I am a person who rarely uses AI, but I cannot say I have not tried it. I remember when ChatGPT first came out, and almost every one of my classmates was using it as a tool for their studies. I decided to give it a try and inserted an essay into the machine and prompted it to make it sound better. What I failed to realize is that this machine takes your tone and style of writing and converts it to its own formulation of writing. I did not know you have to be specific about how to prompt ChatGPT. Then I learned that generative AI feeds off the information you insert and keeps it. Never did it again. After reading Kalota’s article, I have gained a deeper understanding of how artificial intelligence works. It is a complex system, but still in need of continued work. If a colleague asked why AI gets something wrong, I would tell them that AI is a machine that collects data and sorts it into categories. These categories become patterns that the machine identifies and outputs. It only knows information to a certain extent, resulting in inaccurate information. Never depend only on AI. When I think about AI in a social work practice, I am still uncomfortable with it. The main reasons are the environmental factors and confidentiality. I question myself on how I can, as a social worker, advocate for a community facing data center issues when I am using AI. How can I serve as a social worker when I am risking private information by using AI?

My complicated perspective on AI makes it difficult to accept AI in the social work field, but I know it is inevitable. An AI tool could be useful if it adheres to the NASW. As I mentioned before, AI is a complex system that needs continued work, so if AI engineers could focus on having the machine recognize social work ethics and keep the information confidential, I beleive it can be useful with social workers’ workload. In a perfect world, AI can help lessen the social worker’s data collection workload, like the administrative and client intake screening part of the role. This would then allow social workers to focus more on the human interaction aspect of the role. Therefore, from my perspective, AI is better for data collection. Moore et al. (2025) demonstrated how AI, specifically LLMs, lacks a deeper understanding of the appropriate response to delicate situations like suicidal ideation. LLMS showed stigma and made dangerous statements, risking the client’s safety (Moore et al., 2025). It helped me understand that AI lacks the capacity for therapeutic intervention. My understanding is that AI could be better at information input, such as data collection, but not at acting as a therapist. In my experience, I have seen how human oversight can cause a ripple effect in an organization. Working in a substance abuse organization has shown me that not only does inaccurate data impact a client’s process, but also the organization could fail to represent the numbers for their financial documents. It leads to miscommunication and inaccurate data.

Kalota, F. (2024). A Primer on Generative Artificial Intelligence. Education Sciences, 14(2), 172. https://doi.org/10.3390/educsci14020172

Moore, J., Grabb, D., Agnew, W., Klyman, K., Chancellor, S., Ong, D. C., & Haber, N. (2025). Expressing stigma and inappropriate responses prevent LLMs from safely replacing mental health providers. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (FAccT). https://doi.org/10.1145/3715275.3732039