Why is a Large Language Model So Kindly Evil? The Case of GPT-3, an Artificial Intelligence Lab, and the Search giant Nabla
Causality will be hard to prove—was it really the words of the chatbot that put the murderer over the edge? No one will know for sure. But the perpetrator will have spoken to the chatbot, and the chatbot will have encouraged the act. Or maybe a chatbot has ended someones life, so they felt compelled to take their own life? Some machines are making their users depressed. The bot may come with a warning, but dead is not. In 2023, we may well see our first death by chatbot.
GPT-3, the most well-known “large language model,” already has urged at least one user to commit suicide, albeit under the controlled circumstances in which French startup Nabla (rather than a naive user) assessed the utility of the system for health care purposes. Things began well, but deteriorated quickly.
There is a lot of talk about “AI alignment” these days—getting machines to behave in ethical ways—but no convincing way to do it. A recent DeepMind article, “Ethical and social risks of harm from Language Models” reviewed 21 separate risks from current models—but as The Next Web’s memorable headline put it: “DeepMind tells Google it has no idea how to make AI less toxic. None of the other labs is to be fair. Berkeley professor Jacob Steinhardt recently reported the results of an AI forecasting contest he is running: It is moving slower than people thought, and by some measures, it is moving faster than predicted.
In case you missed it, a recently fired tech worker accused the search giant of having a large language model that was sentient. That a trained engineer could believe such a thing goes to show how credulous some humans can be. The large language models are not more than an imitation of a large database of human interaction, but they can easily fool the uninitiated.
Source: https://www.wired.com/story/large-language-models-artificial-intelligence/
The Case for Theory of Mind: How Many Brains Are Needed to Read a Speech? An Empirical Paper by Michal Kosinski
There is no regulation on how these systems are used or how they are supposed to be used, despite the fact that we may see product liability lawsuits after the fact.
Mind reading is common among us humans. Not in the ways that psychics claim to do it, by gaining access to the warm streams of consciousness that fill every individual’s experience, or in the ways that mentalists claim to do it, by pulling a thought out of your head at will. Everyday mind reading is more subtle: We take in people’s faces and movements, listen to their words and then decide or intuit what might be going on in their heads.
It is a known fact that theories of mind and intuitive psychology are associated with the development of many mental disorders. Theory of mind allows us to understand and communicate with one another, while also allowing us to play games and enjoy literature and movies. The capacity is an essential part of being human.
Recently, Michal Kosinski, a psychologist at the Stanford Graduate School of Business, made just that argument: that large language models like OpenAI’s ChatGPT and GPT-4 — next-word prediction machines trained on vast amounts of text from the internet — have developed theory of mind. His studies have not been peer reviewed, but they prompted scrutiny and conversation among cognitive scientists, who have been trying to take the often asked question these days — Can ChatGPT do this? — and move it into the realm of more robust scientific inquiry. How might models change our understanding of ourselves?
The capacity of young children is not being claimed based on anecdotes about your interactions with them, according to Alison Gopnik, a psychology professor at the University of California, Berkeley. “You have to do quite careful and rigorous tests.”