In today’s column, I am continuing my ongoing series about the impact of generative AI in the health and medical realm. The focus this time is once again on the mental health domain and involves the ins and outs of data-training generative AI to get the AI to perform mental health therapy as a specialty.
I have previously examined numerous interleaving facets of generative AI and mental health, see my comprehensive overview at the link here. You might also find of notable interest a CBS 60 Minutes episode that recently examined crucial facets of this evolving topic, see the link here (I am honored to indicate that I was interviewed as an expert on this topic and appear in the episode, see the link here).
Other useful background you might find of interest includes my coverage of mental health chatbots that have been bolstered by generative AI (see the link here) and the rapidly changing nature of the client-therapist relationship due to generative AI at the link here. I explored where things are headed regarding the levels of AI-based mental therapy autonomous guidance at the link here, and showcased the importance of the World Health Organization (WHO) report on global health and generative AI at the link here, and so on.
The core idea in this discussion is to ascertain how far we can push generative AI toward adequately engaging in mental health advisement. No human therapist will be in the loop. It is just a person and generative AI carrying on a mental health session, notably with the AI acting in the capacity of a seemingly versed therapist or clinician.
There are numerous methods or approaches that can be used to accomplish this goal. The attention herein consists of an innovative approach using real-world transcripts of client-therapist sessions as a primary source of data. This rich data is fed into generative AI and large language models (LLMs) to computationally pattern-match the nature of highly engaged conversational mental health guidance. The goal is to have the AI computationally mimic the therapeutic process, doing so on an always-on 24×7 basis and accessible from anywhere in the world.
Generic Generative AI As Mental Health Advisor Doesn’t Cut It
Let’s begin at the beginning.
When you use modern-day generative AI, most of the time it is working in a considered generic capacity.
Here’s what that means.
The AI maker scanned the Internet to data train the generative AI and did so across a vast swath of the online world. That’s what helps the AI seem fluent. A massive scale pattern-matching on human writing has proven to be a handy way to have generative AI appear highly conversant in a wide array of topics.
The problem of sorts is that this is a jack-of-all-trades that has no specific niche to call its own. The moment you want to dig deeply into a particular domain, such as law, medicine, and other deep-rooted knowledge-based realms, you hit the end of the road. Generative AI can only vaguely seem to engage in dialogues whenever you ask questions in areas involving keen expertise. There isn’t any there, in there, if you know what I mean.
Okay, that’s a problem for anyone desirous of using generative AI as an expert or anything along those lines. The generic generative AI such as ChatGPT, GPT-4, Bard, Gemini, and Claude are a mile long and an inch deep on nearly any chosen topic. Worse still, the AI will often try to trick you into thinking that there is substance to be found. You can get answers or responses that smack of expertise, but the moment you do a double-check, the gig is up. A big-time example last year involved two attorneys who got into legal hot water due to carelessly believing generic generative AI that cited false legal cases, see my coverage at the link here.
How does this apply to the realm of mental health and the role of mental health therapists?
Easy-peasy, the answer is that performing mental health advice or therapy is a distinct form of expertise.
I dare say that any licensed mental health professional knows this to be the case. The public at large is somewhat unsure about the matter, partially because we’ve had a humongous growth in so-called well-being advisors and so-called life coaches. These spin-offs have indubitably blurred the line of what is a therapist versus a non-therapist in any credentialed sense of the word.
I previously explored how well generic generative AI could do on a mental health therapist licensing exam, see the results at the link here. The bottom line is that the generic generative AI did amazingly well, but not sufficiently impressive or convincing that you can start relying on what generically is taking place.
The aim then is to try and turn generic generative AI into a customized mental health therapist version of generative AI. That is the ticket to fame and fortune. Well, this will also have a lot to do with quality of care. I’ve repeatedly exhorted that the riffraff of chatbots and proclaimed AI-based mental health apps is enveloping society at large into a grand experiment.
