Polanyi’s Paradox and the Shape of Employment Growth

Income Distribution, September 2014

In 1966, the philoso­pher Michael Polanyi observed, “We can know more than we can tell… The skill of a dri­ver can­not be replaced by a thor­ough school­ing in the the­ory of the motor­car; the
knowl­edge I have of my own body dif­fers alto­gether from the knowl­edge of its phys­i­ol­ogy.” Polanyi’s  obser­va­tion largely pre­dates the com­puter era, but the para­dox he identified—that our tacit  knowl­edge of how the world works often exceeds our explicit under­stand­ing— fore­tells much of the  his­tory of com­put­er­i­za­tion over the past five decades. This paper offers a con­cep­tual and empir­i­cal  overview of this evo­lu­tion. I begin by sketch­ing the his­tor­i­cal think­ing about machine dis­place­ment of human labor, and then con­sider the con­tem­po­rary incar­na­tion of this displacement—labor mar­ket polar­iza­tion, mean­ing the simul­ta­ne­ous growth of high‑ edu­ca­tion, high-wage and low‑education, low‑wages jobs—a man­i­fes­ta­tion of Polanyi’s para­dox. I dis­cuss both the explana­tory power of the polar­iza­tion phe­nom­e­non and some key puz­zles that con­front it. I then reflect on how recent advances in arti­fi­cial intel­li­gence and robot­ics should shape our think­ing about the likely tra­jec­tory of occu­pa­tional change and employ­ment growth. A key obser­va­tion of the paper is that jour­nal­ists and expert com­men­ta­tors over­state the extent of machine sub­sti­tu­tion for human labor and ignore the strong com­ple­men­tar­i­ties. The chal­lenges to sub­sti­tut­ing machines for work­ers in tasks requir­ing adapt­abil­ity, com­mon sense, and cre­ativ­ity remain immense. Contemporary com­puter sci­ence seeks to over­come Polanyi’s para­dox by build­ing machines that learn from human exam­ples, thus infer­ring the rules that we tac­itly apply but do not explic­itly understand.