Operational Listening and Computational Eugenics

Seminar/Forum

Operational Listening and Computational Eugenics

Operational Listening: Mark Andrejevic

This pre­sen­ta­tion takes as its start­ing point the work of Harun Farocki and Trevor Paglen on the rise of the ​'oper­a­tional image' to con­sider the related rise of what might be described as auto­matic or ​'oper­a­tional­ised' lis­ten­ing. This type of lis­ten­ing is becom­ing increas­ingly famil­iar thanks to the deploy­ment of smart speak­ers and a grow­ing array of net­worked audio sen­sors (from gun­shot sen­sors to work­place mon­i­tors, smart phones, and audio sur­veil­lance on public trans­port).

The talk describes the stakes of oper­a­tional­ism as the dis­place­ment of sym­bolic inter­pre­ta­tion by action, and draws on psy­cho­an­a­lytic theory to con­sider the impli­ca­tions for sub­jec­tiv­ity. The goal of the talk is to raise the ques­tion regard­ing what is lost in the shift from com­pre­hen­sion and inter­pre­ta­tion to oper­a­tion. What does it mean to say that Alexa will gather infor­ma­tion about your words but it doesn’t know what you mean (beyond purely oper­a­tional com­mands)? The talk makes some spec­u­la­tive claims about the emer­gence of a world in which sym­bolic effi­ciency is replaced by oper­a­tional effi­ciency. This is, per­haps need­less to say, a fun­da­men­tally unde­mo­c­ra­tic process that is already becom­ing all-too famil­iar in these post-truth, post-delib­er­a­tive, post-polit­i­cal times.

Computational Eugenics: Jake Goldenfein

Over the past decade, researchers have been inves­ti­gat­ing new tech­nolo­gies for cat­e­goris­ing people based on phys­i­cal attrib­utes alone. Unlike pro­fil­ing with behav­ioural data cre­ated by inter­act­ing with infor­ma­tional envi­ron­ments, these tech­nolo­gies record and mea­sure data from the phys­i­cal world (i.e. signal) and use it to make a deci­sion about the ​‘world state’ – in this case a judge­ment about a person.

Auto­mated per­son­al­ity analy­sis and auto­mated per­son­al­ity recog­ni­tion, for instance, are grow­ing sub-dis­ci­plines of com­puter vision, com­puter lis­ten­ing, and machine learn­ing. This family of tech­niques has been used to gen­er­ate per­son­al­ity pro­files and assess­ments of sex­u­al­ity, polit­i­cal posi­tion and even crim­i­nal­ity using facial mor­pholo­gies and speech expres­sions. These pro­fil­ing sys­tems do not attempt to com­pre­hend the con­tent of speech or to under­stand actions or sen­ti­ments, but rather to read per­sonal typolo­gies and build clas­si­fiers that can deter­mine per­sonal char­ac­ter­is­tics.

While the knowl­edge claims of these pro­fil­ing tech­niques are often ten­ta­tive, they increas­ingly deploy a vari­ant of ​‘big data epis­te­mol­ogy’ that sug­gests there is more infor­ma­tion in a human face or in spoken sound than is acces­si­ble or com­pre­hen­si­ble to humans. This paper explores the bases of those claims and the sys­tems of mea­sure­ment that are deployed in com­puter vision and lis­ten­ing. It asks if there is some­thing new in these claims beyond ​‘big data epis­te­mol­ogy’, and attempts to under­stand what it means to com­bine com­pu­ta­tional empiri­cism, sta­tis­ti­cal analy­ses, and prob­a­bilis­tic rep­re­sen­ta­tions to pro­duce knowl­edge about people.

All Seminar/Forums