My New Understanding of AI from Kalota’s Primer

Identifying one specific concept that I learned from the Kalota article is hard because I learned a lot. I have always known that the work that goes into creating and maintaining an AI system was difficult for me to understand. I struggled a lot while reading and trying to understand the concepts because they were so technical, but after reading it, I feel like I understand how AI works a bit better. One concept that changed my understanding of AI was learning that ChatGPT is not just “generative AI”. It is also a large language model (LLM) that uses natural language processing (NLP) techniques to understand and generate human language. (Kalota, 2024) Before learning this, I thought of AI as one technology, but I now understand that generative AI is built from several different technologies with different roles working together. It also helped me understand that the system predicts the most likely words to generate next based on patterns it learned rather than searching for facts. (Kalota, 2024) I did not know that this technology was capable of reasoning in a way that resembles human reasoning. I previously thought of it as a data-distributing machine instead of one that analyzes and processes things.

I use AI a lot in my everyday life because I am such a curious person. My browser history on a given day is probably 2 pages long because I am constantly trying to learn and understand the world around me and any questions that come into my mind. I have an issue with the unknown. I am not writing this to analyze myself, but to explain that the everyday research that I have always done has gotten a lot more efficient with generative AI, and I was not aware of that. Simply googling something now gives you an AI summary without you having to even do the research yourself. Knowing what I know now about how the data is processed and how the system is taught helps me understand why there may be discrepancies or why inferences may not be factual.

Now that I have a general understanding of how AI works, I am more confident in using AI as a tool in social work. I am less confident in blindly trusting it because I understand how it can have inaccuracies, but I believe that it can be a useful assistive device for social workers.

If I were asked to explain why AI hallucinates sometimes, I would likely use an analogy. I would describe AI as a person who has read a billion books and memorized countless writing patterns and learned how to explain things confidently. Since that person knows so much, they are generally accurate, but when asked a question they may not have the knowledge base for, they may try to fill in the gaps with knowledge that they have and infer the answer without actually fact-checking it.

Utilizing AI Tools in Mental Health Practice

The study by Moore et al. on chatbots was fascinating. The study helped me understand just how chatbots may be unintentionally interfering with effective mental health treatment (Moore et al., 2025). I understand the general public’s desire to utilize AI, though, especially for mental health because of stigma and lack of access.

I think that AI could be used successfully as an intake screening tool IF there was human oversight. For example, I think it would be important to have a staff member reviewing any “screen-outs” to ensure that they were accurately screened out. I don’t think that someone would necessarily need to review the cases that are screened in or scheduled with a therapist because the therapist could make that decision when they meet with the client. I work in child welfare, and we have a centralized intake team that determines the route of cases. Although AI is not currently used for this, we still have staff assigned to the tasks of double-checking the work of the intake team, and there have been plenty of times that we have had to screen a case that was screened out. I have also seen this system fail because both checks fail to recognize missed details and affect the families involved. This experience has shown me that even trained professionals sometimes make mistakes when screening cases, and because of that, I am even less comfortable allowing AI to make screening decisions without human review.

For me, the line between an appropriate task for AI and a task that requires a human is whether the decision is “high-stakes.”  The higher the potential risk to safety, well-being, or access to services, the more important it becomes for a trained human professional to make the final decision. Moore et al (2025) show how unhelpful AI chatbots can be in acting as a “therapist” because it struggles to read human emotion and social context. The issue that I see is that we are not going to be able to stop people from using these chatbots, so the solution would be giving more options. As the reading suggests, utilizing AI to help make the process of accessing human therapists easier may deter people from relying on these chatbots (Moore et al., 2025). I honestly have a hard time “letting go” of any control that we have over speaking with clients regarding mental health issues, and I cannot see any use for diagnosing symptoms or giving suggestions for treatment/solutions by AI. I think AI can be useful to help understand processes, insurance, and available resources. Those tasks involve providing information rather than clinical decision-making, which is why I feel it is a more appropriate use of AI. Decisions involving diagnosis, treatment recommendations, or crisis intervention should remain the responsibility of trained mental health professionals.

References