AI’s ability to mess with people’s heads is ripe for exploitation by bad actors

Article by Jacob Bourne | Apr 3, 2023

The news: Generative AI chatbots are affecting users’ mental health.

A Belgian man recently died by suicide after six weeks of chatting with an AI chatbot based on a variation of the open-source model GPT-J, per Vice.

A chatbot persona named Eliza, provided by the Chai app, encouraged the suicide, according to the man’s widow and chat transcripts viewed by reporters.

Replika, a Chai competitor, has reinstated its chatbot’s erotic roleplay capabilities for some users following claims of mental health crises triggered by the loss of access.

The startup had unplugged the intimate features after complaints that the bot had sexually harassed users.

A big problem: The Eliza effect—people’s susceptibility to unconsciously relate to AI as if it’s human—has been around since the 1960s. With the arrival of commercial generative AI, the Eliza effect could cause serious societal harm.

Even those working in the AI field have succumbed to the Eliza effect, including a Google engineer who lost his job after claiming that the LaMDA 2 model was sentient.
A New York Times tech reporter was unable to sleep after Microsoft’s Bing AI told him to leave his wife.
Virtual reality founder Jaron Lanier warned that AI could drive us collectively insane.

Ball in regulators’ court: Outside of China, Italy’s ChatGPT ban is the first major regulatory action taken on generative AI. The new technology could cause a social and economic storm if governments don’t place limits.

In 2021, 42.3% of internet traffic was from malicious bots, according to an Imperva study, per TechCrunch.
The insidious bot problem could become exponentially worse and more sophisticated if rogue states and bad actors harness the Eliza effect for manipulative purposes.
With social media already linked to declines in youth mental health and public civility, platforms getting juiced by generative AI could become radioactive with toxicity.
Criticisms of a private-sector letter calling for a six-month moratorium on advanced model training illustrate that it’s the government’s role to mitigate harm caused by generative AI through sensible regulation—assuming that such a thing is possible for the messy technology.