AI Model Wiki

    Skip to main content
    15 min
    Lesson 6 of 6

    Common Pitfalls & How to Avoid Them

    From the course AI for Social Workers

    Even well-intentioned AI adoption in social work can go wrong. Understanding the most common failure modes helps you avoid them before they affect clients.

    Pitfall 1: Using Unapproved Public AI Tools with Client Data

    What happens: A worker pastes case notes into a free chatbot to get a quick summary. The tool's terms of service allow training on user inputs. Client data is now outside agency control.

    How to avoid it:

    • Treat any AI tool not explicitly approved by your agency as off-limits for real client information.
    • Use fictional vignettes for practice and skills-building.
    • Advocate for your agency to establish a clear approved-tools policy rather than working around the gap.

    Pitfall 2: Treating AI Output as Clinical Judgment

    What happens: An AI risk assessment score is recorded in a case file without annotation. A supervisor later acts on the score without knowing it came from an automated tool.

    How to avoid it:

    • Always label AI-generated content as such in case records.
    • Add your own professional assessment as a distinct field.
    • Document that you reviewed and either agreed with or modified the AI output.

    Pitfall 3: Overlooking Algorithmic Bias in Risk Tools

    What happens: A child welfare agency adopts a predictive tool trained on historical removal data. Because historical decisions reflected systemic racial bias, the model flags Black families at higher rates even when controlling for poverty.

    How to avoid it:

    • Ask vendors for bias audits before adopting any risk-scoring tool.
    • Track your own outcomes: are certain demographic groups receiving different recommendations?
    • Report disparities through proper channels — this is part of your ethical obligation.

    Pitfall 4: Automation Complacency

    What happens: A worker relies on AI-generated resource lists without verifying that phone numbers, addresses, and eligibility rules are still current. A client shows up at a shelter that closed six months ago.

    How to avoid it:

    • Verify time-sensitive information (contact details, eligibility, wait times) before passing it to clients.
    • Build a habit of spot-checking one or two items per AI-generated resource list.
    • Note the verification date in the resource document.

    Pitfall 5: Informed Consent Gaps

    What happens: AI is used to summarise client conversations or generate risk scores, but clients are never told. This undermines trust and may violate agency policy or law.

    How to avoid it:

    • Include AI use disclosure in your agency's standard consent forms.
    • Explain to clients in plain language what AI does in their case (e.g., "a computer helps me write up our session notes, and I review everything before it goes in your file").
    • Allow clients to opt out and document that option.

    Quick Reference: Pitfall Checklist

    • Tool is agency-approved for client data
    • AI output is labelled and my judgment is documented separately
    • Risk tool has a published bias audit
    • Resource information has been verified
    • Client has been informed and consented

    Key Takeaways

    • •Never use unapproved public tools with real client data — advocate for an agency-approved list instead
    • •Always label AI output in case records and document your own separate professional judgment
    • •Demand bias audits for any risk-scoring tool and track your own demographic outcome data
    • •Verify time-sensitive resource information before sharing with clients, and document client consent for AI use
    0 of 6 lessons marked complete

    Progress is kept in this browser only.