Researching with AI — From Case Conceptualization to Publication
The essence of research has not changed. What has changed is where human effort is most needed. As tools accelerate, the researcher's questions become even more important.
Period 2 Experience: AI Case Conceptualization
This Week's Reading: Can AI Assist Counseling Supervision?
Supervision is the educational process where experienced senior counselors (supervisors) review and provide feedback on novice counselors' (trainees') counseling. “AI supervision” means AI analyzes verbatim transcripts, simultaneously shows cases from multiple theoretical perspectives, and provides immediate feedback to support trainee learning.
Case conceptualization is the process of systematically understanding a client's problems within a theoretical framework and developing treatment strategies. Gainer(2025) argues in Chapter 6 that AI can “augment” rather than “replace” this process.
세네카 R. 게이너
AI supervision isn't trying to replace the wisdom of human supervisors. It's trying to supplement the 'theoretical tunnel vision' that can occur when a novice counselor depends on only one supervisor.
— Gainer (2025, p. 168)
AI supervision's greatest advantage is the ability to “show multiple perspectives simultaneously.” AI can analyze the same case from multiple perspectives without theoretical bias.
What Happens When You View the Same Case Through Three Theories?
The core question of case conceptualization is “why is this person, at this point, experiencing this problem?” By analogy, it's like taking the same landscape with an infrared camera, a regular camera, and X-ray — three completely different photos.
CBT Frame:Beck's cognitive model analyzes cases in five layers.
Triggering event — The specific situation where the problem began.
Automatic thoughts — Thoughts that arise automatically in that situation.
Cognitive distortions — Thinking traps contained in those thoughts.
Core beliefs are usually formed in childhood and operate for life. 'I am unlovable,' 'I am incompetent,' 'The world is dangerous' — these beliefs become the factory of automatic thoughts.
— Beck (1995)
Psychodynamic Frame:Luborsky's CCRT method analyzes the client's relationship patterns. Three elements: Wish, Response of Others, Response of Self.
레스터 루보스키
People repeat their core conflictual relationship themes without being aware of it. The experience of rejection from parents replays in relationships with bosses, friends, and even counselors.
— Luborsky & Crits-Christoph (1998)
Humanistic Frame:Rogers's person-centered therapy focuses on “how clients experience their feelings” rather than diagnosis.
In one sentence: CBT sees “this person's thinking is the problem,” psychodynamic sees “this person's relationship pattern is the problem,” and humanistic sees “this person's environment is the problem.” The core value of AI supervision is the ability to use all three lenses simultaneously.
Verbatim Analysis: Where AI Particularly Excels
Verbatim is text that records every word of a counseling conversation. AI can analyze emotional flows, thematic patterns, and counselor-client interaction patterns across the entire verbatim transcript.
마이클 램버트
If meaningful change doesn't appear within the first 3-5 sessions, the treatment strategy should be revisited. The sooner the feedback, the better.
— Lambert (2013)
Lambert's research raises another implication. It's important to quickly notice when counseling isn't working. AI can detect subtle emotional changes within sessions from verbatim transcripts alone.
칼 로저스
The most powerful healing tool in counseling is not technique but relationship. The experience of one human accepting another as they are within that relationship creates change.
— Rogers (1961)
Ultimately, AI supervision's role is clear. AI is a “first-pass analysis tool.” But the subtle texture of the counseling relationship, nonverbal cues, and interpretation of cultural context remain in the domain of human supervisors.
Class principle: Only HIPAA/BAA-compliant tools are used for actual client data. Today's exercises use fictional cases.
Creating Verbatim Transcripts: AI Transcription Tool Comparison
The first step in case conceptualization practice is creating a verbatim transcript. In the past, this meant typing word by word while listening to recordings (30-minute session = 3-5 hours of transcription). AI transcription tools reduce this to minutes. There are three key criteria for choosing a tool.
Korean language accuracy — Counseling conversations contain colloquial speech, interjections, and silences
Speaker diarization — Must distinguish between counselor and client speech
Security/Ethics — Client information protection is essential. Check HIPAA compliance
Practice Workflow: Recording → Transcript → AI Analysis
Record — Conduct a mock counseling session and record using Clova Note or Otter
Transcribe — Download the AI auto-transcribed text (txt or docx)
Review — Scan the transcript and correct errors (especially technical terms)
AI Analysis — Input the reviewed transcript into Claude and request case conceptualization
Compare — Compare analysis results from CBT, psychodynamic, and humanistic perspectives
The entire process can be completed within class time. In the past, creating a verbatim transcript alone took days, but thanks to AI transcription tools, there is now time to focus on 'analysis.' The tool handles transcription; the human handles interpretation.
Period 2: Research & Paper Writing with Claude Cowork
The bottleneck in research has shifted. In the past, data collection and analysis took time. Now AI can read hundreds of papers, find patterns, and write drafts. The remaining bottleneck for researchers is judging 'what question to ask' and 'why these results matter.'
What Is Claude Cowork?
Claude Cowork is an agent feature provided by Anthropic's Claude Desktop app. It differs from regular chat in three ways.
Direct local file access — It reads, modifies, and creates files and folders on your computer. No need to upload paper PDFs, notes, or data files — just specify the folder path.
Multi-step autonomous execution — If you say 'Read the 10 papers in this folder and organize them by topic,' it creates a plan and executes it step by step. It asks for confirmation when needed.
Direct output generation — It saves organized documents, Excel tables, and presentations directly as files on your computer.
Practice: Organize Materials and Write Papers with Cowork
What AI Does Well vs. What the Researcher Must Do
What AI Does Well
What the Researcher Must Do
Summarizing/comparing large volumes of papers
Setting research questions
Discovering data patterns
Interpreting why those patterns matter
Structuring in APA format
Designing the logical flow of argumentation
Grammar and expression correction
Ethical judgment (IRB, consent forms, etc.)
Generating statistical analysis code
Interpreting clinical significance of results
The core value of Cowork is 'automating repetitive tasks.' It can reduce the 2 weeks spent reading and organizing literature to 2 hours. In that saved time, the researcher can think about 'what do these results mean for counseling practice?' As tools get faster, human judgment becomes even more important.
We Can Build It Ourselves
Tools like Upheal and Mentalyc are ultimately combinations of three AI capabilities: (1) speech-to-text transcription, (2) text-to-clinical-note conversion, (3) multi-perspective analysis. Today in class, we performed all three steps using prompts. Combining these into a single app creates a tool for counselors.
Upheal charges $49/month because of HIPAA infrastructure and medical regulation compliance costs. For our class project, the goal is to build a prototype with fictional data and document the security requirements needed for actual clinical deployment. The technology already exists — today in class we ran the entire pipeline with prompts. Packaging it into an app is an engineering problem, not a research problem.
Discussion Topics
As AI accelerates the information-to-knowledge conversion, the bottleneck in research has shifted. Researchers are now most needed at the stage of turning knowledge into wisdom. Share your AI experiences and discuss how to apply agile research methodology to our projects.
