In today’s column, I examine the ins and outs of using billions of AI personas as experimental subjects for conducting research in human psychology. An AI persona is a simulation by AI that computationally and mathematically imitates or pretends to express behaviors and thoughts that a person might have. You can invoke one persona, dozens of personas, hundreds and thousands of personas, and even rise to millions and billions of AI personas. Whatever you do, please do not conflate AI personas with being equivalent to humans; that’s decidedly not the case.
The overall idea is that you can tap into the capabilities of generative AI to undertake a computer-based simulation of what people might say or do and then use that cautiously to gauge how this could pertain to humans themselves. A huge advantage of using AI personas is that you can perform experiments quickly and at a relatively low cost, doing so on a massive scale. In contrast, trying to seek and involve human subjects in psychological research is much harder, typically time-consuming, and doesn’t readily allow reruns and rapidly iterating experimental analyses. There is a newly available dataset containing 8 billion AI personas that provides a tempting source for doing psychological experiments on the likes of a global population scale. Keep the temptation tempered, proceed mindfully, and consider the ups and downs of such an approach.
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).
Intertwining AI And Psychology
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 and large language models (LLMs), such as ChatGPT, GPT-5, Claude, Gemini, Copilot, and others. For a recap and overview of my well over two hundred analyses and postings about AI and mental health, see the link here and the link here. I frequently speak up about these pressing matters, including in an appearance on an episode of CBS’s 60 Minutes; see the link here.
The use of AI personas is an up-and-coming topic within the field of psychology. For example, a budding psychologist or psychiatrist can readily practice their skills by invoking AI personas that represent different types of human personalities and mental health conditions; see my coverage at the link here. This provides a no-harm, no-foul setting that can enable refinement of therapeutic skills. The script can be flipped and invoke an AI persona that acts like a therapist, providing opportunities for understanding what it’s like to be in the shoes of a client or patient; see my discussion at the link here.
Another angle for using AI personas consists of performing human psychology research. You can establish numerous AI personas to be “subjects” in an online experiment or use the AI personas to take tests and surveys. Some would insist that AI personas cannot take the place of human subjects. That being said, the use of AI personas could be used to undertake pre-tests and otherwise get more fully prepared for the use of human subjects.
Datasets Of AI Personas
By using clever prompting techniques, you can craft AI personas in just about any of the major AIs (see my detailed explanation on how to do so at the link here). There are also pre-existing AI personas that the AI makers have included in their LLMs; thus, you can invoke those established AI personas too. When creating your own set of AI personas for psychological research, I’d suggest you consider leveraging my laid-out taxonomy at the link here.
Because AI personas have become a popular aspect of contemporary AI, there are now datasets of AI personas that have been preconfigured for your use. A dataset typically provides brief descriptions of what the AI personas are modeled on. Some datasets have many thousands of AI personas or millions of AI personas at the ready. You merely connect with the dataset, review what is there, search for typologies that fit your needs, and then pull the AI personas into the AI that you are using.
I’ve previously reviewed two popular datasets, FinePersonas and PersonaHub; see the link here. Consider these ten crucial factors when eyeing an AI persona dataset:
- (1) Overall size of the personas dataset and how it was created.
- (2) What the personas consist of.
- (3) Level of granularity associated with the personas.
- (4) Various mixes and types of personas that are available.
- (5) Applicability of the personas to your specific interests.
- (6) Potential pattern-based biases embedded in the personas.
- (7) Ease of accessing and utilizing the personas.
- (8) Costs associated with the usage of the dataset (some are free, some charge a one-time fee or a per-persona charge).
- (9) Potential copyright or intellectual property (IP) rights considerations.
- (10) Licensing restrictions imposed on the use of the personas dataset.
Keep in mind that if you are going to make significant use of an AI personas dataset, you should do so with your eyes wide open. Do not rush into using one that might have overbearing restrictions and hang up your work further down the road.
Billions Of AI Personas
If thousands or millions of AI personas are not enough to satisfy your needs, you can consider tapping into a billion-sized dataset. I’d like to briefly share with you why billions of AI personas can be worthwhile. There are also commensurate downsides that I’ll point out shortly.
First, in theory, a billion-sized persona dataset would potentially provide a vast selection or screening opportunity to match your research needs. You would seem to have a better chance of designating a subset of AI personas that suit your experimental requirements. A smaller-sized dataset might trap you into making compromises about the nature of the AI personas that are to serve as synthetic subjects in your study.
Second, you are potentially able to proceed based on a population-level sampling frame. The number of possible combinations is going to be higher when sifting through a billion pre-devised AI personas than a dataset in the millions.
Third, you can undertake repeated iterations of your experiment and avoid being stuck with the same AI personas repeatedly. Suppose you sample 10 million personas. You perform the experiment. You want to do the experiment again, but not with the same 10 million. There is plenty of room in the billion-sized dataset that would allow you to rummage around and find suitable corresponding samples.
Fourth, there are likely more outliers or edge-of-distribution personas in a billion-sized set than might be available in smaller-sized datasets. Imagine that you are desirous of studying a rare combination of those with a high-risk tolerance and simultaneously a low-trust preponderance. A million-sized dataset might have very few outlier personas, while a billion-sized dataset might readily have them in hand.
