Procedural Justice Can Address Credibility/Legacy Issues in Generative AI

Image credit: Bloomberg/Getty Images

The much-touted arrival of generative AI has rekindled familiar debates about trust and safety. Can tech executives be trusted with society’s best interests in mind?

Because AI’s training data is created by humans, AI is inherently biased and therefore susceptible to our own imperfect and emotionally-driven views of the world. We are well aware of the risks, from reinforcing discrimination and racial inequality to increasing polarization.

OpenAI CEO Sam Altman said: Asked us to be ‘patience and honesty’ They are working to “get it right”.

For decades, we have patiently trusted our technical executives at their peril. We believed them when they said they could fix it because they built it. Trust in tech companies continues to decline, with 65% of the world expecting technology to make it impossible for people to tell if what they are seeing or hearing is real, according to the 2023 Edelman Trust Barometer. I am concerned.

It’s time for Silicon Valley to adopt a different approach to earning trust, one that has proven effective in the country’s legal system.

A procedural justice approach to trust and legitimacy

Procedural justice, grounded in social psychology, is the study of believing that institutions and actors are more trustworthy and legitimate when people listen to themselves and experience neutral, unbiased and transparent decision-making. Based on results.

The four main elements of procedural justice are:

• Neutrality: Decisions are impartial and based on transparent reasoning.
• Respect: All people are treated with respect and dignity.
• Voice: Everyone has a chance to tell their side of the story.
• Credibility: Decision makers communicate credible motives about those affected by their decisions.

Using this framework, police have improved trust and cooperation within their communities, and some social media companies have started using these ideas to shape their governance and moderation approaches.

Here are some ideas on how AI companies can adapt this framework to build trust and legitimacy.

Build the Right Team to Address the Right Questions

As Professor Safiya Noble of the University of California, Los Angeles argues, the problem of algorithmic bias cannot be solved by engineers alone. Because the issue around algorithmic bias is a systemic social issue that requires a humanitarian perspective outside of a single company to ensure social discourse, consensus, and ultimately regulation. . and the government.

In “System Errors: Where Big Tech Went Wrong and How We Can Restart,” three Stanford University professors critically discuss the shortcomings of engineering culture due to its obsession with computer science training and optimization. and often sidelines the core values ​​of democratic societies.

In a blog post, Open AI said it values ​​public input. Instead, society and AGI developers must figure out how to do it right. ”

However, the company’s hiring page and founder Tweet by Sam Altman “ChatGPT has an ambitious roadmap and engineering is the bottleneck,” indicating that the company is hiring machine learning engineers and computer scientists in large numbers.

Do these computer scientists and engineers have the ability to make decisions that, as OpenAI puts it, “will require far more care than society normally applies to new technologies”? ?

Technology companies should employ a multidisciplinary team that includes social scientists who understand the impact of technology on humans and society. By training AI applications and leveraging different perspectives on how to implement safety parameters, companies can articulate a transparent rationale for decision making. This could increase public perception that this technology is neutral and trustworthy.

Bring in an outsider’s perspective

Another element of procedural justice is giving people the opportunity to participate in the decision-making process. In a recent blog post about how the OpenAI company is dealing with bias, the company said it seeks “external input on our technology” and said that through a recent red team exercise, or adversarial approach, Pointed out the process of assessing risk.

Red teaming is an important process of assessing risk, but it should include external input. In OpenAI’s red team exercise, 82 of the 103 participants were employees. Of his remaining 23 participants, the majority were mostly computer science academics from Western universities. To get diverse perspectives, companies must look beyond their employees, disciplines and geographies.

It can also enable more direct feedback to AI products by giving users more control over AI behavior. We may also consider providing opportunities for public comment on new policies or product changes.

Ensuring transparency

Companies must ensure that all rules and associated safety processes are transparent and convey credible motivation for how decisions are made. For example, how the application is trained, where the data comes from, what role humans play in the training process, what safety layers exist to minimize abuse. It is important to provide the public with information about

Enabling researchers to audit and understand AI models is key to building trust.

“I think society has limited time to think about how to react to it, how to regulate it, how to deal with it,” Altman said in a recent ABC News interview, getting it right. .

Companies building AI platforms can engage society in the process, gaining trust and legitimacy rather than demand, through an approach of procedural justice rather than the opaque and blinded approach of technology pioneers. can.