About | HeinOnline Law Journal Library | HeinOnline Law Journal Library | HeinOnline
LOG IN
LOG IN

42 Admin. & Reg. L. News 7 (2016-2017)
Improving the Administrative State with Machine Learning

handle is hein.journals/admreln42 and id is 137 raw text is: 8. The Federal government should partner with the industry to ensure that connected vehicles are sufficiently protected from hacking and cyber attacks. 9. The Federal government should partner with the industry to determine how to do the inter-manufacturer testing that will be necessary for market certification. 10. The Federal government needs to spec a uniform method for emer- gency services to stop or disable an automated vehicle. There are far more than these ten things to do, but we can all start working on these things now. Through every step of the process, manufacturers should work closely with government to ensure that the process will actually work, without having to take steps backward. Improving the Administrative State with Machine Learning Cary Coglianese* & David Lehr** Machine learning refers to computer algorithms that, on their own, discover patterns in historical datasets, which can then be used for making predictions. Such algorithms undergird important inno- vations throughout the private sector, making possible high-frequency trading, online video and retail recom- mendations, and self-driving cars, among other applications. A prolif- eration of seemingly autonomous decision-making tools has raised alarms over the potential loss of human auton- omy. Particularly when government agencies begin to rely more extensively on machine-learning algorithms to support decisions previously made by human officials, the wisdom and legality of machine learning will likely come under increased attention. Can agencies use artificial intelligence while remaining faithful to principles of law and consonant with government of the people and by the people? The answer is basically yes. Administrative agencies have numer- ous opportunities to improve their operations through the responsible application of machine-learning technology. Contrary to growing * Edward B. Shils Professor of Law and Professor of Political Science, University of Pennsylvania Law School; Director, Penn Program on Regulation. ** Research Fellow, Georgetown University Law Center; Research Affiliate, Penn Program on Regulation;J.D. Candidate,Yale Law School, 2020. alarmism, governments can actually use algorithms in important ways without contravening constitutional and administrative law doctrines. We have recently offered an extensive legal analysis of government use of machine- learning algorithms in our article, Regulating by Robot: Administrative Decision Making in the Machine-Learning Era, 105 GEO. L.J. 1147 (2017). Here we highlight what machine learning promises government agencies and why standard doctrines should not form insuperable barriers to governmental use of machine-learning algorithms. Machine Learning's Administrative Promise At an intuitive level, machine- learning algorithms are those that learn from the data. But what does this really mean? It means these algorithms find patterns or correlations between variables in a set of data, which can then be used to make predictions. Importantly, this discovery is done on the algorithms' own; humans do not explicitly program them to look for certain patterns. As a result, it can be difficult to explain exactly how or why a machine-learning algorithm keys in on certain correlations or makes the predictions that it does. In other words, machine-learning algorithms are often considered black boxes. Furthermore, and in contrast with some conventional techniques like regression analysis, machine Summer 20177 learning is not used to support causal inferences about the learned relation- ships between different variables. Nevertheless, just as machine learning is leading private-sector improvements, its ability to make sense of large quan- tities of data can help agency officials make smarter decisions, allocate scarce administrative resources more wisely, and improve the accuracy and fairness of governmental processes. Government agencies are already exploring how to use machine learn- ing. The Environmental Protection Agency (EPA) has used machine learning to forecast chemical toxic- ity. The Internal Revenue Service (IRS) prioritizes audits based on machine-learning predictions of tax violations. The Securities and Exchange Commission (SEC) similarly relies on machine learning to identify potential instances of insider trading. These are just a few examples, but they show that agen- cies are already embracing artificial intelligence. That said, current administrative applications are limited. Many of these applications assist with discre- tionary aspects of enforcement, such as deciding which facilities to inspect. These existing uses are unlikely to be problematic from the standpoint of administrative law because they involve algorithms as decision- support tools to inform actions that are committed to agency discretion. But in the future, federal agencies Administrative & Regulatory Law News Summer 2017

What Is HeinOnline?

HeinOnline is a subscription-based resource containing thousands of academic and legal journals from inception; complete coverage of government documents such as U.S. Statutes at Large, U.S. Code, Federal Register, Code of Federal Regulations, U.S. Reports, and much more. Documents are image-based, fully searchable PDFs with the authority of print combined with the accessibility of a user-friendly and powerful database. For more information, request a quote or trial for your organization below.



Short-term subscription options include 24 hours, 48 hours, or 1 week to HeinOnline.

Purchase Access

Contact us for annual subscription options:

Contact Us

Already a HeinOnline Subscriber?

Log In

profiles profiles most