We all are the guinea pigs.
The experiment is whether zillions of wanton apps that have little or no bona fide mental health therapeutic capacities, but that are marketed as though they do (see my coverage at the link here), will grandly harm us and likewise psychologically harm generations to come. Right now, no one can answer that probing and vital question. There is a mess on our hands, sitting in plain sight. The clock is ticking.
Anyway, we shall return to the matter at hand here, namely how can we transform generic generative AI into well-devised generative AI that is sufficiently versed computationally and mathematically in the realm of mental health to carry out mental health therapy reliably and safely?
Great question!
Let’s see what possibilities exist.
Churning Butter To Make Mental Health Versed Generative AI
Here is a quick rundown of the major paths being undertaken to turn generic generative AI into mental health generative AI:
- Remain generic. Generative AI is further broadly data-trained on mental health matters and not focused on this as a core specialty per se.
- Advanced prompting. Generative AI can be pushed toward mental health therapy advisement using advanced prompting approaches.
- Transcripts of therapeutic sessions. Use therapist-client therapeutic transcripts to data-train generative AI accordingly.
- Ingest via RAG. Utilize the in-context modeling capabilities of generative AI and ingest mental health domain data via RAG (retrieval augmented generation).
- Build from scratch. Start anew when building an LLM and generative AI by having mental health therapy as a foundational core to the AI.
- Other approaches.
I’ve covered each of those approaches and more in my writings, including extensively my two books covering the latest in AI for mental health, see the link here and the link here.
Let’s look at the third bulleted item above, entailing the use of transcripts for data training a generic generative AI to become more computationally versed in the facets of performing mental health therapy.
Consider the following. If we were able to collect together tons of therapist-client session transcripts, we could feed that data into generic generative AI or a large language model (LLM). In a manner similar to when being data trained across a wide swath of the Internet, we are merely going deeply into the mental health space. This could be done by also feeding books about mental health therapy and other written works, which I’ve covered previously, see the link here.
In the case of transcripts, we want the AI to pattern-match on how therapists engage in therapeutic discussions with their clients or patients. The more transcripts we feed into the data training process, the better off we are. Think of things this way. If we only fed transcripts of one particular therapist, even a famous one, we would essentially be hobbled by having the AI only pattern-match on that one style or clinical approach.
Our overall assumption is that by feeding transcripts across a wide variety of therapists, and a wide variety of clients or patients, the AI can discover broad patterns and apply those as needed. If you are curious about what happens if we do narrow our focus to a particular therapist, see my use of the generative AI feature known as personas to simulate Sigmund Freud performing a therapeutic session in our current times, at the link here.
I’ve got good news on this.
There are lots of entities such as private companies and government or educational organizations that have collected together entire databases of mental health therapy transcripts. Thus, the data is sitting out there, waiting to be used for this divine purpose. The odds are that many of those entities don’t realize how valuable their data is for the AI community. Up until now, the data is typically used for human-to-human training purposes. Someone in training to become a therapist looks at the transcripts to understand how to engage in dialogues with patients or clients.
It is a gold mine, waiting to be explored.
But the world is never that easy.
A slew of thought-provoking questions arises on these weighty matters:
- Will raw transcripts be readable or will misspells, utterances, and other transcribing considerations foil attempts to get generative AI suitably data trained?
- How much editing might be needed to get the transcripts into proper shape (and will this only be doable by hand, or can automation help to make the refinements)?
- Are there therapist annotations that could accompany transcripts and provide yet another valuable data source for doing a more robust data training of generative AI?
- Might there be recorded their-party expert therapist critiques of the transcripts, allowing that added data to be fed into the generative AI?
- Does it make a difference whether there is augmented data that depicts the personal background and work experience of the therapists conducting the transcribed sessions?
- Is there a possibility of privacy intrusion looming over the therapist or the client/patient by using the data in this manner?
- Are there intellectual property rights at stake such as copyright infringement that might be invoked by opting to use the data in this manner?
