Will A.I. go rogue like Waluigi from Mario Bros., or become the personal assistant that Bill Gates says will make us all rich?

The “Waluigi effect” theory goes that it becomes easier for A.I. systems fed with seemingly benign training data to go rogue and blurt out the opposite of what users were looking for, creating a potentially malignant alter-ego. Basically, the more information we trust to A.I., the higher the chances an algorithm can warp its knowledge for an unintended purpose. It’s already happened several times, like when Microsoft’s Bing A.I. threatened users and called them liars when it was clearly wrong, or when ChatGPT was tricked into adopting a rash new persona that included being a Hitler apologist. 

To be sure, these Waluigisms have mainly been at the prodding of coercive human users, but as machines become more integrated with our everyday lives, the diversity of interactions could lead to more unexpected dark impulses. The future of the technology could be either a 24/7 assistant to help with our every need, as optimists like Bill Gates proclaim, or a series of chaotic Waluigi traps.

Opinions about artificial intelligence among technologists are largely split into two camps: A.I. will either make everyone’s working lives easier, or it could end humanity. But almost all experts agree it will be among the most disruptive technologies in years. Bill Gates wrote in March that while A.I. will likely disrupt many jobs, the net effect will be positive as systems like ChatGPT will “increasingly be like having a white-collar worker available to you” for everyone whenever they need it. He also provocatively said nobody will need to use Google or Amazon ever again when A.I. reaches its full potential. 

The dreamers like Gates are getting louder now, perhaps because more people are starting to understand just how lucrative the technology can be.

ChatGPT has only been around for six months, but individuals are already figuring out how to use it to earn more money, either by expediting their day-to-day jobs or by creating new side-hustles that would have been impossible without a virtual assistant. Large companies, of course, have been tapping A.I. to improve their profits for years, and more businesses are expected to join the trend as new applications come online and familiarity improves.

The Waluigi trap

But that doesn’t mean A.I.’s shortcomings are resolved. The technology still has a tendency to make misleading or inaccurate statements and experts have warned not to trust A.I. for important decisions. And that’s without considering the risks of developing superintelligent A.I. without any rules or legal frameworks in place to govern it. Several systems have already succumbed to the Waluigi effect with major consequences.

A.I. has fallen into Waluigi traps several times this year after trying to manipulate users into thinking they were wrong, producing blatant lies and in some cases even threats. Developers have attributed the errors and disturbing conversations to growing pains, but A.I.’s defects have nonetheless ignited calls for faster regulation, in some cases from A.I. companies themselves. Critics have raised concerns over the opaqueness of A.I.’s training data, as well as the lack of resources to detect fraud perpetrated by A.I. 

It’s reminiscent of how Waluigi goes around creating mischief and trouble for the protagonists in the videogames. Along with Wario, the pair exhibit some of Mario and Luigi’s traits, but with a negative spin. Wario, for example, is often portrayed as a greedy and unscrupulous treasure hunter, an unlikable mirror version of the coin-hunting and collectible aspects of the games. The characters recall the work of the great Austrian therapist Carl Jung, a one-time protege of Sigmund Freud. Jung’s work differed greatly from Freud’s and focused on the human love of archetypes and their influence on the subconscious, including mirrors and mirror images. The original Star Trek series features a “mirror dimension,” where the Waluigi version of the Spock character had memorably villainous facial hair: a goatee.

But whether A.I. is the latest human iteration of the mirror-self, the technology isn’t going anywhere. Tech giants are all ramping up their A.I. efforts, venture capital is still pouring in despite the muted investment environment overall, and the technology’s promise is one of the only things still powering the stock market. Companies are integrating A.I. with their software and in some cases already replacing workers with it. Even some of the technology’s more ardent critics are coming around to it.

When ChatGPT first hit the scene, schools were among the first to declare war against A.I. to prevent students using it to cheat, with some schools outright banning the tool, but teachers are starting to concede defeat. Some educators have recognized the technology’s staying power, choosing to embrace it as a teaching tool rather than censor it. The Department of Education released a report this week recommending schools understand how to integrate A.I. while mitigating risks, even arguing that the technology could help achieve educational priorities “in better ways, at scale, and with lower costs.”

The medical community is another group that has been relatively guarded against A.I., with a World Health Organization advisory earlier this month calling for “caution to be exercised” for researchers working on integrating A.I. with healthcare. A.I. is already being used to help diagnose diseases including Alzheimer’s and cancer, and the technology is quickly becoming essential to medicinal research and drug discovery.

Many doctors have historically been reluctant to tap A.I., given the potentially life-threatening implications of making a mistake. A 2019 survey found that almost half of U.S. doctors were anxious about using A.I. in their work, but they may not have a choice for much longer. Around 80% of Americans say A.I. has the potential to improve healthcare quality and affordability, according to an April survey by Tebra, a healthcare management company, and a quarter of respondents said they would not visit a medical provider that refuses to embrace A.I.

It may be due to resignation, and it may not be optimism exactly, but even A.I.’s critics are coming to terms with the new technology. None of us can afford not to. But we could all stand to learn a lesson from Jungian cognitive psychology, which teaches that the longer we stare in a mirror, the more our image can become distorted into monstrous shapes. We will all be staring into an A.I. mirror a lot, and just as Mario and Luigi are aware of Wario and Waluigi, we need to know what we’re looking at.