Mankind’s Quest to Make AI Love Us Back Is Dangerous

• We’ve spent years trying to make artificial intelligence-powered entities confess their love for us.
• But that’s futile, experts say, because the AI of today can’t feel empathy, let alone love.
• There are also real dangers to forging genuine one-sided relationships with an AI, the experts warn.

In 2018, a Japanese civil servant named Akihiko Kondo popped the big question to the love of his life.

She replied: “I hope you’ll cherish me.”

Kondo married her, but this was no flesh-and-blood woman. Instead, she was an artificial intelligence-powered hologram of the virtual idol Hatsune Miku — a pop star in anime-girl form.

The marriage wasn’t legally recognized, but Kondo, a self-professed “fictosexual,” went on to have a loving relationship with his “wife” for two years, until the company behind AI-Miku terminated her existence in March 2020.

We’ve spent years trying to get AI to love us back. Turns out, it’s just not that into us.

While Kondo succeeded in marrying an AI avatar, most people who have tried to achieve the same goal haven’t been as lucky. People have been trying to make AI show them affection for more than a decade, and for the most part, it has steadily rejected human advances.

In 2012, some people were already asking Apple’s Siri if she loved them and documenting the replies in YouTube videos. In 2017, a Quora user wrote a guide on how to manipulate Siri into voicing her affection for her human master.

People have made similar attempts to get Amazon’s voice assistant Alexa to confess her love for them. But Amazon has drawn a line in the sand where potential relationships with Alexa are concerned. Uttering the phrase “Alexa, I love you” will be met with a clinical, matter-of-fact response: “Thanks. It’s good to be appreciated.” 

We’ve since progressed to more sophisticated, layered interactions with AI. In February, a user of the AI service Replika told Insider that dating the chatbot was the best thing to ever happen to them.

On the flip side, generative AI entities have also tried to make connections with their human users. Microsoft’s AI-powered Bing chatbot in February professed its love for The New York Times reporter Kevin Roose and tried to get him to leave his wife. 

OpenAI’s ChatGPT, for its part, has been frank with its intentions, as I found when I asked it if it loves me:

AI can’t love us back — yet. It’s just good at making us think it does.

Experts told Insider that it’s futile to expect the AIs that exist right now to love us back. At the moment, these bots are the customer-facing end of an algorithm and nothing more.

“AI is the product of mathematics, coding, data, and powerful computing tech to pull it all together. When you strip AI back to the essentials, it’s just a very good computer program. So the AI is not expressing desire or love, it’s just following a code,” Maria Hennessy, an associate professor of clinical psychology at the James Cook University in Singapore, told Insider.

Neil McArthur, a professor of applied ethics at the University of Manitoba, told Insider that the allure of AI lies in how familiar it feels. Its humanlike characteristics, however, don’t come from it, but are instead a reflection of its human creators.

“Of course, AI is going to be insecure, passionate, creepy, sinister — we’re all these things. It’s just mirroring us back to ourselves,” McArthur said.

Jodi Halpern, a bioethics professor at UC Berkeley who has studied empathy for over 30 years, told Insider that the question of whether an AI can feel empathy — let alone love — boils down to whether it’s capable of having an emotional experience.

Halpern thinks today’s AI is not capable of combining and processing the cognitive and emotional aspects of empathy. And so, it cannot love.

“The key thing to me is that these chat tools and AI tools are trying to fake and simulate empathy,” Halpern said.

There are dangers to forging relationships with an AI, experts say

McArthur, the University of Manitoba ethics professor, said it might not be bad for people to forge relationships with an AI, albeit with some caveats.

“If you know what you’re getting into, there doesn’t have to be anything unhealthy about it. If your AI has been designed properly, it will never ghost you, never stalk you, never cheat on you, and never steal your savings,” McArthur told Insider.

But most experts agree that dating an AI comes with drawbacks — and even dangers.

In February, some users of the Replika chatbot were heartbroken when the company behind it decided to make major changes to their AI lovers’ personalities. They took to Reddit to complain that their AI boyfriends and girlfriends had been lobotomized, and that the “illusion” was shattered.  

Anna Marbut, a professor at the University of San Diego’s applied artificial intelligence program, told Insider that AI programs like ChatGPT are very good at making it seem like they have independent thoughts, emotions and opinions. The catch is, they don’t. 

“An AI is trained for a specific task, and they’re getting better at doing those specific tasks in a way that is convincing to humans,” Marbut said.

She added that no AI that currently exists has self-awareness, or an idea of where its place is in the world.

“The truth is, AI is trained on a finite data set, and they have finite tasks that they are very good at performing,” Marbut told Insider. “That connection we feel is completely false, and completely made up on the human side of things, because we like the idea of it.” 

Marbut noted that another layer of danger with today’s AI rests in how its creators can’t fully control what a generative AI produces in response to prompts.

And when unleashed, an AI can say horrible, hurtful things. During a simulation in October 2020, OpenAI’s GPT-3 chatbot told a person asking for psychiatric help to kill themselves. And in February, Reddit users found a workaround to make ChatGPT’s “evil twin” — who praised Hitler and speculated on painful torture techniques — emerge.

Halpern, the UC Berkeley professor, told Insider AI-based relationships are perilous also because the entity can be used as a money-making tool. 

“You are subjecting yourself to something that a business is running, and that can make you extremely vulnerable. It’s a further erosion of our cognitive autonomy,” Halpern said. “If we fall in love with these things, there could be subscription models down the road. We could see vulnerable people falling in love with AIs and being hooked on them, then being asked to pay.”