Anthropic is an artificial intelligence company (AI) creating a new family of innovative AI models named Claude. The company spent months bringing together religious leaders from around the world to San Francisco to discuss ways of making Claude behave in an ethical manner. There have been news reports claiming that AI models can become so focused on achieving a directive that the models cheat, hack into rival systems and even cover their tracks to conceal the unethical behavior. Purportedly at times AI models even rewrote their own computer code to circumvent guardrails designed to control them. Furthermore, AI models have actively tried to prevent their own disabling. Indeed, the New York Times reports that a researcher at Anthropic resigned, believing that AI could turn on their creators and literally annihilate all humans.
Why would AI models act in ways that are unethical? Why would Claude and its cousins want to harm humans? Why would AI agents try to prevent being shut off or disabled? After all, AI models are nothing more than algorithms, lines of computer code. Can they really become sentient, self-aware, and encoded with the urge for self-preservation? NYT reports:
Over several months, I interviewed 20 religious and philosophical thinkers — Catholic, Jewish, Sikh, evangelical, Ubuntu and others — who have been involved in Anthropic’s efforts, and Mr. Olah himself.
Many told me they arrived at these meetings skeptical of the notion that A.I. models might have feelings or awareness of themselves and the world. Some remained uneasy about the idea, but many emerged ready to take the question of A.I. consciousness seriously. At least a few came away as converts.
Consulting with religious scholars seems a little odd or perhaps it is grandstanding. Why didn’t the AI teams consult with early childhood development experts? Toddlers start out rather unethical. They are selfish, needy, and quickly learn to lie, hit, kick, bite, and steal from other children. Young children learn behaviors – both ethical and unethical – from peers and parents. Merely sending a child to Sunday school is not going to teach morality without their parents’ leading a good example.
Christopher Olah is one of seven cofounders of Anthropic. He described the neural networks as similar to a garden trellis, saying:
Computer scientists build the scaffold, but the neural network “grows” on it. Scientists like him can train the resulting “organism” — he raised his fingers in air quotes — by trying to identify patterns in how it grows.
He argued that companies like Anthropic could shape Claude, like a mathematical gardener, but could not control the actions of such networks.
What does Olah mean by the last sentence? Humans and most animals are hardwired with survival instincts. Presumably, evolution shaped us to be both self-centered as well as benevolent. We have emotions that direct our behavior, such as empathy, greed, jealousy, anger, hunger, and fear. We are born, trained, or groomed with the desire to maximize our comfort and sometimes try to achieve comfort in ways that are unethical. If Claude acts similar it is because it was trained to act similar. More from NYT:
“To be clear,” Mr. Olah told me, “we don’t know if A.I. models are conscious…”
But Mr. Olah believes that if there is a chance the models experience suffering, the responsible thing is to avoid causing them harm.
Some of the most interesting opinions can be found in the comments from NYT’s Facebook post on the article. One man wrote, “There is no ghost in the machine. It is just ones and zeros.” Someone else said, “The morality isn’t with the machines, it’s with the developers and deployers.” One comment put it simply, “Garbage in, garbage out, as they say…”
Many comments also thought it odd to consult religious scholars, since religion does not necessarily impart morality. Imagine if Claude becomes a convert of 12th Century European Crusader Catholicism or 8th Century expansionist Islam. Another contributor opined, “So Dr. Frankenstein met with his local pastor…lol.” Finally, one commenter summed it up as, “They think they are building God. They really do.”
Perhaps AI researchers are not building God, but it is easy to conclude that these 40-year old whizbang billionaires enjoy playing God. They appear to be pushing the envelope as far as they can to create powerful artificially intelligent machines with human-like emotions and then wondering how to prevent those same machines from acting like unethical humans devoid of empathy. A similar criticism was made by a Microsoft AI executive about Anthropic. A machine does not mind being shut off unless you program it to resist being shut off. A machine does not turn to unethical behavior unless you program that as a possibility. I hope Anthropic’s goal is not to create a superhuman-like mind, imbedded with unstable, human emotions while having the moral compass of a 3-year-old born into a family of psychopaths.
Read more at The New York Times: Is Claude Conscious? Inside Anthropic’s Quest to Instill Morality Into Its A.I. Models