The Ethics of Inaction: When Avoiding AI May Compromise Client Care

Overview & Clinical Significance

Adam Ventura's presentation makes a provocation that few practitioners will have encountered before: in a field driven by data, effectiveness, and client-centered outcomes, refusing to use beneficial technologies may itself constitute an ethical risk. The prevailing discourse around AI in ABA has focused on the risks of using it—confidentiality, data integrity, professional displacement. Ventura inverts this framing and asks the harder question: what happens to clients when practitioners categorically avoid tools that could meaningfully improve service quality?

The BACB Ethics Code (2022) Code 2.01 requires behavior analysts to rely on current scientific knowledge when designing and implementing services. Code 2.09 requires that services be effective. If AI tools genuinely improve documentation accuracy, reduce assessment time, support better supervision delivery, or enable practitioners to see more clients without compromising quality—then declining to use them, on principle rather than on evidence, may produce a worse outcome for clients than thoughtful adoption would.

This does not mean practitioners are obligated to adopt every AI tool that becomes available. It means the evaluation of AI tools should be symmetric: the same evidence-based reasoning practitioners apply to intervention selection should be applied to technology adoption decisions. Rejecting AI without examining the evidence is no more defensible than adopting it without examining the evidence.

Kaur et al. (2026) documented that social functions of behavior are frequently missed when assessment is restricted to a narrow set of tools. The same principle applies to practice enhancement: restricting the practitioner's toolkit to only pre-AI methods may produce blind spots in domains where AI provides genuine incremental value.

Background & Context

The history of behavioral service delivery includes repeated cycles in which new technologies were initially resisted on grounds that turned out not to be proportional to the actual risks. Telehealth is the most recent example: initial resistance based on concerns about fidelity, rapport, and clinical quality gave way—partly under COVID-19 pressures—to an evidence base showing that telehealth-delivered ABA can produce comparable outcomes to in-person delivery for many client populations and many treatment targets.

AI is the current iteration of this cycle. The concerns raised about AI—confidentiality, accuracy, displacement of clinical judgment—are legitimate and require structured management. They are not inherently prohibitive, and treating them as prohibitive without examining the evidence is an evidence-based failure.

Dawson et al. (2026) reviewed functional communication training and found that the response topographies most effective for individual clients varied based on their specific motivating operations and communication history. The lesson for AI adoption is similar: the value of AI tools varies substantially by application context and client profile, which means blanket adoption and blanket rejection are both epistemically inadequate responses.

Kaye et al. (2025) demonstrated that combining multiple assessment approaches—antecedent analysis plus functional analysis—produces more accurate function hypotheses than either approach alone. Technology adoption decisions benefit from the same combined-methods logic: rather than evaluating AI tools in isolation, evaluate them in the context of the specific practice problems they are being applied to and the specific clients they are intended to serve.

Clinical Implications

The clinical implications of Ventura's argument are most direct in three areas: documentation, assessment support, and supervision delivery. In each area, AI tools have demonstrated incremental value in specific applications, and practitioners who have not evaluated these tools may be missing efficiency gains that would allow them to provide better or more frequent services to clients.

In documentation, AI-assisted note drafting can reduce the time BCBAs spend on administrative tasks and reallocate that time to direct service or supervision. For practitioners with high caseloads, this reallocation can be clinically significant: more time for supervision means more development for RBTs, which means better service quality for clients. The ethical obligation is not to use AI for documentation—it is to ensure that documentation quality does not suffer because of time constraints that AI could address.

Tong et al. (2026) found that behavior problems in autistic children are associated with a complex array of variables including mealtime behavior and sibling risk profiles. Assessment of complex, multi-dimensional presentations requires efficient data collection and synthesis tools.

AI tools that help practitioners organize and identify patterns in complex behavioral data can improve assessment quality—provided the practitioner validates the output against their clinical knowledge.

