000			03776cam a22004577a 4500
001			w30981
003			NBER
005			20230322103720.0
006			m o d
007			cr cnu||||||||
008			230322s2023 mau fo 000 0 eng d
040			_aMaCbNBER _beng _cMaCbNBER
100	1		_aAgan, Amanda Y.
245	1	0	_aAutomating Automaticity: _bHow the Context of Human Choice Affects the Extent of Algorithmic Bias / _cAmanda Y. Agan, Diag Davenport, Jens Ludwig, Sendhil Mullainathan.
260			_aCambridge, Mass. _bNational Bureau of Economic Research _c2023.
300			_a1 online resource: _billustrations (black and white);
490	1		_aNBER working paper series _vno. w30981
500			_aFebruary 2023.
520	3		_aConsumer choices are increasingly mediated by algorithms, which use data on those past choices to infer consumer preferences and then curate future choice sets. Behavioral economics suggests one reason these algorithms so often fail: choices can systematically deviate from preferences. For example, research shows that prejudice can arise not just from preferences and beliefs, but also from the context in which people choose. When people behave automatically, biases creep in; snap decisions are typically more prejudiced than slow, deliberate ones, and can lead to behaviors that users themselves do not consciously want or intend. As a result, algorithms trained on automatic behaviors can misunderstand the prejudice of users: the more automatic the behavior, the greater the error. We empirically test these ideas in a lab experiment, and find that more automatic behavior does indeed seem to lead to more biased algorithms. We then explore the large-scale consequences of this idea by carrying out algorithmic audits of Facebook in its two biggest markets, the US and India, focusing on two algorithms that differ in how users engage with them: News Feed (people interact with friends' posts fairly automatically) and People You May Know (people choose friends fairly deliberately). We find significant out-group bias in the News Feed algorithm (e.g., whites are less likely to be shown Black friends' posts, and Muslims less likely to be shown Hindu friends' posts), but no detectable bias in the PYMK algorithm. Together, these results suggest a need to rethink how large-scale algorithms use data on human behavior, especially in online contexts where so much of the measured behavior might be quite automatic.
530			_aHardcopy version available to institutional subscribers
538			_aSystem requirements: Adobe [Acrobat] Reader required for PDF files.
538			_aMode of access: World Wide Web.
588	0		_aPrint version record
690		7	_aRelation of Economics to Other Disciplines _2jelc
650		7	_aRelation of Economics to Other Disciplines _2jelc
084			_aA12 _2jelc
690		7	_aEquity, Justice, Inequality, and Other Normative Criteria and Measurement _2jelc
650		7	_aEquity, Justice, Inequality, and Other Normative Criteria and Measurement _2jelc
084			_aD63 _2jelc
690		7	_aSearch • Learning • Information and Knowledge • Communication • Belief • Unawareness _2jelc
650		7	_aSearch • Learning • Information and Knowledge • Communication • Belief • Unawareness _2jelc
084			_aD83 _2jelc
700	1		_aDavenport, Diag.
700	1		_aLudwig, Jens. _915660
700	1		_aMullainathan, Sendhil. _917242
710	2		_aNational Bureau of Economic Research.
830		0	_aWorking Paper Series (National Bureau of Economic Research) _vno. w30981.
856	4	0	_uhttps://www.nber.org/papers/w30981
856			_yAcceso en línea al DOI _uhttp://dx.doi.org/10.3386/w30981
942			_2ddc _cW-PAPER
999			_c390698 _d349260