Beneficence and Creativity

I use AI when it meaningfully contributes to thinking, learning, access, or communication. With tools such as ChatGPT and Claude, I may test an argument, ask for counterarguments, experiment with organization, identify questions I have overlooked, or explore how an idea might be communicated differently.

This is not a conversation with an independent interlocutor. It is closer to thinking through a technologically mediated reflection of human knowledge. Sometimes the reflection exposes a weakness in my argument. Sometimes it produces an unfamiliar combination that opens another direction. Sometimes it simply returns a polished version of an assumption I already brought to it. The work is determining which is which.

This is where Vallor’s (2024) distinction between reflection and wisdom becomes especially useful. A mirror can contribute to practical wisdom, but the reflection itself is not wisdom. Creativity still requires judgment about which possibilities deserve to survive, which should be rejected, and which require further investigation.

I therefore distinguish assistance from delegation. AI can help me discover another way of approaching a problem. It should not create the appearance that I possess knowledge, reasoning, or expertise I have not developed myself.

Accessibility

Accessibility is also central to my use of technology. Academic work does not begin only when words appear on a page. Thinking can occur through speech, listening, movement between modalities, and recursive conversation. Technologies that allow movement among these forms can reduce barriers without reducing intellectual expectations.

I regularly use Plaud and Voicepal for dictation and audio processing. Speaking allows me to externalize ideas while they are developing, particularly when thoughts are moving faster than conventional writing allows or are becoming diffuse and repetitive. I can then return to those ideas, organize them, interrogate them, and revise them deliberately. Speechify allows me to listen to written material at increased speeds, which can support attention and reduce distracting behaviors during reading.

These practices are consistent with emerging research on AI and disability. Pierrès and colleagues (2025) found that students with disabilities used generative AI to assist with writing, reading, research, teaching, and self-organization. Their findings are important because they complicate the assumption that technological assistance necessarily represents avoidance of academic work. Sometimes technology reduces barriers that are extraneous to the learning objective.

Mullan's (2023) critique of professional systems similarly asks us to value choice, accessibility, and multiple ways of knowing rather than treating one normative pathway as inherently superior. Accessibility does not mean lowering standards. It means questioning whether a barrier is actually necessary to demonstrating the competence being assessed.

Research, Knowledge, and Non-Maleficence

My research workflow also includes SciSpace, Elicit, Litmaps, and Connected Papers for discovering scholarship and tracing relationships among sources, alongside Zotero for organizing references and maintaining a scholarly library. These tools help me locate possibilities. They do not determine what counts as evidence.

I maintain a clear distinction between discovery and verification. A citation generated by an AI system is not evidence that a source exists. A summary is not a substitute for reading the original scholarship when that source becomes consequential to an argument. References, quotations, findings, and substantive claims must be checked against the original source before I use them. This is especially important because generative systems can fabricate information while presenting it with considerable linguistic confidence (Bowen & Watson, 2024; Heersmink et al., 2024).

Non-maleficence also requires attention to privacy, bias, and deskilling. I do not enter identifiable clinical, student, supervisee, research participant, or other confidential information into generative systems without appropriate authorization and safeguards. I remain attentive to whether repeated reliance on a tool is strengthening my ability to think or quietly replacing a skill I need to retain. Pierrès and colleagues (2025) similarly found that students valued AI's accessibility benefits while remaining concerned about privacy, overreliance, inaccurate information, diminished critical thinking, and the loss of human interaction.

Transparency and Accountability

Transparency means that I should not allow another person to reasonably misunderstand how significant intellectual work was produced. I follow the requirements of the instructor, university, journal, conference, funder, or research setting governing a particular project. When generative AI materially shapes the argument, organization, analysis, coding, or language of scholarly work, I disclose that use when required or when nondisclosure would create a misleading impression of the process.

Not every technological intervention carries the same ethical weight. Dictation, text-to-speech, citation management, and literature mapping serve different functions from asking a generative model to produce substantive prose or analysis. Ethical disclosure should therefore attend to what the technology actually contributed rather than collapsing all digital assistance into a single category.

Power and the Politics of Knowledge

My position also extends beyond individual academic integrity. AI participates in systems of knowledge production that are already structured by unequal power. McDowell and colleagues (2023) argue that critical scholarship requires asking who benefits from knowledge claims, whose experiences are rendered silent, whose perspectives dominate, and what gets recognized as legitimate knowledge. These questions become more urgent when systems trained on enormous collections of existing human material begin mediating what information appears plausible or important.

Mohamed and colleagues (2020) likewise argue that values and power are inseparable from AI development. Decisions about what enters a dataset, what is ignored, what counts as valid knowledge, and whose interests technologies serve are ethical decisions rather than technically neutral ones. Mullan (2023) similarly calls for the democratization of professionalized knowledge and challenges systems that treat institutional knowledge as though it holds a monopoly on legitimate ways of knowing. AI does not reflect humanity evenly. Vallor (2024) argues that contemporary systems reflect only partial records of human experience and can amplify patterns that already dominate those records. What appears prominently in the mirror may therefore tell us as much about historical visibility and power as it does about human reality.

For this reason, I do not understand ethical AI use simply as checking citations and avoiding plagiarism. It also means asking whose voices a system amplifies, whose language becomes normative, what forms of knowledge are absent, who benefits from technological efficiency, and who absorbs its harms. AI may increase access while simultaneously reproducing ableism, racialized assumptions, economic inequality, and dominant epistemologies. These dialectic possibilities must remain visible and we must confront: Whose knowledge is being reflected? What is absent from the reflection? Which assumptions are being repeated because they are statistically common? And when should I look away from the mirror and seek knowledge elsewhere?

My Ethical Standard

The specific technologies I use will change.

Beneficence asks whether a tool contributes something worthwhile. Non-maleficence asks what it may damage or displace. Transparency asks whether I am representing my process truthfully. Accessibility asks whether technology can remove unnecessary barriers to participation. Creativity asks whether it expands rather than constrains thought. Attention to power asks whose knowledge and interests are being reproduced.

Across all questions I pose, accountability remains with me. I must not outsource judgment, interpretation, citation, critical thinking, or ethical responsibility to a language model. If my name appears on the work, I must be able to understand it, verify it, defend it, and answer for what it does in the world.