Machines learn from biology, the speed of coronavirus and how to build an ethical self-driving ca

Here are links to a few recent articles by LML External Fellow Mark Buchanan.
Machines Learn from Biology (Nature Physics 16, 238; 6 March 2020). The number of connected devices is projected to be over 70 billion by 2025. Until now, our computing devices have remained within the scope of our oversight and maintenance, but that will increasingly become practically impossible. Much future computing will begin to emphasize energy use over performance, and many devices will come to live on their own, unattended or aided by us, essentially as independent entities. Not quite organisms, but certainly on their way to becoming like them.
To Beat Coronavirus, Win the Containment Battle (Bloomberg Opinion, 25 February, 2020). New research (as of mid-February) suggested that the virus is far more contagious than initially believed. Early estimates of the basic reproductive number— a key epidemiological figure that reflects the number of new cases, on average, resulting from a single infection in a fully susceptible population — looked to be in the range of 2 to 3. But several new studies of the epidemic’s earliest stages have reported significantly larger numbers, from 4 to 7. These higher numbers suggested nearly one month ago that the simplest measures for stopping new outbreaks — quarantine of infected individuals and tracking the people showing symptoms were previously in contact with — would likely be insufficient. It would take more aggressive social-distancing methods, such as closing schools, discouraging travel and locking down cities. Unfortunately, events of the past three weeks have made this clear.
The Certainty of Uncertainty (Nature Physics 16, 120; 6 February 2020). Climate scientists have not managed to convince much of the public of the urgency of responding to global warming.  Of course, much of the delay is the result of a decades-long propaganda campaign funded by fossil fuel interests. But scientists could do better in communicating uncertainty as well. In a recent study, researchers undertook an experiment with more than 1,000 individuals testing how they responded to messages about climate risks which conveyed uncertainty in different ways. One positive finding: people didn’t naively prefer messages expressed with no uncertainty, but were more receptive to a message with “bounded uncertainty” listing upper and lower bounds. For most people, discussions of clearly defined uncertainties make scientists more trustworthy.
How to Build an Ethical Self-Driving Car (Bloomberg Opinion, 6 February 2020). Self-driving cars, whenever they begin to arrive in large numbers, will face difficult ethical dilemmas. Should the car save its passengers, if that means sacrificing some nearby pedestrians? A recent survey asked 40,000 people in different nations how cars should respond, building an empirical picture of human ethics. But a better way to identify reliable rules, some experts argue, is to combine the survey-based approach with analysis based on prevailing ethical theories. One might start with public views but put these through the filter of ethical theory to see if a rule is truly defensible. Ethicists refer to views that survive this test as “laundered preferences.” For example, all ethical theories would reject preferences for one gender over another, even though the survey found such preferences in some regions. In contrast, preferences to save the largest number of people would survive, as might a preference for the very young over the very old. In this obviously messy area, policies will have to be guided by some mixture of the empirical and the theoretical. When a self-driving car makes a choice and kills some children, people will want to know how it made the decision. And the rules had better survive systematic ethical scrutiny.

Leave a Reply

Your email address will not be published. Required fields are marked *