testing rates over time. METHODS: Data were extracted from electronic HIV testing program records and administrative hospital databases for January 2003 to December 2008 to obtain the monthly number of ED visits and HIV tests. We calculated the proportions of (1) patients tested in the program who reported a previous HIV test or had been previously tested in the program, and (2) the cumulative number of unique ED patients who were tested in our program. RESULTS: During the study period, 165,665 unique patients made 491,552 ED visits and the program provided 13,509 tests to 11,503 unique patients. From 2003 to 2008, tested patients who reported a history of an HIV test increased by 0.085% per month (95% confidence interval [CI] 0.037% to 0.133%), from 67.7% to 74.4%; the percentage of tested patients who had previous testing in the program increased by 0.277% per month (95% CI 0.245% to 0.308%), from 3.2% to 21.2%; and the percentage of unique ED patients previously tested in the program increased by 0.100% per month (95% CI 0.096% to 0.105%), reaching a cumulative proportion of 6.9%. CONCLUSION: Our HIV testing program increased the proportion of ED patients who have been tested for HIV at least once and repeatedly tested a subset of individuals. HIV screening, even during a minority of ED visits, can have important cumulative effects over time.

Relationship of self-reported prior testing history to undiagnosed HIV positivity and HIV risk.

Screening everyone for HIV at least once is estimated to be cost-effective. Screening in health care settings is recommended to help achieve that goal. Health care settings often encounter the same patient repeatedly, and it is unknown if limited resources are better allocated to conduct repeat screening, or to screen patients not yet tested. We reviewed data for a targeted ED based HIV screening program for 2003-2007. The role of prior testing history as a predictor of undiagnosed HIV positivity was assessed using a negative binomial model adjusted for demographics and risk behaviors. HIV testing was provided to 8,450 unique patients. There were 5,781 (70%) self-reporting a prior HIV test. Compared with patients reporting no prior test, the relative risk of HIV positivity for those reporting a test within the prior year was 0.90 (95%CI 0.48-1.66), and for those reporting a prior test more than a year previously the relative risk was 0.91 (95%CI 0.48-1.73). Among patients testing positive, those who did not report a prior test had a median CD4 count that was 228 cells/mm(3) lower than those with a prior test (CI(95) of the difference in medians 20-436 cells/mm(3)). Diagnosis of prevalent HIV among those who are at risk but have never been tested should be a priority. However, repeat screening of target populations for incident infection remains important and results in earlier diagnosis. Recent self-reported testing history is not associated with undiagnosed positivity among targeted patients irrespective of the timing of the prior test.