In the absence of legal guidelines, companies will need to establish internal processes for using AI responsibly. Oscar Wong/Moment via Getty Images
The rush to adopt powerful new generative AI technologies such as ChatGPT has raised vigilance about potential harm and abuse. The law’s dispassionate response to such threats requires companies developing these technologies to deploy AI “ethically.”
But what does that mean exactly?
The simple answer is to align enterprise operations with one or more sets of dozens of AI ethical principles created by governments, multiple stakeholder groups, and academics. But that is easier said than done.
Over the course of two years, we and our colleagues interviewed and surveyed AI ethics experts from various fields to find out how they are trying to achieve ethical AI and what they are missing. I tried to understand why. We believe that pursuing AI ethics in the field is not about mapping ethical principles to corporate behavior, but about implementing management structures and processes that enable organizations to detect and mitigate threats. I learned
This could be disappointing news for organizations looking for clear guidance around the gray areas, and for consumers who want clear and protective standards. But it shows a growing understanding of how companies can pursue ethical AI.
Fighting ethical uncertainty
Our research is the basis for an upcoming book focused on managing AI ethics issues at leading companies that use AI. In late 2017 and early 2019, we interviewed 23 such managers. Their titles ranged from privacy officers and privacy advisors to data ethics officers, a novelty at the time but becoming more common today. Four main takeaways emerged from our conversations with these AI ethics managers.
First, while there are many benefits to using AI in business, there are also significant risks, and businesses know it. AI ethics managers expressed concerns about privacy, manipulation, bias, opacity, inequality, and labor migration. In a well-known example, Amazon has developed an AI tool that classifies resumes and trained it to find candidates similar to those it has hired in the past. Male dominance in the tech industry meant that most of Amazon’s workforce was male. In response, the tool has learned to reject female candidates. Unable to resolve the issue, Amazon was ultimately forced to abandon the project.
Generative AI raises additional concerns about mass misinformation, hate speech, and intellectual property misappropriation.
Second, companies pursuing ethical AI do so primarily for strategic reasons. They want to maintain trust among their customers, business partners and employees. And they want to stay ahead of, or prepare for, new regulations. The Facebook and Cambridge Analytica scandal alleges that Cambridge Analytica uses Facebook user data shared without their consent to infer users’ psychological types and target them with manipulative political advertising. We have shown how unethical use of analytics can and can even eviscerate a company’s reputation. , bring it down, as is the case with Cambridge Analytica itself. The companies we spoke to wanted instead to be seen as responsible stewards of people’s data.
The challenge faced by AI ethics managers was to find the best way to achieve “ethical AI”. They first looked to AI ethical principles, especially those rooted in bioethics and human rights principles, but found that they were not enough. It’s not just that there are many competing principles. Principles such as justice, fairness, charity, and self-government were contested, subject to interpretation, and could conflict with each other.
This gave us our third point. The manager needed more than her high-level AI principles to decide what to do in a given situation. One AI ethics manager explained that they are trying to translate human rights principles into a set of questions that developers can ask themselves in order to create more ethical AI software systems. “I stopped after 34 pages of questions,” the manager said.
Fourth, experts addressing ethical uncertainty have looked to organizational structures and procedures to arrive at judgments about what to do. Some of these were clearly inadequate. However, while most are still in development, some are more useful, such as:
- Hire an AI ethics officer to build and oversee the program.
- Establish an internal AI Ethics Committee to discuss and make decisions on difficult issues.
- Create a data ethics checklist and require frontline data scientists to complete the checklist.
- Encourage academics, former regulators and advocates to seek alternative perspectives.
- Conduct algorithmic impact assessments of the type already used in environmental and privacy governance.
Ethics as responsible decision-making
Key ideas that emerged from our research include: Companies seeking to use AI ethically should not expect to discover a simple set of principles that will lead to the correct answer from an omniscient God’s perspective. Instead, we need to focus on the very human task of trying to make responsible decisions in a world of finite understanding and changing circumstances, even if some decisions are imperfect.
In the absence of clear legal requirements, companies, like individuals, should be aware of how AI impacts people and the environment, and stay abreast of public concerns and the latest research and expert ideas. I can only do my best. It also allows us to seek input from a large and diverse set of stakeholders and to take seriously high-level ethical principles.
This simple idea changes conversations in important ways. This suggests that rather than AI ethics experts focusing their energies on identifying and applying AI principles, AI principles are still part of the story, and the decision-making structures and processes they employ. , to ensure that it considers the impacts, perspectives and public expectations with which information should be provided. their business decisions.
Ultimately, we believe that laws and regulations should provide a substantive standard for organizations to aim for. But responsible decision-making structures and processes are a starting point, and over time should help build the knowledge needed to create protective, enforceable and substantive legal standards.
In fact, new AI laws and policies are focused on process. New York City has passed a law requiring companies to audit AI systems for harmful biases before using them to make hiring decisions. Members of Congress introduced legislation that would require companies to conduct algorithmic impact assessments before using AI for lending, hiring, insurance, or other consequential decisions. These laws emphasize the process of proactively addressing many AI threats.
Some generative AI developers take a completely different approach. His CEO of OpenAI, Sam Altman, initially said that in releasing ChatGPT to the general public, the company aimed to “expose the chatbot sufficiently to the real world so that we can uncover examples of exploitation that we might not have thought of.” I explained. It allows us to build better tools. ” For us, it is not responsible AI. We treat humans as guinea pigs for dangerous experiments.
Altman’s call for government regulation of AI at a Senate hearing in May 2023 signals a growing awareness of the issue. But we think he’s gone too far in shifting the responsibility that the developers of generative AI should have to the government. Maintaining public trust and avoiding harm to society requires companies to take their responsibilities more seriously than ever before.
Dennis Hirsch, Professor of Law and Computer Science. Director of the Program on Data and Governance. Core Faculty TDAI, Ohio State University and Piers Norris Turner, Associate Professor of Philosophy and PPE Coordinator.Ethics and Human Values Center Director Ohio State University
This article is republished from The Conversation under a Creative Commons license. Please read the original article.
