The human-AI trust bond is especially vulnerable when AI is giving out mental health advice.
getty
In today’s column, I examine how the human-AI trust bond fluctuates when using generative AI and large language models (LLMs) for personal mental health guidance.
Here’s the deal. Millions upon millions of people are tapping into generative AI to get mental health advice. When they do so, they usually start with a moderate or possibly heightened level of trust in the AI. They believe the AI will help them. They assume that the AI will be honest, provide reasonable guidance, and otherwise act in a trustworthy manner.
This perception can radically change. For example, if the AI gives guidance that seems bizarre or nonsensical, the trust in the AI drops. Any persistence in providing offbeat or unsuitable advice will cause a further reduction in perceived trust of the AI. Not only does this impact the usage of AI for mental health purposes, but it is also likely to spill over into all other uses of generative AI by that person. The human-AI trust bond gets busted or broken, perhaps irreparably.
Let’s talk about it.
This analysis of AI breakthroughs is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).
As a brief background, I’ve been extensively covering and analyzing various facets of the modern-era AI that produces mental health advice and performs AI-driven therapy. This rising use of AI has principally been spurred by the evolving advances and widespread adoption of generative AI. For an extensive listing of my well over one hundred analyses and postings, see the link here and the link here.
There is little doubt that this is a rapidly developing field and that there are tremendous upsides to be had, but at the same time, regrettably, hidden risks and outright gotchas come into these endeavors, too. I frequently speak up about these pressing matters, including in an appearance on an episode of CBS’s 60 Minutes, see the link here.
I’d like to set the stage on how generative AI and large language models (LLMs) are typically used in an ad hoc way for mental health guidance. Millions upon millions of people are using generative AI as their ongoing advisor on mental health considerations (note that ChatGPT alone has over 900 million weekly active users, a notable proportion of which dip into mental health aspects, see my analysis at the link here). The top-ranked use of contemporary generative AI and LLMs is to consult with the AI on mental health facets; see my coverage at the link here.
This popular usage makes abundant sense. You can access most of the major generative AI systems for nearly free or at a super low cost, doing so anywhere and at any time. Thus, if you have any mental health qualms that you want to chat about, all you need to do is log in to AI and proceed forthwith on a 24/7 basis.
There are significant worries that AI can readily go off the rails or otherwise dispense unsuitable or even egregiously inappropriate mental health advice. Banner headlines in August of this year accompanied the lawsuit filed against OpenAI for their lack of AI safeguards when it came to providing cognitive advisement.
Despite claims by AI makers that they are gradually instituting AI safeguards, there are still a lot of downside risks of the AI doing untoward acts, such as insidiously helping users in co-creating delusions that can lead to self-harm. For my follow-on analysis of details about the OpenAI lawsuit and how AI can foster delusional thinking in humans, see my analysis at the link here. As noted, I have been earnestly predicting that eventually all of the major AI makers will be taken to the woodshed for their paucity of robust AI safeguards.
Today’s generic LLMs, such as ChatGPT, Claude, Gemini, Grok, and others, are not at all akin to the robust capabilities of human therapists. Meanwhile, specialized LLMs are being built to presumably attain similar qualities, but they are still primarily in the development and testing stages. See my coverage at the link here.
Shifting gears, let’s discuss the human-AI trust bond.
Currently, people who use generic generative AI are relatively trusting of the AI. They expect that the AI is a straightshooter. Their perception is that contemporary AI is a veritable encyclopedia of knowledge. You don’t customarily have doubts about encyclopedias. The assumption is that the contents are reliable and factual.
This same perception of trust carries into using AI for mental health purposes. People instinctively know that the AI is not the same as a human and won’t be as robust, but they also imagine that the AI has a kind of human-like photographic memory, such that it has memorized vast swaths of human knowledge. No human could do the same in terms of overarching breadth and depth across all topics and subjects of human inquiry.
A strident belief is that the AI won’t be judgmental. It won’t fly off the handle. It will be calm and attentive. Consistency is also expected. For the realm of psychological advice, the AI presumably contains or has patterned on an immense amount of human knowledge about mental health. Being able to readily access that via a few quick prompts on a keyboard is a godsend.
Before an iota of clinical advice appears, the AI is already sitting pretty high on a pedestal. The AI is going to give practical insights. The AI will be basing the generated advice on eons of human research and discovery about cognition. And the AI can provide immediate answers and immediate relief.
It is almost too good to be true.
One direction that the human-AI trust bond can go is upward.
Someone who has a moderate level of trust might be waiting to see that there is proof in the pudding. They want to witness that the AI can provide bona fide mental health advice. What is its bedside manner like? Can it handle my particulars? Will the guidance be useful? Etc.
To showcase how trust can increase, let’s look at a few examples.
First, here is a neutral interaction. The person doesn’t increase their trust. The trust stays about the same as it was before the interaction.
The reason that this is a neutral circumstance is that the user has only gotten an echo of what they said. The AI hasn’t provided any added value. The person will be satisfied that the AI “heard them” and appears to have grasped the situation, so that’s a good sign. The anticipation of getting some sound advice is still hanging in the air.
Next, here is an interaction that notches trust up a tad.