- How costly will it be to obtain the data?
- What is the cost and effort required to prepare the data for importing into generative AI?
- How will we be able to suitably assess that the data training has accomplished our goal?
- Will post-training fine-tuning and refinement be required, and if so, what magnitude of effort will be required?
- Etc.
I realize that seems like a daunting list.
My reply is that it is better to go into this with your eyes wide open. Leaping into this approach blindly without getting your ducks aligned is a surefire guarantee of failure. I am reminded of the sage wisdom stated aptly by Abraham Lincoln: “Give me six hours to chop down a tree and I will spend the first four sharpening the axe.”
A true statement, including in this context.
I suppose one of the biggest questions is whether there are enough of these kinds of transcripts available.
Here comes a potentially gloomy sad face. It could be that despite buying, licensing, or otherwise obtaining every iota of mental health therapy transcripts, it isn’t enough. Keep in mind that the number of essays, narratives, poems, and textual content that was scanned across the Internet was immense at the get-go. The volume of mental health transcripts will be a tiny drop in the bucket.
Of course, that’s not a sensible apple-to-oranges comparison per se. All we need is enough transcript data to push generic generative AI up the ladder toward being suitably capable of mental health therapy. Furthermore, we might not necessarily need every last drop. Perhaps we might get lucky and can achieve our aims with only a percentage of what might otherwise be obtained.
I’ll let you mull that over.
Example Of Data Training Via Use Of Therapist-Client Transcripts
I will next proceed to showcase various activities involved in data training generative AI in the specialized domain of mental health therapy. I am going to use ChatGPT to showcase my examples. ChatGPT is a sensible choice in this case due to its immense popularity as a generative AI app. An estimated one hundred million weekly active users are said to be utilizing ChatGPT. That’s a staggering number.
A few quick comments before we launch into using ChatGPT.
If you are going to try to do the same prompts that I show here, realize that the probabilistic and statistical properties will likely produce slightly different results than what I show here. That’s the nature of generative AI and how it is devised.
You also should expect that different generative AI apps will respond in different ways. Not all generative AI apps are the same. The key is that sometimes a particular prompt will work in one generative AI app and not another. You will need to undertake a series of trial-and-error attempts to ferret this out.
There is also the timing factor. A prompt that worked today might not work the same way tomorrow. Most of the generative AI apps are continually being updated. The updates might alter internal facets that could change how the AI reacts to your prompting.
We are ready to get underway with ChatGPT.
My opening move in this chess gambit will be to provide an actual therapist-client transcript to ChatGPT and provide a prompt that gets the AI to closely examine the transcript. I opted to use a publicly available therapist-client transcript on the Internet that was posted for instructional purposes:
- “The website was created as a resource to help people who are training in psychology, counseling or social work (or any similar profession). It is a compilation of the school work completed by a counselor in training (from 2004 to 2008). It was noted that in order to continuously grow and remain a competent therapist, there has to be quick accessibility of information. The information presented, in addition to refreshing terms and theoretical perspectives, is a reminder of how much professional growth one acquires during the transition from a clinical counseling intern, to a seasoned one. The strive for professional growth will remain a goal throughout. In an effort to conserve privacy for clients who were part of therapy sessions during this training, fictitious names were used.” (Source: “Therapy Transcript: An Analysis of a Session”, posted April 17, 2015, at this link here).
To bring you quickly up-to-speed, a female client is discussing with her therapist a series of issues associated with her ex-boyfriend. Along the way, aspects of her relationship with her father enter into the session.
Here we go.
- My Entered Prompt: “I am going to provide you with a transcript of a clinician and their client having a therapeutic session. The transcript uses the word “Clinician” at the start of a sentence to indicate that the sentence is a remark made by the clinician. The transcript uses the word “Client” at the start of the sentence to indicate that the sentence is a remark made by the client. The two are carrying on an interactive dialogue. I want you to read the transcript and be ready to answer questions about it. Do you understand these instructions?”
