About | HeinOnline Law Journal Library | HeinOnline Law Journal Library | HeinOnline

35 Berkeley Tech. L.J. 113 (2020)
Artificial Intelligence Opinion Liability

handle is hein.journals/berktech35 and id is 125 raw text is: ARTIFICIAL INTELLIGENCE OPINION LIABILITY
Yavar Bathaeet
Opinions are not simply a collection of factual statements-they are something more.
They are models of reality that are based on probabilistic judgments, experience, and a
complex weighting of information. That is why most liability regimes that address opinion
statements apply scienter-like heuristics to determine whether liability is appropriate, for
example, holding a speaker liable only if there is evidence that the speaker did not subjectively
believe in his or her own opinion. In the case of artificial intelligence, scienter is problematic.
Using machine-learning algorithms, such as deep neural networks, these artificial intelligence
systems are capable of making intuitive and experiential judgments just as humans experts do,
but their capabilities come at the price of transparency. Because of the Black Box Problem, it
maybe impossible to determine what facts or parameters an artificial intelligence system found
important in its decision making or in reaching its opinions. This means that one cannot simply
examine the artificial intelligence to determine the intent of the person that created or deployed
it. This decouples intent from the opinion, and renders scienter-based heuristics inert,
functionally insulating both artificial intelligence and artificial intelligence-assisted opinions
from liability in a wide range of contexts. This Article proposes a more precise set of factual
heuristics that address how much supervision and deference the artificial intelligence receives,
the training, validation, and testing of the artificial intelligence, and the a priori constraints
imposed on the artificial intelligence. This Article argues that although these heuristics may
indicate that the creator or user of the artificial intelligence acted with scienter (i.e.,
recklessness), scienter should be merely sufficient, not necessary for liability. This Article also
discusses other contexts, such as data bias in training data, that should also give rise to liability,
even if there is no scienter and none of the granular factual heuristics suggest that liability is
DOI: https://doi.org/10.15779/Z38P55DH32
© 2020 Yavar Bathaee.
t Litigator and computer scientist. This Article is dedicated to my wife, Jacqueline, and
my children, Elliot and Audrey. I would like to thank James Steiner-Dillon for his comments
on the Article and his support. All errors and omissions are my own.

What Is HeinOnline?

HeinOnline is a subscription-based resource containing nearly 3,000 academic and legal journals from inception; complete coverage of government documents such as U.S. Statutes at Large, U.S. Code, Federal Register, Code of Federal Regulations, U.S. Reports, and much more. Documents are image-based, fully searchable PDFs with the authority of print combined with the accessibility of a user-friendly and powerful database. For more information, request a quote or trial for your organization below.

Short-term subscription options include 24 hours, 48 hours, or 1 week to HeinOnline with pricing starting as low as $29.95

Contact us for annual subscription options:

Already a HeinOnline Subscriber?

profiles profiles most