The Un-Modeled World: Law and the Limits of Machine Learning

tags

Predictive Analytics

Notes

innovation operates as a self-corrective —a built-in feature that limits the ability of technology to fully objectify human life and law error-free.

NOTER_PAGE: (2 . 0.3126022913256956)

many legal scholars mistakenly presuppose the unlimited expansionary power of a machine’s capacity to learn. It is as if one observes the moon landing, and then immediately anticipates full colonization of the universe on the basis of a “natural” and “ineluctable” trajectory of human achievement without ever considering the physical limitations of time and space.

NOTER_PAGE: (2 . 0.6202945990180033)

machine learning occurs when two conditions are satisfied: learnable regularity and contextual invariance

NOTER_PAGE: (3 . 0.2905073649754501)

unplanned innovations guarantee that perfect predictability in law and elsewhere cannot be achieved.

NOTER_PAGE: (3 . 0.6636661211129297)

The pervasive concern in law that unchecked ascendance of a technical rationality will degrade law’s human and intersubjective character may to some extent be true, but many would suggest that this is desirable in legal domains that are relatively static and sufficiently predictable

NOTER_PAGE: (4 . 0.472176759410802)

Martin Heidegger

NOTER_PAGE: (5 . 0.09819967266775778)

Technological instrumentality is temporal, but life is intra-temporal.

NOTER_PAGE: (6 . 0.18821603927986907)

Herbert Marcuse

NOTER_PAGE: (6 . 0.22422258592471359)

Technology is, at its core, a social project of purpose-fulfillment. The purpose is the domination of nature and the improvement of living conditions.

NOTER_PAGE: (6 . 0.8592471358428806)

scientific rationality will extinguish itself as a social project of worth once humans no longer need to perform “socially necessary but individually repressive labor.”

NOTER_PAGE: (7 . 0.49018003273322425)

technology’s ascent is therefore limited by its ability to deliver on that promise.

NOTER_PAGE: (8 . 0.10065466448445172)

Jürgen Habermas

NOTER_PAGE: (8 . 0.1783960720130933)

Over time, legitimization through language and “talking things through” came to be replaced with empirical rationales. Slowly, technology began displacing humans as the legitimizers of economic and state action.

NOTER_PAGE: (8 . 0.4386252045826514)

Superstructure is no longer determined by deliberation among humans. It, too, is empirically rationalized.

NOTER_PAGE: (8 . 0.6153846153846154)

Old-style politics defined itself in relation to practical goals; today’s politics is defined in terms of solutions to technical problems.

NOTER_PAGE: (8 . 0.7618657937806874)

People can, and often do, stand in the way, which generates a need for the de-politicization of mass populations. Habermas contends that this is accomplished by replacing the bourgeois ideology of free exchange with an ideology of scientific and technological legitimization

NOTER_PAGE: (8 . 0.781505728314239)

A world in which language has been absorbed by technology is a world denatured.

NOTER_PAGE: (9 . 0.12684124386252046)

Technological rationality is not a means after all. It is a process of provocation because it is temporal and we are not.

NOTER_PAGE: (9 . 0.6194762684124386)

II. The Un-Modeled World

NOTER_PAGE: (9 . 0.7054009819967267)

if humans change, then the law must change, too.

NOTER_PAGE: (10 . 0.10147299509001637)

Valiant’s two assumptions imply three hard boundaries for the use of machine learning in law: (1) inherently small sample sizes, which limit the ability to observe a pattern; (2) the reflexive behavior of humans and a tendency toward outsmarting rules, which accomplishes the same; and (3) our contemporary state of perpetual innovation and dynamism, which breaks down both patterns and contexts, especially over time.

NOTER_PAGE: (10 . 0.3657937806873977)

Small Sample Sizes

NOTER_PAGE: (10 . 0.5114566284779051)

“Empirically minded legal academics are often impressed with large data sets when it is the quality of hypotheses that matters most.”

NOTER_PAGE: (11 . 0.5188216039279869)

Reflexive Behavior

NOTER_PAGE: (12 . 0.18003273322422259)

Humans act in anticipation of the future. Anticipatory and interdependent action distorts learning because people modulate their behavior in response to others.

NOTER_PAGE: (12 . 0.21685761047463176)

Previous observations are unhelpful when alternative rules, policies, and legal environments are introduced. They must be recalibrated, on the basis of a model of the world, in order to retain their predictive value

NOTER_PAGE: (12 . 0.3625204582651391)

Innovation

NOTER_PAGE: (12 . 0.8142389525368249)

We will abide by Galileo’s maxim to “measure what is measurable, and make measurable what is not so,” but what is made measurable will be made un-measurable again and again.

NOTER_PAGE: (14 . 0.7978723404255319)

Conclusion

NOTER_PAGE: (15 . 0.24386252045826515)

The predictability of all things is a mechanical delusion.

NOTER_PAGE: (15 . 0.279050736497545)