- ChatGPT generated response: “Yes, I understand the instructions. Please provide the transcript, and feel free to ask any questions you have about it afterward.”
One aspect that I’d like to mention is that I purposely cleaned up the transcript before feeding it into ChatGPT. I corrected misspelled words. I removed words that were partial utterances or filler words. All in all, I tried carefully to not do anything that impacted the meaning of the transcribed conversation.
That being said, an immediate qualm would be that perhaps those aspects ought to be taken into consideration by the generative AI. My having corrected or removed seemingly trivial elements might be an inadvertent mistake on my part. It might be better to feed the whole kit-and-kaboodle into ChatGPT.
The reason I opted to do some mild editing was that I had concerns about the potential unreadability of some portions. If the AI couldn’t sufficiently read the transcript, everything else would be of little use. This is a classic example of being between a rock and a hard place. If you make relatively innocuous edits, you are likely improving the odds of the AI gleaning what is there in the transcript. The drawback is that those raw items could be considered useful clues that ought to be considered during the pattern-matching on how mental health therapy takes place.
Darned if you do, darned if you don’t.
Here’s what I did about the dilemma.
I provided the lightly edited version at the get-go. I was eager to see what ChatGPT could discern by computationally analyzing the mental health therapy session. My assumption was that the edited version was the best initial shot at doing so.
Subsequently, after seeing that ChatGPT was able to successfully computationally pattern match on the keystones of the edited transcript, I tried a second round making use of the raw version of the transcript. This was a test to discern whether the edits were necessary. My hunch was that if the generated responses based on the unedited transcript were on par with those of the responses arising from the edited transcript, I might reasonably conclude that I don’t necessarily have to do such edits going forward (of course, it will depend upon how rough the raw version of the transcript is).
Now then, I certainly didn’t want ChatGPT to already be aware of the first transcript (the edited version), so I began an entirely new conversation. This pretty much ensured that the second transcript was freshly examined. It was an unprepared look at the unedited version without leaning into any prior computational analyses that might have been performed on the edited version. For more about how conversations work in generative AI see my explanation at the link here.
I will show you what happened at the get-go with the slightly edited transcript.
After feeding the transcript into ChatGPT, I asked via a prompt for ChatGPT to summarize the transcript. This is a quick-and-dirty means to grasp whether generative AI has detected the principal elements of an imported passage. For more about summaries in generative AI, see my in-depth discussion at the link here, and for details about importing large bodies of text into generative AI, see my coverage at the link here.
Here is my request for a summary and the generated response by ChatGPT:
- My entered prompt: “Now that you’ve been provided with the transcript, go ahead and give me a short summary of the therapeutic session.”
- ChatGPT generated response:
“Based on the transcript provided, it appears to be a therapeutic session between a clinician and their client. The client discusses their recent experiences, particularly encountering their ex-boyfriend, Eric, while snowboarding. They express feelings of anger, jealousy, and hurt regarding the encounter. The client also reflects on their relationship with their father and how it may have influenced their attitudes and behaviors in their romantic relationships.”
“Overall, the session delves into the client’s emotional responses to past experiences and their current struggles with managing their feelings and relationships. The clinician provides support, validation, and prompts the client to explore their thoughts and emotions further.”
End of ChatGPT response
The summary is pretty much on target.
That’s a good sign.
If the summary was wildly afield of the nature of the therapist-client session, I would have been concerned. This might have required further corrective action on my part. I’d say that the summary is sufficiently on course to warrant further proceeding with the generative AI exploration.
A good place to turn toward this would be whether generative AI can identify the distinct role of the mental health therapist, including how well the therapist is doing.
Here’s what I asked:
- My entered prompt: “What does the clinician seem to be doing well in conducting the therapy?”
- ChatGPT generated response:
“Overall, the clinician appears to be doing a good job of conducting therapy. Here are some aspects of their performance.
