business, legal, and ethical aspects of artificial
TRANSCRIPT
1Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
ARMA Boise Valley Chapter – October 17, 2019
Business, Legal, and Ethical Aspects of Artificial Intelligence
in Information Governance
2Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
3Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Defining “Artificial Intelligence”
Machinelearning
nting layers of compounding non-linear combinations of variables…
4Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
“A number of technologies under the umbrella of artificial intelligence, such as machine learning,
natural language processing, expert systems (the ability to emulate decision-making of a human expert)
and others, that allow computers to perform things that normally require human intelligence.”
Defining “Artificial Intelligence”
5Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Differences in degree (volume and complexity) Differences in kind (“brute force” versus
independent agency)Renders decisions based on statistical correlations,
not “cause and effect”
Defining “Artificial Intelligence”
6Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Machine Learning “Algorithms that parse data, learn from that data, and then
apply what they’ve learned to make informed decisions” Common example: playlist recommendations, based on other
listeners with similar interests “Fine-tunes” itself with outside input
Deep Learning Subset of machine learning Built-in layer of automated evaluation to constantly get better
on it’s own
Defining “Artificial Intelligence”
7Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
From “Deep Blue” to “AlphaGO”
8Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Ancient Scrolls of Herculaneum
9Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Practical Applications of AI
The best use cases for AI involve seeing things, especially patterns, that humans
can’t or aren’t good at.
10Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Practical Applications of AI
Line Functions Predictive analytics for
sales and marketing Chat Bots for customer
service Knowledge Management
on steroids Contract management
Staff Functions Job applicant screening Auto classification Data remediation Monitoring
communications
11Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Practical Applications of AI
Example 1 (no endorsement intended): Pactum® assists with contract negotiation by
analyzing contract clauses and negotiation points from a variety of sources (legacy contracts, drafts, emails, interviews with employees and clients, etc.) much the same way as AI is used to play games
It can get “smarter” over time May sit on top of a contract management
system
12Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Practical Applications of AI
Example 2 (again, no endorsement intended): Evidence Optix® creates “heat maps”
that identify relevant data custodians and sources and ranks them by relative accessibility
Development scenario: 40 product liability class actions involving 11 products and 2,000 potential custodians world-wide
13Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
IBMs Watson helps Judge
Practical Applications of AI
14Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Practical Applications of AI
Yes, an Israeli small claims court ruledthis was evidence of a lease contract.
15Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Practical Applications of AI
The same ASCII code is rendered very differently by different platforms. AI can accurately interpret it, regardless.
16Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
PredPol Designed to predict when and where crimes will take
place, with the goal of reducing human bias in policing Simulation of PredPol’s algorithm to drug offences in
Oakland, California, repeatedly sent officers to neighborhoods with a high proportion of people from racial minorities, regardless of the true crime rate in those areas
Epic AI Fails in the Law
17Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
COMPAS Algorithm used for bail and sentencing decisions Black defendants were twice as likely to be incorrectly
labeled as higher risk than white defendants 60 % rate of accuracy in COMPAS scores was the same
for black and white defendants, so developers claimed that a test correct in equal proportions for all groups could not be biased
Epic AI Fails in the Law
18Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
“There is not a single instance of AI that truly replicates, let alone beats, human intelligence.” Self-driving cars: not really there yet Narrow AI v. Artificial General Intelligence (AGI)
“AI is not free from bias; the danger is that it can automate bias.” In a Google Images search for “CEO,” just 11 per cent of
the people it displayed were women, even though 27 per cent of the chief executives in the US are female
Critiques of AI
19Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
AI development relies on large data sets, raising privacy and security concerns “Data lakes” typically fall outside of IG
AI, and in particular “deep learning,” cannot be explained in human terms “Transparency” is not a realistic solution
Critiques of AI
20Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
What “Transparency” in AI Looks Like
21Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
The more advanced the AI, and therefore the more accurate it is, the harder it is to explain.
(But when someone asks you to “explain yourself” do you explain how the neurons in your brain are
triggered?)
The Transparency Tradeoff
22Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Direct v. indirect explainability, or procedural v. substantive explainability Can a human understand the machine’s “reasoning” Can a human understand the machine’s decision,
regardless of the “reasoning” behind it?
