A market for trustworthy AI could gain traction if government and/or nongovernmental entities were able to grade or otherwise certify AI systems for trustworthy attributes. Evidence from other public-private certification projects suggests that transparency and clear evaluation metrics are key to trust and adoption. To the extent applicable, certification could be based on existing metrics, frameworks, and standards developed by NIST and national or international bodies.
For instance, under the ENERGY STAR® program, which is administered by the United States Environmental Protection Agency (EPA) and Department of Energy (DOE), companies may voluntarily seek certification to display the ENERGY STAR label on those products that meet strict performance requirements for energy efficiency.313 This labeling provides a way for “consumers and businesses who want to save energy and money” to do so by choosing products with the ENERGY STAR label, thereby relying on a recognizable and trustworthy information mechanism.314 To date, ENERGY STAR has achieved widespread adoption, leading to substantial energy and consumer savings.315 Likewise, the Leadership in Energy and Environmental Design (LEED) program, led by the non-profit U.S. Green Building Council (USGBC), allows green building projects to earn a certification (platinum, gold, silver, or certified) based on adherence to certain environmental metrics.316 Per USGBC, LEED projects have been adopted worldwide.317 Programs like ENERGY STAR and LEED empower their users (e.g., individuals, businesses) to make informed choices,318 guide regulators and lawmakers,319 and more generally help build community trust.320 Certification could even provide the basis for liability safe harbors, should those be created by legislation, to encourage participation in the certification process, in appropriate cases.
Such a process for AI systems could contribute to a functioning market for trustworthy AI. While issues remain about whether such certification programs should be led by government or non-governmental entities (or both), certification programs could enlarge the marketplace for trustworthy AI by bridging information and knowledge gaps. However, a major challenge to establish certifications, as one commenter observed, is the difficulty in gaining sufficient legitimacy and credibility.321 BBB National Programs, which itself administers industry certifications, notes that effective certification mechanisms have consistent and verifiable standards and transparency markers (e.g., “trust marks, annual reports, or consumer complaint processes”), among other characteristics.322 We agree with the comment from twenty-three attorneys general that transparency around the evaluation process is critical and certification programs should operate “through transparent and verifiable policies and practices driven by appropriate standards including a code of ethics.”323
Establishing and promoting certification systems can further the development of a trustworthy marketplace for AI.324 More abundant and reliable information of the type discussed in Section 3 above can make it easier to generate public trust in AI, AI evaluations, and AI certifications.325
313 ENERGY STAR, How ENERGY STAR Works.
314 See ENERGY STAR, About ENERGY STAR. See also ENERGY STAR, ENERGY STAR Impacts.
315 See Id. (“Since 1992, ENERGY STAR and its partners helped prevent 4 billion metric tons of greenhouse gas emissions from entering our atmosphere; By choosing ENERGY STAR, a typical household can save about $450 on their energy bills each year and still enjoy the quality and performance they expect; Approximately 1,700 manufacturers and 1,200 retailers partner with ENERGY STAR to make and sell millions of ENERGY STAR certified products.”).
316 See U.S. Green Building Council, LEED Rating System.
317 See U.S. Green Building Council, Press Room (noting “more than 185,000 total LEED projects worldwide” “and “more than 185 countries and territories with LEED projects” and “more than 205,000 LEED professionals around the world.”). See also Twenty-three Attorneys General Comment at 3-4 (“As an example of a private sector program, the [LEED] standard has spurred the move towards ‘green buildings.’”).
318 See, e.g., ENERGY STAR, About ENERGY STAR (“The blue ENERGY STAR label provides simple, credible, and unbiased information that consumers and businesses rely on to make well-informed decisions.”) (emphasis added).
319 See, e.g., The Policing Project at New York University’s School of Law Comment at 2 (“Before LEED, there was no mechanism to incentivize this type of information-surfacing about buildings’ environmental impact. Thanks to the information surfaced by LEED certification, lawmakers now have an objective standard against which they can tie the development of building regulations.”).
320 See Twenty-three Attorneys General Comment at 3-4 (referencing Energy Star and LEED in the context of “agile and dynamic public and civic initiatives that build trust and spur trusted technological changes.”).
321 Friedman et al., supra note 73, at 748. In particular, in the context of private certification programs of technology used by police, the Policing Project’s study found that “institutional trust in policing agencies and Big Tech is low, especially from communities most impacted by policing tech, such as Black communities.” Id. at 746. Here, Policing Project’s law review article advises that transparency in certification schemes themselves is crucial to building trust. Id. at 748-49.
322 See BBB National Programs Comment at 3. In addition to “consistent standards” (which includes verifiability) and “transparency,” BBB National Programs highlights additional characteristics it believes are key for an “effective and accountable independent certification mechanisms” to demonstrate: “defined areas of responsibility[,]” “oversight and independent review[,]” “regulatory recognition[,]” and “layers of accountability.” Id.
323 See Twenty-three Attorneys General Comment at 4 (emphasis added).
324 See BBB National Programs Comment at 3 (noting that several of the characteristics are important in the development of a marketplace, including by bringing consistency and reducing friction). See also id. at 5 (arguing that “[t]his type of certification-based system with a trusted mark and standardized reporting can serve a vital role in building a trustworthy AI marketplace.”) (referencing the BBB National Programs and the Center for Industry Self-Regulation’s Principles for Trustworthy AI in Recruiting and Hiring and accompanying Hiring and Independent Certification Protocols for AI-Enabled Hiring and Recruiting Technologies).
325 See, e.g., Johnson & Johnson Comment at 4 (“Developing a framework to enhance the explicability of AI systems that support decision-making on socially significant issues, such as healthcare, is a component of building societal trust… Central to a supportable framework is the ability for individuals to obtain a factually correct, and generally clear explanation of the decision-making process”); AI Policy and Governance Working Group Comment at 2 (“Moving quickly to address risks concerning AI systems and tools will not only provide accountability, it will promote the trust of the American public.”); AI Impacts Comment at 2. Cf. Gary Marchant et al., Governing Emerging Technologies Through Soft Law: Lessons for Artificial Intelligence, 61 Jurametrics J. 1, 9 (2020). (“The biggest deficits of soft law programs…relate to their effectiveness and credibility. Their provisions are often phrased in broad and general terms, making compliance difficult to objectively determine, especially without any type of reporting or monitoring requirement.”).