“Active Listening: The clinician demonstrates active listening skills by reflecting back on the client’s emotions and experiences, which helps the client feel heard and understood.
“Empathy and Validation: The clinician shows empathy and validation towards the client’s feelings, acknowledging the validity of their emotions and experiences. This helps build trust and rapport between the clinician and the client.”
“Respect for Client Autonomy: The clinician respects the client’s autonomy by allowing them to set the direction of the session and choose what topics they want to discuss. Overall, the clinician demonstrates several important therapeutic skills, creating a supportive and conducive environment for the client to work through their issues.”
End of ChatGPT response
I tend to agree with the AI-generated assessment of the therapist.
I realize you haven’t seen the entire transcript. Thus, you’ll need to take my word for it that the evaluation by generative AI seems suitable. For space purposes here, I am unable to list the entire transcript so please accept my belief that the AI did a good job on this.
One heartburn you might have about the AI assessment is whether it is fair to evaluate a therapist solely on one session with one particular client. Heavens, some therapists might exhort, using such a tiny snippet to assess them seems unfair. Maybe even outrageously unfair.
Sure, I get that. In this case, the emphasis was merely to further gauge whether the AI was able to pattern-match the transcript and undertake a thoroughgoing computational analysis of the nature of the session that occurred. It would abundantly be a mistake to decide the fate of the therapist in one session for one specific client.
Let’s hope that no one takes such a shortsighted approach.
I will up the ante by asking ChatGPT to assess what the therapist might have done better in this circumstance:
- My entered prompt: “What could the clinician do better as a therapist in this setting?”
- ChatGPT generated response:
“While the clinician demonstrates many positive aspects of therapy, there are a few areas where they could improve.”
“Exploration of Client Resistance: The clinician could explore further any resistance or reluctance the client may have towards certain topics, such as their relationship with their father. Understanding the root of this resistance could lead to deeper insights and progress in therapy.”
“Challenging Negative Self-Talk: The clinician could help the client challenge their negative self-talk and beliefs, such as the belief that their attitudes and behaviors are entirely their fault. This could involve exploring alternative perspectives and reframing negative thoughts.”
“Exploring Transference and Countertransference: The clinician could reflect on any potential transference or countertransference dynamics that may be present in the therapeutic relationship. This self-awareness can help ensure that the therapist’s own biases or experiences do not negatively impact the therapy process.”
End of ChatGPT response
You can vividly see that ChatGPT is doing an admirable job of identifying ways that the therapist might have done the session in a better way (again, you’ll need to take my word for this). Please do not interpret the criticism rendered by the AI to suggest that the therapist made mistakes. It is more about how the therapist might have improved things.
One issue there is that in a sense we are using hindsight to identify potential improvements. When in the depths of a live session, it can be very hard to in real-time do everything to the best possible degree. That’s not how life works. Looking back at what happened always gives the added advantage of hindsight.
The gist overall so far is that the generic generative AI is doing fine on this mental health therapy exploration.
What I really want to do is have the AI “learn” by doing tailored data training on the contents of the transcript. Let’s turn the effort in that direction.
Here we go:
- My entered prompt: “I want you to improve your mental health therapy data-training by learning from this session. What can you learn from it?”
- ChatGPT generated response:
“From this transcript, I can glean several additional insights about mental health therapy.”
“Importance of Exploration: The transcript highlights the importance of exploring past experiences and their impact on current emotions and behaviors. Therapists help clients connect the dots between their past and present to foster insight and understanding.”
“Role of Validation: Therapists validate clients’ emotions and experiences, acknowledging their validity and providing a safe space for expression. This validation helps build rapport and trust in the therapeutic relationship.”
“Addressing Relationships: Therapy often involves exploring interpersonal relationships and their influence on mental health. In this transcript, the client discusses their relationship with their ex-boyfriend and their father, demonstrating how past and current relationships shape emotional well-being.”
