Agentic World Models Are Bringing Embodiment And Psychological Grounding When It Comes To Improving AI Mental Health Advice
In today’s column, I examine the use of agentic world models to improve the mental health advice generated by generative AI and large language models (LLMs).
I’ve previously discussed the advent of agentic world models at the link here. The twist this time entails focusing on two novel factors that come into play for extending AI-generated mental health guidance. First, in a broad sense, an agentic world model provides an opportunity for AI to gain a semblance of embodiment. Second, the AI can react within the agentic world and garner important psychological grounding that otherwise would not be feasible without the world model encapsulation.
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).
AI And Mental Health
As a quick background, I’ve been extensively covering and analyzing a myriad of facets regarding the advent of 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.
Background On AI For Mental Health
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.
Improving AI In Mental Health Guidance
A crucial goal of researchers and developers is to improve AI toward generating highly suitable and realistically valid mental health advice. A novel approach to pursue this goal consists of leveraging agentic world models. See my coverage at the link here. Let’s briefly cover what agentic world models consist of. Once that’s been introduced, we will be ready to dive into how they are an innovative tool for advancing the realm of AI and mental health.
The usual approach to data training for generative AI and LLMs consists of using a vast corpus of data, such as text found across the Internet, and having the AI pattern-match on that encountered data. An additional and quickly emerging augmented approach involves establishing a so-called world model that interacts with the budding LLM. During those interactions, the LLM further refines its pattern-matching and logic-based reasoning capabilities accordingly.
Consider an example. Suppose you wanted generative AI to become expertly versed in baseball. The odds are that during the initial data training, the LLM encountered lots of online text that generally described the sport of baseball. Based on that text, it is possible to discuss baseball in general with just about any generative AI and get a reasonably sensible response.
If you want to go further and have the generative AI be deeply responsive about baseball, you might connect the AI to an online video game via an API to involve playing baseball in a simulated environment. I’m sure you’ve seen or used such online video games. You manipulate online controls to swing a virtual bat, run the bases, catch a pop-up flyball, and otherwise are immersed in a simulated baseball game.
What if we had the AI do the same?
The AI would derive additional training about playing baseball. By doing so, the AI is augmenting the “book learning” associated with the text about baseball that was scanned and patterned during initial training. This gives the AI a type of “experience” related to playing baseball.
AI And Psychology
One of the ongoing debates in the field of psychology and AI is that AI cannot “experience” the real world and therefore cannot grasp what it means to undergo psychological development and adjustment, see my detailed explanation at the link here.
This is the now-classic embodiment problem. Some ardently believe that AI will never reach human-levels of cognition because AI doesn’t have a body. Humans have a body. They learn about the real world via the use of that body. Since AI lacks a body, it can never fully be cognitively on par with humans. Period, end of story.
Not everyone agrees with that assertion.
There are various counterarguments afoot. For example, generative AI is gradually being incorporated into humanoid robots. These robots are made to look like humans, containing mechanical arms, hands, fingers, feet, toes, and the like. The latest advances include artificial electronic skin that reacts to touching and sends digital signals to the AI that is running the humanoid robot. The gist is that perhaps this will finally get us past the debate about AI not having “experiences” in line with human experiences. For more on the latest in AI and humanoid robots, see my discussion at the link here.
Since humanoid robots are still being crafted and don’t yet fully exist per se, another alternative to the embodiment problem is to encapsulate AI within an artificial environment. We can use agentic world models to do this. The world model simulates a real-world setting. The AI must make its way through this simulated real world.
Learning About Psychology
Think about the human mind and how humans develop psychologically. We interact in the real world. The behavior of others shapes our ways of thinking. You learn about frustration, adaptation, and expectation formulation via how the world pushes back at you. People exist within an environment that resists their intentions. The real world imposes constraints on you.
By embedding AI within an agentic world model, the aim is to force AI to interact in a simulated world and learn about frustration, adaptation, and expectation formulation. The model pushes back at the AI. The AI deals with an environment that resists the actions of the AI. An agentic world model imposes constraints on the AI.
The agentic world model becomes a controlled mirror to psychological “experiences” for AI. That being said, there is a risk that this is an overstating of what takes place, namely that we are anthropomorphizing AI. To suggest that the AI will experience human suffering via such a simulation is a bridge too far for some critics.
New Approach To World Models
To ease the effort to make use of agentic world models overall, a new approach known as web world models (WWM) has been devised. In a research paper “Web World Models” by Jichen Feng, Yifan Zhang, Chenggong Zhang, Yifu Lu, Shilong Liu, Mengdi Wang, arXiv, December 29, 2025, these salient points were made (excerpts):
– “Creating environments that support long-term agent interaction requires robust memory and social simulation capabilities.”
– “A Web World Model is a world whose state and ‘physics’ are defined by ordinary web code (e.g., TypeScript modules, HTTP handlers, and database schemas), while large language models generate context and narratives on top of this structured latent state.”
– “In this view, code specifies what kinds of entities exist, how they interact, and which actions are possible; the model is invoked to enrich these entities with descriptions, stories, or task-specific reasoning.”
– “WWMs thus inherit the controllability, observability, and tooling of web frameworks, yet they can procedurally expand to an effectively unlimited state space by using language models to synthesize new content on demand.”
The distinctive value-add is that a web world model is generally easier to construct and utilize than other, more arcane world models.
Applying The Model To AI And Psychology
In my lab, we are experimenting with having AI embedded within a free-ranging world model and discerning if a focus on psychological learning can be undertaken. If you think of psychological development as the gradual shaping of internal models through repeated interaction with a stable environment, then world models are surprisingly well-suited for this purpose.
Some distinct advantages include:
– Persistence: Various entities, relationships, and consequences endure across time.
– Observability: Internal states can be logged, inspected, and replayed.
– Counterfactual control: You can rewind, branch, or modify developmental conditions.
– Narrative scaffolding: AI can translate raw state transitions into self-models, explanations, and meaning-making.
This especially matters for mental health because many psychological constructs, such as identity, trust, attachment, avoidance, and coping, are fundamentally relational and temporal. A purely stateless chatbot cannot meaningfully exhibit them. AI embedded in a world model, however, can develop patterns that look like psychological traits. For example, traits including learned helplessness under certain reward structures, risk aversion after simulated loss, and attachment-like dynamics to persistent agents.
Our overarching goal in this pursuit is this:
– Experimental goal: To examine whether an AI embedded in an agentic world model exhibits stable, history-dependent behavioral patterns that resemble psychological constructs such as avoidance, self-modeling, and maladaptive risk aversion.
Initial Results Are Promising
The work is showing promising results so far, but it is only at an early stage. I am hopeful that other researchers will take a similar approach, and, if so, I will aim to cover such research as it becomes available. Stay tuned.
Initial findings encompass:
– Persistent world rules create expectations.
– Repeated negative outcomes produce avoidance.
– Narrative generation converts state transitions into self-belief.
– World model internal state changes, rather than internal rewiring, alter behavior.
On an AI and mental health guidance basis, this research suggests that maladaptive patterns can emerge from history, constraints, and meaning-making, even in a simulated world. That makes world models a promising sandbox for studying which therapeutic levers operate at the level of environment, narrative, or agency.
The World We Are In
Let’s end with a big picture viewpoint.
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.
We need to decide whether we need new laws or can employ existing laws, or both, and stem the potential tide of adversely impacting society-wide mental health. 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. You might remember that in The Wizard of Oz, the Scarecrow sings a song about the lament and desire if he only had a brain. Perhaps AI will one day sing a song about desiring a body, though it could be that humanoid robots are sufficient and the request will be otherwise fulfilled. Time will tell.
