In January 2012, I wrotea about the past and future of artificial intelligence (AI). I reiterated Bill Joy's 2000 question: Does the future need us? Little did I know then that a revolution was already brewing. By 2011, GPUs had accelerated considerably the training of deep neural networks, finally making a technology whose roots go back to the early 1940sb competitive. By 2011–2012, AlexNet, a deep neural network, won several international competitions, launching the deep-learning revolution. A decade later, generative AI, which refers to AI that can generate novel content rather than simply analyze or act on existing data, has become all the rage. Over the past few weeks, ChatGPT, the "newest kid on the generative-AI block," is practically everywhere.
Media reporting of these new AI technologies often focuses on their societal risk, and various proposals have been put forward aiming at containing the risk of AI. In October 2022, the U.S. Office of Science and Technology Policy (OSTP), which is part of the Executive Office of the President of the United States, published a Blueprint for an AI Bill of Rights. This bill identified five principles that "should guide the design, use, and deployment of automated systems to protect the American public in the age of artificial intelligence." In November 2022, the ACM Technology Policy Council released a statement on principles for responsible algorithmic systems, identifying nine principles "intended to foster fair, accurate, and beneficial algorithmic decision-making."
Principlism is an approach developed in biomedical ethics that uses a framework of universal ethical principles that should underlie biomedical decisions. Its recent rise in computing has, however, been criticized:c "they are isolated principles situated in an industry and education system which largely ignores ethics; and they are toothless principles which lack consequences and adhere to corporate agendas."
Consider the ACM policy statement. The statement addresses system builders and operators, AI system developers, and operators of AI systems. In other words, the statement is not about responsible algorithmic systems, but about responsible people and corporations. Nevertheless, as I have previously pointed out,d ACM has been reluctant to address the unethical behavior of technology corporations and their leaders, even when these corporations blatantly violate ACM's Code of Ethics and Professional Conduct.
OSTP's mission is to "maximize the benefits of science and technology to advance health, prosperity, security, environmental quality, and justice for all Americans." OSTP, however, is a governance body, not a philosophy department. Governance happens via actions, either by executive actions or congressional bills. So far, OSTP has shared nothing about a plan to turn the AI Bill of Rights into an actionable policy.
Worries about societal harm caused by AI are not new. About a decade ago, the philosopher Nick Bostrom worried about the existential risk of super-intelligent AI. In his "Paperclip Maximizer," thought experiment he hypothesized about a super-intelligent agent whose goal is to maximize the number of paperclips in its collection. In its zeal to accomplish its mission, the agent may transform "first all of earth and then increasing portions of space into paperclip manufacturing facilities." The purpose of this experiment was to demonstrate that an artificial agent with apparently innocuous values could pose an existential threat to humanity. The value-alignment approach is a response to the super-intelligent-agent risk. AI-alignment research aims to steer AI systems toward their designers' intended goals and interests. In his 2021 report, Our Common Agenda, the UN Secretary General called for ensuring AI is "aligned with shared global values."
AI agents, however, are developed by tech corporations. Our fundamental problem is not paperclip maximization but unregulated profit maximization. Adam Smith's argument in 1759 in favor of the invisible hand of the free market was to "advance the interests of the society as a whole." Since then the argument that unregulated profit maximization advances the interests of the society as a whole, for example, by former U.S. Federal Reserve chairman, Alan Greenspan, has been shown to fail both theoreticallye and practically.f While profit maximization has led to some impressive societal benefits, for example, mRNA vaccines, it has also led to serious adverse consequences, for example, financial crises. In general, technology advances faster than regulation. The unregulated use of AI in targeted advertising and content moderation has brought deep polarization to our society, seriously threateningg democracy.
So let us stop talking about Responsible AI. We, computing professionals, should all accept responsibility now, starting with ACM!
©2023 ACM 0001-0782/23/03
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and full citation on the first page. Copyright for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or fee. Request permission to publish from email@example.com or fax (212) 869-0481.
The Digital Library is published by the Association for Computing Machinery. Copyright © 2023 ACM, Inc.
Along with Ricardo Baeza Yates, I was one of the people who led the recent statement on principles for responsible algorithmic systems. I agree that building responsible systems means people acting responsibly and holding corporations responsible.
I don't believe that principles alone are sufficient. ACM itself may not be able to give them sufficient "teeth" in terms of enforcement, but policy makers pay a lot of attention and look to professional societies like ACM as a trusted voice when crafting legislation or regulation or standards that could have more real bite. Europe is ahead of the US in putting some teeth into AI regulation and privacy regulation, but many people see that more is needed in the US as well.
Principles documents also give working engineers something to point to internally when they argue for responsible behavior.
I would encourage folks to check out the USTPC Hot Topic Webinar from December 1 2022, "Toward a fAIr Future: Algorithmic Responsibility in the New Machine Age" for more information.
Principles are not an ending point, but an important starting point. Its the long haul of follow-through and engagement at critical moments of shaping standards, regulation and law where the ACM and other groups professional, community-based, advocacy, research will continue to be very needed, and have already had significant impact. Having voices involved who are not tech lobbyists is critical to moving ethical standards and principles into defined practices and ultimately, legal requirements. I think at this point, few will disagree with high-level principles, which is a victory in itself; the challenge now is methodology, implementation, and enforcement mechanisms (and penalties/redress when legal requirements arent met).
The good news is that we are seeing principles documents like the Blueprint for an AI Bill of Rights move into policy and law. As an example: the AI Bill of Rights came out October 2022. Last month (Feb 2023), the President released Executive Order 14091, Further Advancing Racial Equity, which moves principles from the AI Bill of Rights to a *legal requirement* that federal agencies must follow, including in their procurement practices which impact the private sector significantly.
The executive order puts forth specific directives related to artificial intelligence and equity, including a definition of algorithmic discrimination taken directly from the AI Bill of Rights, and a clear statement that when designing, developing, acquiring, and using artificial intelligence and automated systems in the Federal Government, agencies shall do so in a manner that advances equity. The EO directs agencies to consult their civil rights offices on decisions regarding the design, development, acquisition, and use of artificial intelligence and automated systems. Further, it directs agencies to affirmatively advance civil rights, including by protecting the public from algorithmic discrimination, one of the five protections the Blueprint lays out.
Again, this EO is directed at federal agencies use of algorithmic systems and artificial intelligence, not at the broader private sector. But increasingly there are opportunities to make the case that EOs like this are a first step, and supporting legislation must follow. For instance, earlier this week, the Senate Committee on Homeland Security and Government Affairs held hearings on the Risks and Opportunities of Artificial Intelligence, providing the opportunity for speakers to make the case that Congress needs to act to governing AI, drawing on the AI bill of Rights and NISTs AI Risk Management Framework (which ACM members also provided important input into) as sources.
This is all to say that the work doesnt start or end with a single document or statement, but is ongoing and there are an increasing number of opportunities for ACM colleagues to intervene.
I am very happy to learn that the aI Bill of Rights has gained more teeth with EO 14091!
Displaying all 3 comments