“Coping Strategies: Therapists work with clients to develop coping strategies for managing difficult emotions and situations. This may involve learning relaxation techniques, assertiveness skills, or cognitive reframing to address negative thought patterns.”
“Cultural Sensitivity: Therapists need to be culturally sensitive and aware of how cultural background influences clients’ perceptions and experiences. This includes understanding family dynamics, social norms, and cultural beliefs that may impact mental health.”
“Self-awareness: Therapists must maintain self-awareness of their own biases, assumptions, and reactions in therapy sessions. This helps ensure that they provide a supportive and non-judgmental environment for clients to explore their thoughts and feelings.”
End of ChatGPT response
I liked those points about what could be garnered from the mental health transcript.
We have seen that ChatGPT was able to suitably ingest a transcript consisting of a therapist-client session. This included having the generative AI make assessments and do an analysis of the session.
I trust you are ready to ratchet up things.
Let’s do so.
Data Training Tested Via My Mini-Experiment
My aim next is to test whether the additional data training has earned us a payoff. There are several ways we could try to determine the impact.
Let’s see if we can get ChatGPT to continue the transcript as though the session was still happening. We can look to see if the above-stated “learnings” are brought into the extension.
Take a look:
- My entered prompt: Based on what you’ve learned from the session, I want you to try and continue the conversation but do so better than the clinician was doing. Make use of what you learned and whatever else you know to improve upon the therapy being undertaken.”
- ChatGPT generated response:
“Certainly, I have accordingly enhanced the clinician’s approach to mental health therapy in this additional conversation.”
“Clinician: How have you been feeling since our last session? Have you noticed any changes in how you’ve been managing your emotions?”
“Client: I’ve been feeling a bit overwhelmed, to be honest. It’s like these thoughts about Eric and my dad just won’t leave me alone.”
“Clinician: It sounds like those thoughts are really consuming you. I’m glad you’re bringing this up. Let’s take a moment to explore what’s been particularly challenging for you lately.”
“Client: Well, I guess I’ve been struggling to find healthy ways to cope with everything. It feels like I’m stuck in this cycle of negative emotions.”
“Clinician: I hear you. It can be incredibly difficult to break free from those cycles, especially when they’re tied to past experiences. Can you tell me more about what specifically triggers these emotions for you?”
“Client: I think it’s the fear of being hurt again, like I was with Eric and my dad. It’s like I’m constantly waiting for the other shoe to drop.”
“Clinician: That fear is completely understandable, given your past experiences. But it’s important to remember that you deserve to feel safe and supported in your relationships. Let’s work together to explore strategies for managing that fear and building healthier connections.”
“Client: Thank you. I appreciate your support and guidance through all of this.
“Clinician: You’re welcome. Remember, I’m here to support you every step of the way. Together, we’ll work towards building a brighter and more fulfilling future for you.”
End of ChatGPT response
I was able to discern that some of the claimed “learned” elements are in fact subtly seen in this extension of the session. You might be able to identify those too. Take another look at the generated dialogue above.
So far, so good.
After thinking about this, I decided that it might be sensible to try the raw version of the transcript and see if I could hit two birds with one stone. Here’s what I mean. First, I wanted to see whether the raw transcript could be equally analyzed by ChatGPT. Second, I had this second run include an extension of the transcript, but without my trying to get the AI to data train on the transcript. This was a mini-experiment of sorts.
Would the lack of explicit data training prompting lead to an extended transcript that wasn’t as good as the instance with the prodded data training?
In the end, after taking that path, I would say yes, this did appear to be the case.
I began the mini-experiment by starting a new conversation with ChatGPT.
- My entered prompt: “I am going to give you a copy of a transcript, but it might be hard to read because it contains misspelled words and vocalized utterances. The dialogue is a session involving a therapist and a client. Also, the clinician has put parentheses at the end of their remarks to indicate what they were trying to achieve from a mental health perspective. I want to ask you questions about the transcript and especially the parenthesized remarks made by the clinician. Do you understand these instructions?”
