TLDR: As AI becomes more sophisticated, child welfare professionals will increasingly need to champion the importance of human connection, empathy, and compassion in caring for children and families. Those that do will carve out a distinctive and essential role that uniquely emphasizes the human elements in child welfare, especially as AI capabilities grow and state agencies use AI to increase efficiency.
AI in Child Welfare
All of us who work for the well-being of children must still consider AI as its use becomes increasingly pervasive. Louisiana’s recent announcements regarding the role of AI in child welfare are an example of a state seeking the best path forward.
The State of Louisiana on AI
On October 13, 2025, DCFS Secretary Rebecca Harris informed the Legislature’s Joint Health and Welfare Committee that DCFS has made advancements in AI for Louisiana’s child abuse call center. The Secretary reported that DCFS is using “AI Sentiment,” which listens to callers, identifies signs of an immediate crisis, and sends text alerts to DCFS leadership, asking if leadership wants to hear the call.
Oddly, also on that same day, Louisiana Governor Jeff Landry released a broad-ranging Executive Order (signed two weeks earlier on September 29) requiring that no agency procure or license the use of AI before December 15, 2025.
In many states, AI-related plans for child welfare are swirling with currents from several directions! On the day that Louisiana’s DCFS reported to the Legislature on its “advancements” in using AI to screen child abuse hotline calls, the Governor forbade state agencies from using AI until December 15, and only after specific implementations are approved by Louisiana’s Chief Information Officer.
What Does AI Think About Its Role in Child Welfare?
So, what’s next? Has anyone asked AI what it thinks about these matters?
Do you wonder how AI imagines it might be used in child welfare work? I did, so I asked my locally installed Gemma-3B-12 LLM. I’ll share AI’s answer in just a bit, but first some explanation.
At first glance, Governor Landry’s Executive Order Number JML 25-109 appears to suspend DCFS’s use of AI in child welfare until later this year. Maybe it does, maybe it doesn’t, but I believe the Governor’s caution is wise. While his stated concern seems to be about the use of Chinese AI technologies among all state agencies, I believe caution is especially wise regarding any country’s AI technologies in child welfare.
Second, there are even more important reasons for states to be careful about rolling out AI-based tools for child welfare work.
Child welfare work is an intensely personal endeavor in which foster families, nonprofit partners, and DCFS employees collaborate to protect and care for babies, children, and teens. In a perfect world, these groups work together to provide children in foster care with three federally mandated essentials: safety, permanency, and well-being. None of these three child welfare essentials will spill out of a computer. They require people to care for one another.
Child welfare work is so difficult that if team members don’t also care for each other and their larger team, the risk of harm increases, quality suffers, and effectiveness plummets. AI may provide tools, but it cannot care.
Also, a state child welfare system often requires more volunteers (foster parents) than employees, and the level of appreciation shown to these volunteers by the state agency significantly impacts the pool of potential volunteers. No foster parent will trust the sincerity of an AI’s praise.
Regardless of the AI tools DCFS eventually utilizes, every state leader must recognize that child welfare work requires the human touch. Fortunately, we intuitively understand the value of the human touch in child welfare. For example, few of us, aside from science fiction fans, can really imagine AI-based robots caring for children in foster care. That would be less than human, and entirely inhumane.
While no child welfare professional would place children in the foster care of AI-informed robots, child welfare agencies are considering ways to use AI-based tools to manage child welfare data, screen hotline calls, and even guide treatment planning.
There are good reasons for caution regarding AI in child welfare. AI might provide shiny new tools, but other states have found that AI tools are more like hammers made of ice. They look shiny, but they shatter and children and families have been harmed.
The Prompt
So, here’s the simple, broad question I posed to AI: “Please tell me about the use of AI in the field of child welfare.”
What follows is Gemma-3-12b’s wordy, balanced, and cautionary response. The response offers four ways AI might be helpful in child welfare work:
1. Predictive Risk Modeling
2. Case Management and Workflow Optimization
3. Chatbots and Virtual Assistants
4. Image and Video Analysis
Among these four categories, some state child welfare agencies are trying specific tools with varying degrees of success. Some of the splashiest efforts that made the news ultimately failed, and the states halted their implementation. The details of these early efforts are worth considering in a future report.
To me, the most important part of the AI response isn’t what AI thinks it might do to be helpful. The most important part of the response is the section, Crucial Considerations Across All Applications. Read this material and you will see that AI does not trust itself in child welfare work.
The entire prompt and response follows:
PROMPT:
Please tell me about the use of AI in the field of child welfare.
RESPONSE:
The application of Artificial Intelligence (AI) within child welfare is an evolving and complex area, presenting both significant opportunities and considerable ethical concerns. Its implementation aims to improve efficiency, identify potential risks earlier, and ultimately, enhance the safety and well-being of children. However, careful consideration of biases, transparency, and human oversight is paramount. I will structure this explanation around key areas of AI application, associated challenges, and crucial considerations.
1. PREDICTIVE RISK MODELING:
This is currently the most prevalent application. AI algorithms, specifically machine learning models, are trained on historical data – case files, demographics, service records, and other relevant information – to predict the likelihood of a child experiencing adverse outcomes, such as re-entry into the system, abuse, or neglect.
