When faced with a dilemma, who wouldn’t want to try their luck with a seemingly infinite answer machine? An interactive machine companion doesn’t sound like the worst idea.
We’re living in a time of “friendship recession,” where Americans today have fewer personal friendships and social support.
Recent data from the American Survey Center shows Americans have fewer close friendships now than they did previously, a number that began dwindling decades before the COVID-19 pandemic began. The survey found that 49% of Americans reported having three or fewer close friends.
We also live in a country without socialized mental healthcare, where the depression rate remains at a historic high.
Against all conventional wisdom and supposed right answers, this outpour of feeling in the absence of close friendship should not be directed toward artificial intelligence chatbots. We need to say no to answers that seem right and show respect to the indomitable human spirit. Like it or not, some things have no substitute.
ChatGPT, released in November 2022 and arguably still the most mainstream household-name large language model, has its own Personal Problem Solver. As defined by IBM, Large Language Models are AI learning models that generate text in response to prompts, extracting from a database and breaking down large swaths of text into tokens, which are machine-learnable units.
On paper, it sounds like a futuristic, ingenious idea: Synthesized data from other personal quandaries across the web, streamlined in a quick text quip, impersonal and objective. A robot oracle, or simply a listening ear. An optimized outsource of human interaction and social intelligence.
However, it’s not a suitable replacement. There is an ultimate human quality within advice. Objectivity isn’t everything, nor is it always useful or necessary.
These AI chatbots are no confidantes, and they are certainly no therapists. Feelings are complicated, and there isn’t any formula or universal approach to most situations.
Knowing which details and facts to parse out and which to hone is of utmost importance, and this comes from more than mere language, written or spoken. Body language, tone of voice and gestures all tell a story outside of the narrative being recalled, as does personal context and knowledge of the seeker.
University of Minnesota second-year psychology Ph.D. student Caroline Armstrong said upon ChatGPT’s initial release, she input a friend’s dilemma just to see if it would line up with a real human response, to no avail.
“I feel like it gave generic advice that was not necessarily more thoughtful than what I would be able to give her,” Armstrong said.
It’s not that it is going against its programming, or what it isn’t supposed to do. There’s a mismatch between consumer expectations and the capabilities of this technology.
University first-year computer engineering student Ian Gantner said consulting ChatGPT for personal problems wouldn’t be of any help to him, so he’s never felt the urge.
“I feel like it gives a net average response to like, if everyone in the world had that problem, you know, what’s the average thing that would work?” Gantner said. “So I don’t think it would work for me, because I don’t think the average works for most people.”
There isn’t a one-size-fits-all approach to any problem, outside of conventional knowledge. If someone requires consultation on a personal matter to such a degree that they would consult a machine learning platform, chances are their quandary exceeds the limits of conventional wisdom.
People are smart enough to know when they need help. What they need help with are not the facts of a situation but the in-betweens that can only be felt and perceived by a human witness or confidante.
University computer science and engineering Ph.D. candidate Ash Milton said these chatbots cannot think, making them no substitute for human interaction.
“It doesn’t have feelings, it can’t have empathy, it can’t have sympathy,” Milton said. “It can fake it real well, but it doesn’t actually have it.”
This tech is designed to be user-friendly. People are not.
University first-year student William Marshall said he feels ChatGPT has ulterior motives and would prefer to talk to a peer about any personal dilemma that arises.
“I don’t think it can actually figure out your situation,” Marshall said. “It’s just saying whatever it thinks will appease you and get you to keep using it.”
Milton said validation should be handled delicately and at the discretion of a trained therapist.
“A lot of the systems right now are very prompted to be validating,” Milton said. “But validation isn’t always good in specific cases, and a therapist knows that they know when validation is needed.”
Even with all of the correct programming and prompts for therapeutic uses, it doesn’t stack up to the real thing, Milton added.
“You can feed all the information to pass the certifications for therapy,” Milton said. “It might be able to do it, but at the end of the day, it’s not a person. It can’t think, it can’t have emotions.”
According to the U.S. Centers for Disease Control and Prevention data, one in four U.S. adults says they don’t have social or emotional support.
It seems we’re being backed into a corner where this tech is the only vestige of help. Something to talk to might seem better in the short term when someone isn’t an option.
“You can get out your venting at something that’s a little bit more interactive than journaling is, and people can find that to be helpful,” Milton said. “There’s also the possible side of then kind of having this regression and this atrophy of being able to do your own emotional regulation.”
This technology should not be lauded for the purpose of social or interpersonal problem-solving or mental health support. Just because in some abstract sense it could help doesn’t mean it should.
I don’t want to live in an automated, algorithmic dystopia where people can’t turn to each other or professionals when support is needed.
Feelings can’t be outsourced.