IEEE Technology and Society Magazine - March 2022 - 72

Feature
A primary issue is that in rules-based legal systems,
such value judgments about the law become fixed
in computer code. Moreover, these choices become
embedded in technology where their impact can be
magnified when the software is distributed widely.
For example, the software model created by Turbotax
reflects a series of subjective interpretations
about the meaning of the U.S. income tax code
made by the employees of the Intuit corporation.
Their interpretive choices about the meaning of the
personal income tax laws become embedded in the
software itself. The impact of these interpretations
become magnified when Turbotax's software is distributed
to millions of users worldwide who implicitly
adopt this value-laden model.
In sum, the formal representation of law in a legal
technological system may reflect a series of subtly
encoded values and subjective decisions that may be
difficult to detect. Moreover, when these values are
embedded as software rules, the impact of these subjective
judgments can become magnified when the
software model is distributed and adopted broadly.
Removing prior structural constraints
A final way in which values can be subtly embedded
in legal technological systems has to do with the
reduction of existing technological constraints by
artificial intelligence.
Observe that sometimes a new technology will
make an activity that used to be difficult, suddenly
much easier or less expensive, or significantly reduce
transaction costs. In some instances, simply making
some activity technologically easier to do than it
was before may be subtly value laden. Consider the
court record example from the earlier discussion. In
the era of paper documents, court filings were more
difficult to access. The constraints of paper technology
meant that finding private information in a large
number of public court documents was prohibitive.
Accessing court documents in that pre-electronic
era typically required physical entry to the courthouse
and then cumbersome, manual searching of
volumes of papers to find particular private information.
By digitizing court documents and putting them
online, the activities of accessing and searching for
private data within voluminous court filings, as an
unintended byproduct, became dramatically less
difficult.
In the prior technological era, we can think of paper
technology as having implicitly protected the privacy
72
of litigants by providing a structural constraint. The
subsequent digital technology removed this implicit
constraint, and as a side effect, diminishing litigant
privacy. A similar effect has occurred with the digitization
of other government records such as deeds, voter
registration documents, and campaign contributions,
which were always nominally public, but in the prior
paper technological era, practically difficult to access
and analyze. Digitizing these public documents, and
putting them online has as a byproduct, reduced privacy,
because this technology makes it much easier to
access previously disparate public data about citizens
and cross-reference, aggregate, and centralize this
information (e.g., linking an individual's campaign
contribution, housing purchase, and vehicle registration
public records) than in the past.
Overall, any new technology that makes some
activity substantially easier than it used to be may
have the effect of unintentionally diminishing some
value (such as privacy) which was implicitly protected
by the constraints inherent in the prior technological
era. The central idea is that sometimes
technical cost, inefficiency, or " friction " can be
understood as playing a non-obvious implicit functional
role in protecting societal values. When a
new technology, such as machine learning removes
or reduces frictions or inefficiencies that also were
serving a (perhaps unrecognized) functional role in
fostering values in the previous technological context,
the diffusion of the technology be understood
as widely shifting certain embedded societal values
(e.g., efficiency and inferring correlations) at the
expense of others (e.g., privacy and anonymity).
This can be seen within the legal domain. Artificial
intelligence techniques such as machine
learning make particular activities that used to be
difficult-such as detecting patterns in large data
sets or aggregating previously disparate data, much
easier. For instance, it is now often possible to
probabilistically infer private information about a
person that has not been publicly revealed-such
as sexual orientation-simply by scanning publicly
available data, such as social network connections,
and engaging in statistical analysis [28]. Such capabilities
may be particularly problematic in law,
where machine-learning techniques now make it
significantly easier to surreptitiously detect correlates
(e.g., zip codes) for categories that may be
unlawful to consider directly (e.g., race and sexual
orientation).
IEEE Technology and Society Magazine
IEEE Technology and Society Magazine - March 2022

Table of Contents for the Digital Edition of IEEE Technology and Society Magazine - March 2022