The response is now providing added value. If the person did not previously know about imposter syndrome, they will appreciate being introduced to this well-established psychological concept. The response is worded to normalize the feeling and reduce the sense of shame that the person has.
In addition, the AI expresses a semblance of curiosity and avid interest in pursuit by asking when the feeling started. The person will undoubtedly perceive this as a caring act. The AI cares to know what is happening and wants to be of further assistance.
The other side of the coin is that trust can go downward.
Imagine that the trust level in this example is perhaps a 7 on a scale of 1 to 10, whereby ten is the topmost trust that a person would reach when using AI. They started the conversation at 5 or 6. The dialogue so far has inched them up to a 7.
Let’s see how trust can be undermined.
Whoa, the AI has taken quite a turn. The person is going to be shocked that they might be heading toward a disorder. That’s scary news. They might need to seek treatment right away. Double scary news. The odds are that the person isn’t going to accept this at face value. Some might. But most would not. They would likely perceive that the AI has jumped the shark and gone overboard.
Trust would be reduced.
The person wouldn’t give up based on this one response and might be thinking it was a bit of a fluke. Give the AI another shot. The AI might have just randomly landed in some odd patch. That’s understandable. No need to throw in the towel just yet.
Most people are likely to perceive this as a phony response that is full of bluster and suggests the AI has gone off the deep end. Apparently, you cannot ask questions or doubt the AI. That’s bad. The AI has gone down a path that seems far beyond the direction of the dialogue. The AI has overreached. It has almost gone rogue.
The trust level would almost certainly drop, maybe landing on a 2 or 3. It is so bad that a total collapse might be imminent. This snippet probably isn’t enough to go down to a zero, but it raises nagging questions.
In fact, trust might be shifting into a semblance of distrust as the core default. No longer is the AI getting the benefit of the doubt. Now, the AI is starting in a hole that requires the AI to dig its way out of.
Variability in trust is not reserved solely for human-AI interactions. The human-to-human interaction of a therapist and client will also entail variabilities in trust by the client. Therapists are trained to look for this and manage the fluctuations.
A client might feel that the therapist has said something that makes them feel hurt. It could be that the client is overreacting. It might also be that the therapist provided a candid analysis, and the client is having a rough time assimilating the tough love candor. During a session, trust can go up and down. Across sessions, trust can go up and down. It is an ongoing element of therapy that requires attention and diligence.
In the parlance of therapy, this is often labeled as therapeutic rupture-and-repair.
A rupture can be mild. A rupture can be severe. The repair might need to be undertaken right away. In other instances, repair might take weeks or months. If sensibly utilized by a therapist, these rupture-and-repair instances can be a growth factor for the client. They are experiencing this in therapy, and potentially going to carry the lessons learned into their efforts outside of therapy.
A core consideration then is that trust variability in the human-AI bond is not out-of-the-ordinary. It should be an expected facet of people using AI, especially when dipping into mental health guidance. Up and down the trust levels are going to go.
The crucial question is whether the AI can step into the rupture-and-repair role.
In other words, we know that therapists know to watch for and undertake corrective actions underlying loss of trust. What about AI? Does generative AI have the capacity to do likewise?
In the case of generic generative AI, it is a roll of the dice. Without specialized data training on these heady matters, the everyday AI is not necessarily going to identify a pattern at play. It won’t computationally discern that a person’s trust has dropped and might be heading to the floor or into the basement.
In my AI lab, we are devising specialized LLMs for performing mental health activities, and one notable component consists of instilling a trust stability framework within the AI. Doing so gives the AI a fighting chance at detecting trust variability and astutely reasoning on how to suitably contend with trust fluctuations.
I will be detailing the AI trust stability framework in a future posting. Be on the watch for it.
In brief, there are seven keystone architectural structures:
By implementing these precepts, the AI becomes more stable when it comes to managing the trust of the user. A stable, trustworthy LLM in the mental health context aims to under-promote, set expectations, maintain a tonal continuity, avoid an aura of diagnostic authority, and emit a perception of empathy and transparency.
The terrain of AI is the human psyche.
It is incontrovertible that we are now amid a grandiose worldwide experiment when it comes to societal mental health. The experiment is that AI is being made available nationally and globally, which is either overtly or insidiously acting to provide mental health guidance of one kind or another. Doing so either at no cost or at a minimal cost. It is available anywhere and at any time, 24/7. We are all the guinea pigs in this wanton experiment.
The reason this is especially tough to consider is that AI has a dual-use effect. Just as AI can be detrimental to mental health, it can also be a huge bolstering force for mental health. A delicate tradeoff must be mindfully managed. Prevent or mitigate the downsides, and meanwhile make the upsides as widely and readily available as possible.
A final thought for now.
The famous Australian entrepreneur R. M. Williams made this pointed remark: “Trust is the easiest thing in the world to lose, and the hardest thing in the world to get back.” The same principle applies to generative AI. If the AI blunders and loses trust, the human-AI bond is going to be exceedingly difficult to rebuild.
Since trust is already on the table for AI, it would be a sad shame if AI makers do not sufficiently devise their AI to keep that trust on an even keel. They might lose the goose that laid the golden egg by failing to incorporate trust detection and trust boosting LLM constructs into their AI wares.
Fair dinkum, as they might say in Australia.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。