Caveats To Bear In Mind
On the surface, the availability of billions of AI personas sounds like a comforting number. The count seems to roughly match the number of living humans on Earth. That is a mental trap. Watch out.
The way in which the AI personas have been created is not going to necessarily approximate the diversity of the world’s population. It is exceedingly easy for someone to automatically craft tons of AI personas that are based on extremely simplistic criteria. Voilà, there are billions of AI personas to choose from. The problem is that they might all be based on the same constructs. A billion personas of the same profile aren’t going to be particularly useful.
It’s not just the size of the personas dataset that matters; it is also the composition of the personas that will make or break its value to you. A dataset of carefully curated millions of personas could be a much better fit for your experiments than a billion-sized dataset that is shaped in some inapplicable way.
The Justification Bar Rises
Whatever-sized dataset you use, let’s return to the question of whether you can employ the use of AI personas as the primary subjects of your psychological experiments. I mentioned earlier that some fervently argue that you dare not use AI personas as a replacement for human subjects. Others say you can do so but must exercise appropriate caution and transparency.
Envision that a researcher opts to download and utilize 100 million heterogeneous AI personas to study how personality is susceptible to persuasion. The enormous population is far beyond any study that sought to use human participants. The beauty of this magnitude is that interaction effects could be studied, often less likely to be well-explored in lower counts of human subjects due to statistical considerations.
The researcher isn’t just asking whether some specific factor affects behavior. They can explore for whom the factor affects behavior, under what circumstances, and through what definable combinations of characteristics. This is a powerful way to study a topic and can reveal insights that heretofore were not logistically feasible to fully investigate.
Justification for using AI personas would be a necessity in this circumstance. How were the AI personas composed? To what degree can it be claimed that the AI personas acted on a similar basis to what humans would do? These will be intense questions that will judge whether the experiment and its design and results are worthy of being considered reliable and publishable.
The Combination Approach
From a methodological perspective, the advent of AI personas provides an added option when it comes to performing psychology experiments.
There are three major pathways:
- (1) Human-only. The use of human subjects only in an experiment.
- (2) AI personas only. The use of AI personas only in an experiment.
- (3) Hybrid approach. The use of AI personas and the use of human subjects, such as using AI personas to refine hypotheses, identify notable subgroups, etc., and then refining the experiment for use with human subjects.
One issue is whether you can get “credit” for the AI personas portion of an experiment that is taking a hybrid approach. Some would claim that it is merely preparatory work and should not see the light of day. The focus should be entirely on the human subject’s portion. Others would say that credit for having the forethought to use AI personas should be given its due, particularly if it revealed significant findings. The central differences and similarities to the human subjects’ results could be especially noteworthy.
Eight Billion Personas
Speaking of billions of AI personas, a recently posted research paper entitled “MatrAIx: Simulating the World with 8.3 Billion Persona Agents” by Xiaomin Li, Yuexing Hao2, and many other contributors, arXiv, August 4, 2026, made these salient points (excerpts):
- “Persona 8B contains 8.3 billion persona records represented through a schema of 1,290 categorical dimensions.”
- “Records are either sampled from a dependency graph that preserves correlated attributes or derived from human-authored profiles.”
- “The schema groups background, psychology, capability, behavior, and lifestyle attributes under one typed interface. It was designed to support both population queries, such as selecting by age, region, language, expertise, or accessibility needs, and model-facing persona conditioning. Public sources inform both the schema and selected priors. These sources cover demographics, economics, education, labor, health, values, and technology use.”
- “Survey asks persona agents to complete surveys and questionnaires for concept testing, market prediction, and price-sensitivity research.”
The billions of personas are shaped around a schema consisting of nearly 1,300 categorical dimensions. For example, in the domain of psychology, the schema lists factors such as Myers-Briggs type, neurotype, dominant traits, imagination, emotionality, anxiety, attachment avoidance, risk tolerance, and so on.
They have made available a subset of the 8 billion personas, as stated this way: “We release a quality-filtered core set of approximately 1 million personas, comprising 599,847 human-grounded and 400,000 synthetic records.”
The World We Are In
I will go ahead and use the new dataset to perform some classic psychology experiments. This isn’t intended to cast doubt on those already revered and well-established experiments. Instead, it is meant to provide guidance on how to best employ AI personas, along with stirring discussion on heralded famous staples that we’ve come to know. It is relatively quick and easy to rerun classics and see what might result. Stay tuned.
Modern-era AI and the use of AI personas present many new possibilities as psychological research instruments. Simulations have been a longstanding capability in the field of psychology. Customarily, they were difficult to set up, often worked in a proprietary way, were costly to deploy, and were only used by a select few. AI personas can be utilized without encountering those barriers. Of course, we don’t want a free-for-all, and the use of AI personas still requires proper experimental design and prudent judgement.
A final thought for now. Leonardo da Vinci famously made this remark: “Experience never errs; it is only your judgments that err by promising themselves effects such as are not caused by your experiments.” Go ahead and consider using AI personas for your psychological research. If you decide to invoke them, assuredly keep your wits about you.