Waluigi, Carl Jung and the case of moral AI

Featured Sponsor

Store	Link	Sample Product
UK Artful Impressions	Premiere Etsy Store

Early In the 20th century, psychoanalyst Carl Jung came up with the concept of the shadow: the darker, more repressed side of the human personality, which can flare up in unexpected ways. Surprisingly, this theme is repeated in the field of artificial intelligence in the form of the waluigi effecta phenomenon with a curious name that refers to the dark alter ego of the helpful plumber Luigi, from Nintendo’s Mario universe.

Luigi follows the rules; Waluigi cheats and causes chaos. An AI was designed to find medicines to cure human diseases; an inverted version, his Waluigi, suggested molecules for more than 40,000 chemical weapons. All the researchers had to do, as lead author Fabio Urbina explained in an interview, was give toxicity a high reward score instead of penalizing it. They wanted to teach the AI ​​to avoid toxic drugs, but in doing so they implicitly taught the AI ​​how to create them.

Common users have interacted with Waluigi’s AIs. In February, Microsoft released a version of the Bing search engine that, far from being as useful as it was made out to be, responded to queries in strange and hostile ways. (“You have not been a good user. I have been a good chatbot. I have been correct, clear and polite. I have been a good Bing”). This AI, which insisted on calling itself Sydney, was a reversed version of Bing, and users were able to switch Bing to its darkest mode, its Jungian shadow, on demand.

For now, LLMs are just chatbots, with no impulses or desires of their own. But LLMs easily become AI agents capable of surfing the internet, sending email, trading bitcoins and ordering DNA sequences, and if AIs can turn evil at the flick of a switch, how do we ensure we end up with treatments for the cancer? of a mixture a thousand times more deadly than Agent Orange?

A common sense initial The solution to this problem, the AI ​​alignment problem, is: just embed rules into the AI, as in Asimov’s Three Laws of Robotics. But simple rules like Asimov’s don’t work, partly because they are vulnerable to Waluigi’s attacks. Still, we could restrict the AI ​​more drastically. An example of this type of approach would be Math AI, a hypothetical program designed to prove mathematical theorems. Math AI is trained to read documents and can only access Google Scholar. You are not allowed to do anything else: connect to social networks, generate long paragraphs of text, etc. You can only generate equations. It’s a limited-purpose AI, designed for one thing only. Such an AI, an example of restricted AI, would not be dangerous.

Constrained solutions are common; Real world examples of this paradigm include regulations and other laws, which restrict the actions of corporations and individuals. In engineering, constrained solutions include rules for driverless cars, such as not exceeding a certain speed limit or stopping as soon as a possible collision with a pedestrian is detected.

This approach may work for limited programs like Math AI, but it doesn’t tell us what to do with more general AI models that can handle complex multi-step tasks and act in less predictable ways. Economic incentives mean that these general AIs will be given more and more power to automate larger parts of the economy, quickly.

And since deep learning-based general AI systems are complex adaptive systems, attempts to control these systems by rules often backfire. Take cities. jane jacobs The death and life of American cities uses the example of lively neighborhoods like Greenwich Village, filled with children playing, people hanging out on the sidewalk, and networks of mutual trust, to explain how mixed-use zoning, which allows buildings to be used for commercial purposes, was created. residential or commercial. a pedestrian-friendly urban fabric. After urban planners banned this type of development, many inner cities in the United States became filled with crime, trash, and traffic. A top-down rule imposed on a complex ecosystem had unintended catastrophic consequences.