“Explainability”
23Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
The appropriate level of explanation depends on several factorsWho does the decision affect? How severely does it impact them?What recourse to they have to challenge the decision? Is liability for a poor decision judged on a “strict” or
“fault” basis?
“Explainability”
24Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Laws Affecting the Implementation of AI
General Data Protection Regulation (GDPR) GDPR Article 22 states in relevant part that individuals “have the right not
to be subject to a decision based solely on automated processing.” Even if a legally binding right to explanation is embedded in the text of the
GDPR itself, the right may only apply in limited circumstances (e.g., when a negative decision is solely automated and has legal or other similar significant effects)
The right to an explanation essentially means that users can demand data underlying algorithmic decisions made about them, including in recommendation systems, credit and insurance risk systems, advertising programs, and social networks
25Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Laws Affecting the Implementation of AI
California Consumer Privacy Act (CCPA) of 2018Will take effect on January 1, 2020, with enforceability
to begin July 1, 2020, or six months after publication of the implementing regulations, whichever comes first
Certain provisions require organizations to provide consumers with information regarding the processing of their data during the preceding 12-month period
26Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Laws Affecting the Implementation of AI
Various laws restrict or prohibit the use of certain categories of information, e.g.: Title VII of the Civil Rights Act of 1964 The Equal Credit Opportunity Act (“ECOA”) The Fair Credit Reporting Act The Americans with Disabilities Act The Age Discrimination in Employment Act (“ADEA”)
27Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Laws Affecting the Implementation of AI
Various laws restrict or prohibit the use of certain categories of information, e.g.: Fair Housing Act (“FHA”) Genetic Information Nondiscrimination Act (“GINA”) Health Information Portability and Accountability Act
(HIPAA)
28Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Laws Affecting the Implementation of AI
Various laws restrict or prohibit the use of certain categories of information, e.g.: 4th Amendment (for government actors) Fair Trade and Antitrust laws, e.g.: Search rankings that promote their own products Paid prioritization on search results Social media algorithms trained to promote particular points
of view
29Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Laws Affecting the Implementation of AI
New laws are being proposed every day August 17, 2018: “Proposed Private Right of Action by
New York City’s Automated Decision Systems Task Force” Initiative seeks “a law providing a private right of
action for individuals or groups of individuals that are injured by automated decision system determinations”
30Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
The Sedona Conference Working Group 11 on Data Security and Privacy Liability
Next meeting: March 18-19, 2020 in Denver
Proposed Principles for AI Implementation
Nicholas Economou James Sherer Jason Baron
31Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Proposed Principles for AI Implementation
Principle 1. Any organization that deploys an AI process that significantly impacts others should consider adoption of policies or procedures that provide for a measure of explainability to the subjects of AI decisions. Anticipate demands for explanations from regulators, the courts,
and the general public Have disclosure policies and procedures in place before you are
asked AI system developers must provide understandable documentation
– its an essential part of the project
32Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Principle 2. The level of explainability required depends on the sensitivity of the AI decision being made. “Sensitivity” includes evaluation of the potential impact
of the decision on data subjects, the data inputs used, and the level of human involvement in the decision-making process
Proposed Principles for AI Implementation
33Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Principle 3. Depending on the sensitivity of the AI decision, explainability includes some form of explanation to individuals impacted by the decision and available remedies.
How AI is being used How AI was developed and “trained” Goal or purpose of AI application Kinds of data used Specific data used in a decision How a decision was made Remedies Statistical data on impact
Proposed Principles for AI Implementation
34Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Proposed Principles for AI Implementation
Principle 4. Organizations should adopt an appropriate form of disclosure to subjects of AI decisions depending on the sensitivity of the decision at issue.Notices Information available upon requestRetained information
35Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Proposed Principles for AI Implementation
Principle 5. The ultimate responsibility of ensuring explainability rests with the individual or organization who uses the AI algorithm to make decisions.Software developersEnd users
36Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Proposed Principles for AI Implementation
Principle 6. At least in circumstances involving the most sensitive AI decisions, a role exists for outside observers to provide input. Public accountability Periodic audits
37Business, Legal, and Ethical Aspects
of AI in Information Governance© 2019October 17, 2019
Business, Legal, and Ethical Aspects of Artificial Intelligence
in Information Governance
Ken Withers, Deputy Executive [email protected]