|
2025 will go down as the year in which the debate over the integration of artificial intelligence in all aspects of personal and professional life hit its stride. Both the federal government and state governments were locked in battles over AI regulation, and, in December, Time Magazine identified the “Architects of AI” as their “Person of the Year” for 2025. Of course, AI did not suddenly emerge on the scene in 2025, nor even with the release of ChatGPT in November 2022. I had commissioned three colleagues at the Markkula Center for Applied Ethics in 2014 to specialize in AI and its ethical dilemmas. They have become well-known experts in AI and ethics, actively engaged in the national and global debate. In some ways, arguments regarding how AI should be developed and deployed are merely a continuation of the ethical debates over the introduction of robots on the factory floor and computers in white-collar offices. The capability of our machines has grown steadily at least since World War II. This puts pressure on employment and plays a role in changing our workplaces and occasionally our personal lives as well. But the debate over AI is focused more on its impacts in our personal lives, or at least as much as it is on its economic and employment impacts. The debate over AI began in earnest in November 2022 with the introduction of ChatGPT, the first widely available large language model (LLM) that could answer queries of all kinds, write computer code, and manipulate images. The past year, 2025, saw the effective dates of the first attempts to regulate the technology. In my own life, I find myself comparing AI to three earlier innovations that changed the way I worked: copying machines, fax machines, and computers. (Yes, I am that old!) In 1971, I bought one of the first small Xerox copying machines for a nonprofit I had just launched. It was a wonder. No more cutting mimeograph masters and no more mimeo smell, though the distinctive smell of coated Xerox paper replaced it. Plain paper copiers came later. Fax machines were even more of a wonder. From the early 1980s, you did not need to rush to the post office to catch the 5 p.m. pickup so someone could have something you wrote in only 2-3 days. To send a fax, you simply went down to the FedEx office and paid $1.50 per page for them to send it to a FedEx office near your recipient where they could pick it up—the same day! Personal fax machines and plain paper came later. And computers! I bought six early Macintosh computers in 1988 for another nonprofit I headed, and then comically hard-wired them all together over four offices so they could communicate with each other and share a printer. Computers later brought email, which I resisted for a year before realizing it would not go away; the Stanford Business School in 1993 forced all of us on the faculty to be available to students by email. My nine-year-old daughter had sent her first email before I did. Every new technology has been accompanied by an unavoidable set of ethical questions. Were there safety issues associated with the new technology? Did they replace employees and make unemployment worse? Whose responsibility was it to train employees to upgrade their skills to use new technologies? Who captures the value created by new technologies and rising productivity? Later technologies have raised a much broader set of ethical questions, particularly technologies that impact our personal lives—for example the internet and social media over the past 15 years. How are humans who use the new technology affected by ergonomic issues and, more recently, mental health issues? How does widespread adoption affect the environment? Does the new technology expand or worsen the wealth gap? What occurred in 2025 is that a better understanding of the potential for AI—for both good and bad—has stimulated its rapid implementation—and a much more sophisticated debate over whether and how to regulate and guide its development. Federal regulators, as one would expect in the era of President Trump, have been inclined to let AI develop unimpeded. “Don’t kill the goose that lays the golden eggs” is a familiar argument from the recent development of the internet and social media. Nonetheless, states have led the way in 2025 in AI regulation, as have regulators at the European Union (EU). A new California law targets “frontier” AI programs by requiring safety assessments and reports on how such programs comply with existing safety standards. Other California laws focus on “AI companions,” systems designed to interact emotionally or to simulate human relationships. California law also requires notice whenever one is interacting with AI and reasonable protocols to detect signs of potential self-harm. New York now has similar laws. Its AI Companion law came into effect in November 2025. A New York law aimed at frontier models will go into effect on January 1, 2027. An additional New York law attempts to require disclosure when deepfakes are used in advertising and seeks to reduce their use. Even Texas has new AI laws. The Texas Responsible AI Governance Act, effective January 1, 2026, establishes a structure for governing AI in the future, prohibits certain uses with harmful intents and effects, and requires some disclosures to consumers when they are interacting with AI. The EU has followed its tendency to be more comprehensive in regulating new technologies, including AI. The EU has based its law on the risk that specific AI models represent. As in other areas, the EU has been committed to the “precautionary principle” of delaying the introduction of new technologies until they are proven safe. The EU’s primary AI Act was adopted in 2024, and its implementation has been occurring gradually over 2025 and 2026. It separates new AI systems by the risk they represent to the public, heavily regulating the highest risk applications such as those used in employment and hiring, determining credit-worthiness and insurance eligibility, those employed in health care and medical devices, law enforcement, border control, and migration. In these high-risk areas, the AI Act requires risk management systems, high-quality data regarding the training of the AI systems, human oversight, technical documentation, robust cybersecurity measures, and post-marketing monitoring. Several other EU laws also affect AI deployment. Given that AI regulation is still in its infancy, I think there are five dimensions of AI and ethics that will preoccupy us in 2026. First, we must contend with AI’s flaws. Its tendency to hallucinate false answers and statements, and its sometimes harmful dialogues with users consulting it for mental health or relationship advice, must be addressed with more effective regulation and/or consensus guardrails that protect vulnerable individuals. The “AI companion” laws are just a start. Second, we must address the job displacement and employment impacts of AI. Like with prior technologies, workers will be displaced. There is the possibility that many more workers will lose their jobs than from any previous technology. This makes a vague reliance on “new jobs will be created by the technology” an inadequate answer. At the same time, it would be unfair to impose complete responsibility for displacement on the architects or users of AI. The solution will likely be a combination of company and government-funded training programs that upskill employees who are being replaced, and the creation of other opportunities, including programs to train skilled workers as electricians and mechanics who are desperately needed to fill tens of thousands of jobs. Third, the voracious need for energy and water supplies to run AI data centers must be addressed. The need for mass AI processing will revolutionize the markets for energy and water use, and the interests of the public must be protected. Fourth, we need to consider what AI is doing to our own thinking and writing skills. Is overreliance on AI weakening our ability to resolve problems in our own lives as well as in our work? Should students be encouraged to master the use of AI—or forced to work without it? In writing my current book on the history of the responsible corporation, AI has helped me check to see if I have missed anything critical to the story I am telling—but I would not rely on it to write my chapters! Fifth, we must decide how to manage the new concentration of wealth and power in the hands of a small number of AI creators and entrepreneurs. It appears these individuals will have extraordinary power from a greater concentration of wealth than even the Gilded Age of the late 1800s. California will likely vote in November 2026 on a 5% “wealth tax” on all billionaires paid at 1% per year. Do billionaires wield too much economic and political power? What really are the dangers of having the first trillionaire (likely Elon Musk)? Engagement with these five critical ethical issues raised by artificial intelligence ought to be on everyone’s ethical agenda for 2026. If AI hit its stride in professional and personal life in 2025, then we must make sure the ethics of AI hits its stride in 2026.
|