Treviño & Gerstein (2026) validated an emotion dysregulation assessment instrument, demonstrating the value of precise, psychometrically sound measurement tools. AI tools that assist in analyzing structured assessment data can provide similar precision when used correctly. The practitioner's obligation is not to avoid these tools but to evaluate them with the same rigor applied to any other clinical instrument.

Ethical Considerations

The ethical argument in Ventura's presentation rests on Code 2.01's requirement to use current scientific knowledge and Code 2.09's requirement to provide effective services. Both provisions are forward-looking: they require practitioners to track the state of the science and adopt practices supported by current evidence, not only practices that were standard when the practitioner was trained.

The BACB Ethics Code does not explicitly address AI, but it does require practitioners to stay current with developments in the field (Code 1.01) and to provide services at the highest quality level available (Code 2.01). If AI tools have demonstrated value in specific ABA applications, a practitioner who has categorically declined to evaluate them may be out of compliance with Code 1.01's currency requirement.

Goodhew & Edwards (2026) found that the assessment instruments practitioners choose for theory of mind evaluation vary substantially in their sensitivity and reliability—a reminder that the choice of tools is itself a clinical decision with ethical implications. The same is true of AI adoption: declining to evaluate whether AI tools could improve service quality is a tool selection decision with potential ethical implications.

Samadi et al. (2026) demonstrated the clinical value of validated, domain-specific assessment instruments. AI tools that support the administration, scoring, or interpretation of validated instruments can extend their value.

Practitioners who are not aware of these applications may be making tool selection decisions based on incomplete information about what is currently available.

Assessment & Decision-Making

Ventura's argument suggests a structured evaluation protocol for AI tools that mirrors the practitioner's existing approach to evidence-based practice: identify the clinical problem, review the available evidence for AI solutions, pilot the tool in a controlled context, measure outcomes against a baseline, and make an evidence-based adoption decision.

This protocol transforms AI adoption from a values decision (am I comfortable with AI?) into a clinical decision (does this tool improve outcomes for my clients in this specific application?). The answer may vary by application: an AI tool that adds value for documentation may add no value for assessment, and vice versa.

Kaur et al. (2026) showed that assessment approaches that examine multiple social contexts produce more complete function hypotheses. The same multi-context evaluation logic applies to AI tools: piloting a tool across multiple use cases and client types produces a more accurate picture of its value than evaluating it in a single application.

Dawson et al. (2026) found that functional communication training outcomes depend on selecting response topographies that fit the individual client's profile and environment. AI adoption decisions benefit from the same individualized approach: the question is not whether AI is useful in general but whether this specific tool is useful for this specific clinical problem in this specific practice context.

Al Aqel et al. (2026) found that awareness and attitudes shape engagement with services. Practitioners' own attitudes toward AI—whether skeptical or enthusiastic—should be held as hypotheses to be tested, not conclusions that determine adoption decisions before evidence is gathered.

What This Means for Your Practice

The practical application of Ventura's argument is to identify one area of your current practice where inefficiency is directly affecting service quality—documentation time, assessment synthesis, supervision frequency—and evaluate whether an AI tool could address that inefficiency without compromising accuracy or confidentiality.

This evaluation should be systematic: define the problem, specify the quality threshold the tool must meet, pilot it on a small sample of cases with your own independent verification, measure the outcome, and make a decision based on the data. That process is behavior-analytic in its structure, even though its subject is technology adoption rather than client behavior.

Martín-Díaz et al. (2026) documented that motor function in autistic youth varies substantially across environments and task demands. The same variability characterizes AI tool performance: tools that produce accurate outputs in one context may produce poor outputs in another.

Systematic evaluation across the relevant contexts is required before adoption—but systematic evaluation requires engagement with the tool, not categorical avoidance.

Treviño & Gerstein (2026) validated an assessment instrument that could previously only be estimated clinically. AI tools represent a similar opportunity in some domains: converting clinical estimates into more precise measurements. Practitioners who have not evaluated these tools may be providing services with more measurement error than necessary, which affects the accuracy of clinical decision-making and the quality of care.