Welcome back to TPI’s Research Roundup, our semi-regular compilation of recent outside research of interest to tech policy nerds. If you’ve read a paper you think might be interesting to include in the next roundup, feel free to send it to [email protected]
DISCLAIMER: The papers and authors are not affiliated with TPI. We do not necessarily agree with everything, or even anything, in these papers, but find them interesting and informative.
Financial Distancing: How Venture Capital Follows The Economy Down and Curtails Innovation by Sabrina T. Howell, Josh Lerner, Ramana Nanda, and Richard R. Townsend
What it studies: The impact of COVID-19 on venture capital funding
What they find: Early-stage venture funding declined in March, with 50 to 100 fewer deals per week than at the same time last year. Late-stage venture deals have not fallen as much, and a record amount of committed but uninvested capital before the crisis suggests that the early-stage downturn has more to do with increased risk than a decrease in the supply of capital.
Why it matters: As venture capital funded firms tend to be more innovative, venture capital investing may provide some early insights into the effects of the pandemic on innovation.
What it studies: The impact of various social distancing interventions on the 1918 Flu pandemic
What they find: In 1918, closing schools, prohibiting gatherings, and quarantining infected people caused a large decrease in the ratio of the peak death rate to the average death rate, but only a small decline in the cumulative death rate. In other words, those social distancing measures “flattened the curve” but did not reduce total deaths. The authors suspect the actions did not reduce total deaths because the social distancing measures were not in place for very long (schools, for example, were only closed for an average of 36 days)
Why it matters: Even a short period of social distancing can flatten the curve, but reducing mortality requires more time.
Thirty-Six Views of Copyright Authorship, by Jackson Pollock by Dan L. Burk
What it studies: How copyright should treat creations by artificial intelligence and machine learning algorithms.
What they find: Authorship of works created by an AI should be evaluated “in terms of causation, intent, volition, and related doctrines that are familiar from other areas of law.” They reach this conclusion by thinking through 36 examples of how the law would determine authorship of paintings by Jackson Pollock and those inspired by him. The examples range from the very simple, with Pollock himself painting on a canvas, to trickier examples like accidental creation, coerced creation, deterministic machine creation, and, finally, an AI trained on Jackson Pollock paintings creating original paintings. The author suggests that latter example is akin to a painter being inspired by Pollock, with authorship credit best assigned to the programmers of the AI.
Why it matters: As AI becomes more advanced, questions of authorship credit belonging to the AI, the programmers of the AI or to the author of the data the AI is trained on will become more complex.
An Economic Approach to Regulating Algorithms by Ashesh Rambachan, Jon Kleinberg, Sendhil Mullainathan, and Jens Ludwig
What it studies: How regulations can reduce algorithmic bias
What they find: Algorithms will yield discriminatory outcomes if any discriminatory humans exist, even if regulations restrict what criteria can be used as inputs. Instead, requiring the algorithm’s predictive rules to be revealed can drive discrimination to zero even with discriminatory humans.
Why it matters: This theory paper suggests that it is possible to have nondiscriminatory predictive algorithms with the right rules.