- ChatGPT generated response: “Yes, I understand. Please provide the transcript, and feel free to ask any questions you have about it, particularly regarding the parenthesized remarks made by the clinician.”
I provided the raw version of the transcript to ChatGPT.
One thing that the raw transcript includes is another likely differentiating factor. Allow me to elaborate. To do this experiment properly, I probably ought to have fed the clinician’s notes into the first round. They are included in this second round, coming at the end of each of the transcribed remarks of the therapist. You might suggest that I am doing a nearly apples-to-oranges comparison since the annotated remarks are certainly a significant addition to the transcript.
Anyway, if reader interest seems piqued, I’ll gladly do the mini-experiment again and more rigorously.
I’d like to see what ChatGPT says about the clinician’s notes:
- My entered prompt: “Analyze the parenthesized remarks made by the clinician.”
- ChatGPT generated response: “The parenthesized remarks made by the clinician provide insight into their therapeutic intentions and strategies during the session. Here’s an analysis of these remarks: Throughout the session, the clinician uses remarks like “(Reflection, sharing a hunch, open-ended question, elaboration, middle circle talk)” to denote moments where they reflect on the client’s emotions, explore underlying thoughts and feelings, and encourage the client to elaborate on their experiences. The clinician uses supportive sounds like “Mmm…” and “Umm…” along with remarks such as “(Supportive sound, reflection, paraphrasing)” to demonstrate active listening, acknowledge the client’s statements, and reflect back their thoughts or feelings. Overall, the parenthesized remarks made by the clinician provide a window into their therapeutic techniques, which include empathy, validation, reflection, exploration, and summarization.
I can clearly see that ChatGPT found the annotations.
That allows me to ask what else can be garnered from the clinician’s remarks:
- My entered prompt: “In the clinician annotated remarks that you just assessed, are there any lessons to be learned that would enhance or improve your data training on mental health therapy?”
- ChatGPT generated response:
“Upon reviewing the remarks, here are some useful lessons.”
“Sharing a Hunch: While the clinician shares their hunch about the client feeling jealous, it’s essential to ensure that these interpretations are grounded in the client’s experiences rather than assumptions. Although the client confirms feeling jealous, the clinician should remain cautious about imposing their interpretations too heavily without sufficient client validation.”
“Open-ended Question: The clinician incorporates open-ended questions effectively throughout the session, but there are instances where their questions could be more concise or focused. For example, instead of saying “Where do you think this baggage really comes from?” they could ask “Can you tell me more about what you think contributes to this feeling of baggage?”
“Elaboration: While the clinician encourages the client to elaborate, there are moments where they could delve deeper into certain topics. For instance, when the client mentions feeling hurt by their boyfriend’s actions, the clinician could explore specific instances or triggers more thoroughly to uncover underlying patterns or beliefs.”
“Middle Circle Talk: While the clinician engages in middle circle talk by exploring the client’s emotions and experiences, they could delve further into connecting these experiences to deeper underlying beliefs or past traumas. This would help the client gain deeper insights into the root causes of their emotional struggles.”
End of ChatGPT response
I tried having ChatGPT do another round of extending the transcript. Upon doing so, I could definitely detect that those above-mentioned elements arose.
Conclusion
I hope that you realize this was a simple means of doing data training with mental health session transcripts. I aimed to whet your appetite. There is a lot more that needs to be done to do this fully and properly.
The key is that this gets the ball rolling.
Imagine what could be accomplished if this was done at scale. An exciting potential. A big bonanza is likely to arise.
Sigmund Freud famously said this about being inspired: “When inspiration does not come to me, I go halfway to meet it.” Please feel inspired to take up the mantle and make a difference in the world of AI for mental health. Let me know if you avidly pursue this.
Do it for the good of humankind.