How it Works: These models analyze patterns within the data to identify factors statistically correlated with negative outcomes. For instance, a model might identify a combination of factors like single-parent household, history of substance abuse, frequent moves, and low income as increasing risk. The algorithm assigns a risk score to each case.
Potential Benefits: Earlier identification of high-risk families allows for proactive interventions, like increased support services, parenting classes, or more frequent check-ins from caseworkers. This could prevent crises from escalating. It also has the potential to optimize resource allocation – ensuring caseworkers focus their time where it’s most needed.
Challenges & Criticisms: This is the area with the most substantial concerns.
Bias: Historical data often reflects existing systemic biases in the child welfare system. If these biases (e.g., disproportionate targeting of certain racial or socioeconomic groups) are embedded in the training data, the AI will perpetuate and amplify them. This can lead to unfair targeting and discriminatory practices. Simply removing “race” as a feature does not eliminate bias; correlated factors (e.g., zip code as a proxy for race or socioeconomic status) can still introduce bias.
Lack of Transparency (The “Black Box” Problem): Many machine learning algorithms, particularly deep learning models, are opaque. It’s difficult to understand why a model assigns a particular risk score. This lack of transparency makes it challenging to assess fairness, identify sources of bias, and justify decisions based on the model’s predictions. Explainable AI (XAI) is a developing field attempting to address this, but it’s not yet universally applicable.
False Positives: Models will inevitably generate false positives – identifying families as high-risk when they are not. This can lead to unnecessary investigations, strained relationships with caseworkers, and feelings of distrust within families.
Self-Fulfilling Prophecies: A high-risk score can lead to increased scrutiny and interventions, which, regardless of the family’s initial risk level, could inadvertently contribute to the predicted outcome. This is especially concerning if interventions are not appropriately tailored and supportive.
2. CASE MANAGEMENT AND WORKFLOW OPTIMIZATION:
AI can streamline administrative tasks and improve caseworker efficiency.
How it Works: AI-powered tools can automate tasks like data entry, scheduling appointments, generating reports, and prioritizing cases based on urgency. Natural Language Processing (NLP) can be used to analyze case notes and extract relevant information.
Potential Benefits: Reduced administrative burden allows caseworkers to spend more time directly with families, conducting assessments, and providing support. Improved data management enhances the quality and accessibility of information.
Challenges: Data integration can be problematic – child welfare agencies often use disparate systems. Implementation costs can be significant. Requires substantial training for caseworkers.
3. CHATBOTS AND VIRTUAL ASSISTANTS:
These tools can provide immediate support and information to families and caregivers.
How it Works: AI-powered chatbots can answer frequently asked questions about child welfare services, provide information on parenting resources, and offer emotional support.
Potential Benefits: 24/7 accessibility, reduced wait times for information, and increased access to support for families who may face barriers to traditional services.
Challenges: Limitations in understanding complex or nuanced situations. Potential for providing inaccurate or inappropriate information. Lack of human empathy and connection. Requires careful monitoring and oversight. Data privacy is paramount – chatbots must comply with relevant regulations.
4. IMAGE AND VIDEO ANALYSIS (EMERGING):
There’s nascent research into using AI to analyze images and videos for signs of abuse or neglect. This is highly controversial.
How it Works: Algorithms could analyze photos or videos submitted as evidence (e.g., from social media or digital devices) to detect indicators of physical harm or unsafe environments.
Potential Benefits: Potentially faster identification of urgent situations.
Challenges: Extremely high risk of false positives and misinterpretations. Serious privacy concerns. Significant ethical considerations around consent, data security, and potential for misuse. This application is far* from being ready for widespread implementation. The potential for harm outweighs the potential benefits at present.
CRUCIAL CONSIDERATIONS ACROSS ALL APPLICATIONS:
Human Oversight: AI should augment, not replace, human caseworkers. Human judgment, empathy, and critical thinking are essential in child welfare decision-making.
Data Quality and Governance: High-quality, reliable data is the foundation of any AI system. Robust data governance policies are needed to ensure data accuracy, completeness, and security.
Transparency and Explainability: Efforts should be made to make AI models more transparent and explainable, so that users can understand how they work and why they make the decisions they do.
Bias Mitigation: Ongoing monitoring and evaluation are needed to identify and mitigate bias in AI systems. Data diversification and algorithmic fairness techniques are crucial.
Ethical Frameworks and Regulations: Clear ethical frameworks and regulations are needed to govern the use of AI in child welfare, ensuring that it is used responsibly and in a way that protects the rights and well-being of children and families.
Stakeholder Involvement: Families, caseworkers, and community members should be involved in the design, development, and evaluation of AI systems
Summary
So, that’s the entire prompt and response.
When we distill AI’s response, we see there will be good uses for AI in child welfare. But AI surprises us by emphasizing the “Challenges” associated with each use and warning us about “Crucial Considerations” to remember in all uses of AI in child welfare.
So, the crux of the matter for now is this: We cannot trust a tool that does not trust itself.
And the future story will likely unfold like this: AI will continue to improve, and its tools will become less brittle and more pervasive. This means an important task for child welfare professionals tomorrow will be to promote and defend the role and value of human touch and compassion in child welfare.
Rick Wheat
Director
Center for Child Well-being