To celebrate the fifth anniversary of the Universal Guidelines for AI, we are reopening the 2018 declaration for signature. Sign up below and join the experts and organizations who helped shape one of the most influential frameworks for the governance of AI.
The Universal Guidelines for AI After 5 Years (2018-2023)
Center for AI and Digital Policy
October 5, 2023
In 2018, computer scientists, human rights experts, NGO advocates, and scientific societies issued the Universal Guidelines for AI, a framework for the regulation of Artificial Intelligence. The Universal Guidelines set out 12 principles to maximize the benefits and minimize the risks of AI. Over the past five years, many of the principles in the UGAI have been adopted in national law and international AI policy frameworks. But new issues have emerged, such as sustainability and gender equity. In the fall of 2023, the EU will be finalizing the EU AI Act and the US will begin work on legislation for Artificial Intelligence.
This conference will explore the impact of the Universal Guidelines on the development of AI policies - what was accomplished, which new issues emerged, and what more needs to be done. The outcome will be an assessment of the progress made, a review of implementation, and a reopening of opportunities for endorsement on the occasion of the fifth anniversary of the Universal Guidelines for AI.
Event Format (in-person, invitations only)
Conference chair: Dr. Lorraine Kisselburgh
We invite submissions addressing:
Format: Maximum of 2,000 words
Citations: Standard format.
Extended Deadline: September 30, 2023 (revised)
Notification: October 2, 2023
Event: October 5, 2023 Submission Email: email@example.com
Prizes: 1st Place: ($1,000), 2nd Place: ($500), 3rd Place: ($250)
The Universal Guidelines for AI were released at the Global Privacy Assembly Conference (formerly ICDPPC), held at the European Parliament in October 2018, with the support and endorsement of Giovanni Buttarelli. Giovanni was the European Data Protection Supervisor, a friend and mentor to many in the privacy world, and a true champion of freedom. He devoted his life's work to the defense of fundamental rights in the digital age.
Explanatory Memorandum and References
The Universal Guidelines on Artificial Intelligence (UGAI) call attention to the growing challenges of intelligent computational systems and proposes concrete recommendations that can improve and inform their design. At its core, the purpose of the UGAI is to promote transparency and accountability for these systems and to ensure that people retain control over the systems they create. Not all systems fall within the scope of these Guidelines. Our concern is with those systems that impact the rights of people. Above all else, these systems should do no harm.
The declaration is timely. Governments around the word are developing policy proposals and institutions, both public and private, are supporting research and development of “AI.” Invariably, there will be an enormous impact on the public, regardless of their participation in the design and development of these systems. And so, the UGAI reflects a public perspective on these challenges.
The UGAI were announced at the 2018 International Data Protection and Privacy Commissioners Conference, among the most significant meetings of technology leaders and data protection experts in history.
The UGAI builds on prior work by scientific societies, think tanks, NGOs, and international organizations. The UGAI incorporates elements of human rights doctrine, data protection law, and ethical guidelines. The Guidelines include several well-established principles for AI governance, and put forward new principles not previously found in similar policy frameworks.
The term “Artificial Intelligence” is both broad and imprecise. It includes aspects of machine learning, rule-based decision-making, and other computational techniques. There are also disputes regarding whether Artificial Intelligence is possible. The UGAI simply acknowledges that this term, in common use, covers a wide range of related issues and adopts the term to engage the current debate. There is no attempt here to define its boundaries, other than to assume that AI requires some degree of automated decision-making. The term “Guidelines” follows the practice of policy frameworks that speak primarily to governments and private companies.
The UGAI speaks to the obligations of “institutions” and the rights of “individuals.” This follows from the articulation of fair information practices in the data protection field. The UGAI takes the protection of the individual as a fundamental goal.
Institutions, public and private, are understood to be those entities that develop and deploy AI systems. The term “institution” was chosen rather than the more familiar “organization” to underscore the permanent, ongoing nature of the obligations set out in the Guidelines. There is one principle that is addressed to “national governments.” The reason for this is discussed below.
These Guidelines should be incorporated into ethical standards, adopted in national law and international agreements, and built into the design of systems.
The elements of the Transparency Principle can be found in several modern privacy laws, including the US Privacy Act, the EU Data Protection Directive, the GDPR, and the Council of Europe Convention 108. The aim of this principle is to enable independent accountability for automated decisions, with a primary emphasis on the right of the individual to know the basis of an adverse determination. In practical terms, it may not be possible for an individual to interpret the basis of a particular decision, but this does not obviate the need to ensure that such an explanation is possible.
The Right to a Human Determination reaffirms that individuals and not machines are responsible for automated decision-making. In many instances, such as the operation of an autonomous vehicle, it would not be possible or practical to insert a human decision prior to an automated decision. But the aim remains to ensure accountability. Thus where an automated system fails, this principle should be understood as a requirement that a human assessment of the outcome be made.
Identification Obligation. This principle seeks to address the identification asymmetry that arises in the interaction between individuals and AI systems. An AI system typically knows a great deal about an individual; the individual may not even know the operator of the AI system. The Identification Obligation establishes the foundation of AI accountability which is to make clear the identity of an AI system and the institution responsible.
The Fairness Obligation recognizes that all automated systems make decisions that reflect bias and discrimination, but such decisions should not be normatively unfair. There is no simple answer to the question as to what is unfair or impermissible. The evaluation often depends on context. But the Fairness Obligation makes clear that an assessment of objective outcomes alone is not sufficient to evaluate an AI system. Normative consequences must be assessed, including those that preexist or may be amplified by an AI system.
The Assessment and Accountability Obligation speaks to the obligation to assess an AI system prior to and during deployment. Regarding assessment, it should be understood that a central purpose of this obligation is to determine whether an AI system should be established. If an assessment reveals substantial risks, such as those suggested by principles concerning Public Safety and Cybersecurity, then the project should not move forward.
The Accuracy, Reliability, and Validity Obligations set out key responsibilities associated with the outcome of automated decisions. The terms are intended to be interpreted both independently and jointly.
The Data Quality Principle follows from the preceding obligation.
The Public Safety Obligation recognizes that AI systems control devices in the physical world. For this reason, institutions must both assess risks and take precautionary measures as appropriate.
The Cybersecurity Obligation follows from the Public Safety Obligation and underscores the risk that even well-designed systems may be the target of hostile actors. Those who develop and deploy AI systems must take these risks into account.
The Prohibition on Secret Profiling follows from the earlier Identification Obligation. The aim is to avoid the information asymmetry that arises increasingly with AI systems and to ensure the possibility of independent accountability.
The Prohibition on Unitary Scoring speaks directly to the risk of a single, multi-purpose number assigned by a government to an individual. In data protection law, universal identifiers that enable the profiling of individuals across are disfavored. These identifiers are often regulated and in some instances prohibited. The concern with universal scoring, described here as “unitary scoring,” is even greater.
A unitary score reflects not only a unitary profile but also a predetermined outcome across multiple domains of human activity. There is some risk that unitary scores will also emerge in the private sector. Conceivably, such systems could be subject to market competition and government regulations. But there is not even the possibility of counterbalance with unitary scores assigned by government, and therefore they should be prohibited.
The Termination Obligation is the ultimate statement of accountability for an AI system. The obligation presumes that systems must remain within human control. If that is no longer possible, the system should be terminated.
Countries represented - 31 (as of Sept 26, 2023)