Breakpoint
Author: John Stonestreet and Dr. Glenn Sunshine
Last month, following a series of dire warnings from AI insiders that the technology could bring about the end of humanity, six industry executives gathered at the White House to work out an agreement on regulations. Their proposal included four layers of voluntary controls and audits, including internal controls to prevent hacking and unintended access, internal teams to provide oversight and monitoring of the intended standards, independent external auditors and evaluators, and an independent committee to receive reports and ensure any problems are resolved.
All of which sounds good. However, before we celebrate what this means for the future of humanity, we should, as Confucius warned, remember some history.
In the 1990s, energy giant Enron adopted a lengthy written code of ethics, established an Audit and Compliance Committee, activated internal compliance procedures, and conducted annual reviews of related-party deals. Each year, Arthur Anderson, one of the top five accounting firms in the country, audited the company to ensure compliance. Each year, Enron passed the audit with flying colors. Fortune magazine even listed the company as among the most innovative firms in the country.
As it turns out, Enron was most innovative in how it got around the rules. They were able to hide their losses until the business collapsed in 2001, bringing Arthur Anderson down with them. Enron followed all the ethical procedures in their handbook yet still committed fraud. Significantly, its leaders were evangelical Christians.
Similar stories could be told of Tyco, WorldCom, and Lehman Brothers. Ethics policies are essential, but they are also worthless if the people involved are unethical. When it comes to AI, let’s hope that history is not rhyming.
The dominant ethical theory among AI companies is a philosophy called “Effective Altruism,” or EA, a form of utilitarianism that attempts to use reason and evidence to determine how to maximize the good done for every dollar spent. This is done by multiplying the probability times the size of the expected benefit, which requires that everything is quantified.
Samuel Bankman-Fried, who “orchestrated one of the largest financial frauds in history,” is an advocate of this philosophy. He explained EA this way:
There’s something very quantitative about how you view the world. . .. There’s this really big notion of, in the end, you turn things into numbers, and you decide which number is the biggest.
EA tends to put long-term benefits ahead of immediate ones. For example, if something will cost people their lives today but benefit future generations, it should be done. Similarly, global benefits outweigh national benefits, so national interests can be sacrificed if that would benefit the world at large. Also, EA takes the non-human world into account and, because it isn’t clear that humans are more important than crustaceans, shrimp welfare is, for many advocates, as urgent a project as any.
EA has led to a number of failed analyses. For example, as a low-cost intervention to improve lives in Africa, one EA project supplied villages with insecticide permeated nets to prevent the spread of malaria. However, the nets were repurposed for fishing, possibly polluting the water and definitely contributing to overfishing, particularly as the tight mesh caught juvenile fish. This was not included in EA’s calculations, which points to how difficult it is to fully anticipate all the potentials and ramifications of decisions.
Samuel Bankman-Fried is currently sitting in a federal prison for his fraud to maximize profits, ostensibly so he could give more to good causes. The collapse of FTX and the Future Fund hurt investors and sent several people to jail. EA organizations have disavowed FTX, but Bankman-Fried was one of its most prominent supporters.
EA has all the problems of utilitarianism multiplied by the hubris of those who think that everything can be quantified and anticipated because of how smart they are. Unintended consequences will inevitably follow. Also, just like Enron, even with better intentions and practices, ethical failures have occurred in EA-inspired companies.
In and of themselves, the guidelines agreed upon at the White House will be inadequate to keep AI companies honest. People must be honest and virtuous, and especially, made accountable. Effective altruism cannot replace virtue. Short of a genuine commitment to a higher ethical standard, the foxes are guarding the henhouse.