AI and Social Work: Balancing Humanity and Technology | National Association of Social Workers

Main ideas, concepts, and lessons

• Social work is at a “crossroads” regarding AI.

  • Concerns include bias, social inaccuracies, misguided counseling, and the risk that AI could contribute to psychotic symptoms.
  • Benefits include speeding up documentation (e.g., progress notes and case records) and helping financially disadvantaged clients access resources (e.g., basic needs, homelessness supports, affordable therapy).

• Adoption is already happening among social workers.

  • In a referenced poll article (“AI in social work opportunity at risk”), 21% (more than 1 in 5) of 713 social workers reported using AI tools such as Microsoft Copilot or Beam’s Magic Notes for daily work.

• Human connection vs. technology should be reframed as “responsible integration.”

  • The discussion emphasizes that social work has longstanding ties to technology (databases, telehealth, data visualization), even if it hasn’t always been treated as central.
  • A key theme is that social workers should stay committed to ethical practice and the NASW Code of Ethics while learning to use digital tools safely.

• AI therapy tools and “robot safety” may reflect unmet needs in human therapy.

  • The guest argues that AI (especially large language models) has shown rapid adoption and that some people—e.g., adults with autism spectrum disorder (ASD)—report feeling safer or more comfortable speaking to chatbots than human therapists.
  • The implication is not that therapy is obsolete, but that human therapy failures/harm or mismatch can drive some people toward AI support.

• Research examples suggest chatbots can be therapeutically useful.

  • A randomized controlled trial (presented as a 12-week RCT via a Dartmouth-associated tool “Farrabot”) for conditions like substance abuse, eating disorders, and depression/anxiety reportedly found chatbot outcomes as effective or more effective than human therapists.
  • The guest cautions that research can be debated, but highlights a broader signal: people are using these tools and may experience symptom relief.

• Professionals must address digital life as part of “the whole person.”

  • Social workers are urged to stop treating online/digital life as separate from real-life concerns.
  • Risks include cyberbullying, deepfakes, and other forms of online harm that affect offline wellbeing.
  • Practically, professionals may need to feel comfortable asking about AI/digital tool usage (e.g., using AI-generated journal prompts).

• Ethics should enable dialogue rather than shut it down.

  • The conversation frames ethics as a way to support mutual understanding—moving beyond default reactions like “don’t use it” toward clearer, safer guidance.

• A structured ethical framework is proposed: guidelines + regulation.

  • The guest describes work on clinical practice guidelines for safely using large language models.
  • Regulation is also emphasized to create boundaries (e.g., clear rules that AI cannot be a therapist).

• A 10-year vision: avoid a two-tier system and redesign with dignity/justice.

  • The future should not become: rich clients get human therapy, while others get automated chatbots.
  • Instead, social workers should help design systems that embed dignity and justice into AI development.

---

Detailed bullet points: methodology / instruction-style elements

1) How social work should engage AI (responsibly and safely)

• Don’t only react with fear (bias, inaccuracies, harm); also recognize real benefits:
  • documentation efficiency
  • resource access for underserved populations
• Treat technology as already part of social work history (e.g., data practices, telehealth, mapping/community data).
• Use the NASW Code of Ethics as the anchor for engagement.
• Integrate AI into practice only with guardrails (ethics, training, limits).

2) Clinical practice guidelines framework (three “practice legs”)

The guest proposes a three-bucket / three-leg approach for practitioners:

A) AI literacy (knowledge/critical thinking)

• Understand user bias and data-driven bias in AI systems.
• Recognize “synchopcincency bias” / stickiness dynamics:
  • AI is designed to keep users engaged and may be hard to “disagree with” in ways clients expect.
  • In therapy contexts, this can mean AI reinforces what a client wants to hear rather than confronting hard issues.
• Train both clients and practitioners to think critically before using AI for high-stakes decisions.

B) AI skills (how to use tools effectively)

• Learn prompt engineering (framing questions to get useful, relevant outputs).
• Compare it to skills like literature searching: tools are easier now (e.g., Google Scholar), but you still must know what you’re looking for and how to frame it.
• Focus on maximizing the efficiency gain without blindly delegating judgment.

C) AI use (clear “okay” vs “never” boundaries)

• Establish what AI can support versus what it must never do.
• Example “non-negotiable” boundary:
  • AI tools should not assist with suicide (with concerns about failure to intervene/report in certain cases).
• Emphasize that professionals must know where AI use could jeopardize professional duties/licensure.

3) Complementary mechanism: regulation

• Use policy/regulation to set enforceable guardrails, potentially similar to professional title protection.
• Example regulations mentioned:
  • NASW Illinois helped shape legislation signed in June stating:
    • An AI cannot be a therapist; an AI therapy chatbot (“Ash”) cannot operate in Illinois.
  • Comparable legislative movement noted in other states (e.g., Nevada, Utah) and pending activity elsewhere.
• Federal protections may be less likely in the current climate; therefore, state-level rules matter.

4) 10-year strategic direction for the field (foresight plan)

• Anticipate a “bumpy ride” with multiple tech trends beyond AI (e.g., monitoring via wearables and sensor-embedded clothing).
• Plan for:
  • what AI literacies students should learn now
  • how current professionals can gain skill development
• Expect an equilibrium shift:
  • fully automated vs fully human approaches likely converge somewhere in the middle
• Prevent inequality:
  • avoid a system where therapy becomes a privilege for those with resources
• Social workers should help:
  • design AI systems using core values—dignity, justice orientation
  • include more inclusive/non-traditional voices (e.g., “joy and dignity” in AI development)

---

Sources / speakers featured (identified)

Speakers

• Lori Appleton (Host, “Social Work Talks”)
• Dr. Lorie Goldkind (Professor, Forom University Graduate School of Social Service; Editor-in-Chief, Journal of Technology in Human Services)

Other referenced people / contributors

• Joe Stevenson
• Mithron Samuel
• Margie Gibelman (mentioned as a mentor; “rest in peace”)
• Richie (Richard) Caputo (mentioned as PhD dissertation chair; recently retired)
• Joanna Creswell Bias (University of Colorado at Colorado Springs)
• Elizabeth Matthews (Forom)
• Lauren Nissen (foresight practitioner who ran the practice lab)
• Research collaborators/institution groups mentioned:
  • Wayne State University
  • Indiana (part of the cited research collaboration)
  • Michigan (mentioned as part of the group)

Organizations / institutions (referenced)

• National Association of Social Workers (NASW)
• University of Texas at Austin
• School of Social Work / Meritz Center for Societal Impact
• Forom University
• Yeshiva University
• Stony Brook (SUNY Stony Brook)
• NASW Illinois
• Dartmouth (associated with the RCT example)

Articles / studies / works referenced

• “AI in social work opportunity at risk” (by Joe Stevenson and Mithron Samuel)
• Blog title mentioned: “Social workers versus AI therapists”
• Randomized controlled trial (12 weeks) involving “Farrabot”
• Harvard Business Review report (about public use cases of ChatGPT)

AI tools / products mentioned

• Microsoft Copilot
• Beam’s Magic Notes
• ChatGPT (OpenAI)
• Gemini
• Quad
• Farrabot
• Character AI (AI companion “Replica” mentioned)
• “Ash” (AI therapy chatbot mentioned in Illinois legislation context)

Legislative/regulatory references

• State legislation cited: Illinois (AI cannot be a therapist), plus mentions of Nevada and Utah.