AI in the World of Mental Health
Preface:
The author of this article is Lisa Sojourner. I believe this is important to address first and foremost. Although many of us at Grace and Nature have similar opinions on AI, we are all unique individuals and opinions may vary and shift over time. But generally speaking many of us feel somewhere around where this article lands. Buckle up my friends, this is a long article! But I think you deserve a detailed and nuanced understanding of where we generally stand.
In this article I am going to address four topics that need transparency.
As the world continues to adjust to and understand the impact of AI, we at Grace & Nature wanted to talk about the use of AI in our work as mental health therapists. On this page I want to talk about three things, so anyone reading our materials or thinking about contacting us for help are informed as to where we stand.
In this Article:
Our Understanding of AI
Why We Don't Use AI in Content
Why We Don't Use AI in the Therapy
Where AI has been Utilized
Our Understanding of AI
What AI Is Not
Firstly, the term “AI” is a misnomer. The term ‘intelligence’ supposes that this is something that has intellect. However, intellect is a trait confined solely to human beings. It is a faculty of the soul. Computers and code, no matter how complex and powerful, do not have souls. Our intellect allows us to use judgement, reason, and self-consciousness. It is our ability to perceive the divine, the true, the good and the beautiful. These are abilities gifted to humanity and we cannot abdicate them or imbue inanimate objects with them. Only God has the ability to create in his own image.
What AI Is
AI is a large language model. It breaks texts into pieces or "tokens" and then converts them to numbers, processing them through layer after layer of mathematical networks to predict the next word or string of text. It is a computer system that is “trained on massive amounts of text data to understand and generate human-like language.” (Stanford: What is a Large Language Model) And here is the important distinction in the way AI is discussed and why there is so much confusion surrounding it. Individuals will discuss AI as though is has properties only human beings can possess:
It has intelligence.
It can learn.
It is neural networks.
It can understand.
In reality it is a computer system that, as Stanford has said, generates human-like language. So all of these phrases make much more sense when we re-write them in this reality:
‘It has intelligence’ becomes: It mathematically predicts human language: Meaning it takes human intelligence that we have given the program access to and can mimic language similar to it. But it requires massive amounts of human created intelligence to be able to do this.
‘It can learn’ becomes ‘It uses math to predict relationships’: Large Language Models like GPT-4 are “encoding mathematical relationships between words (actually, tokens), with the aim to make predictions about what text goes with what other text.” (AI Doesn’t Really Learn). I.e. it’s not learning, it’s using math that human beings discovered and encoded and using data sets that are so large it is beyond what any one human could ever read or digest in a lifetime.
‘It has neural networks’ becomes ‘simulated neural networks’: This is directly from IBM: “Neural networks, also called artificial neural networks or simulated neural networks, are a subset of machine learning and are the backbone of deep learning algorithms. They are called “neural” because they mimic how neurons in the brain signal one another.” (AI vs. machine learning vs. deep learning vs. neural networks: What’s the difference?) Again, these are programs working at massive scale that mimic human functions.
‘It can understand’ becomes ‘It can mimic understanding’: Precision of language is important here, what do we mean by understand? The etymology of understand comes from the Old English verb understō̆nden. There are multiple definitional senses for this but my favorites come from the Middle English Compendium: “(a) To exercise the intellectual power;” and “(e) to grasp intellectually”. Which brings us back to intellect, a uniquely human trait. AI mimics intellect by predicting the next most likely portion of a sequence. It’s pretty convincing, but intellect it is not.
Beware of anthropomorphisms:
To anthropomorphize is to attribute human characteristics to non-human entities.
So to say “My phone doesn’t want to work right now.” is to give human characteristics to your phone, your phone doesn’t ‘want’ anything.
Whenever we see articles or pundits talking about AI as though it has intelligence, they are ascribing human features to code that we are actually responsible for. It is a slippery slope, it can go quickly from “AI told me to set my oven to 800°.” to “AI told me it’s ok to cheat on my taxes.” AI hasn’t told you anything, AI doesn’t have an intellect it can use to make prudent or moral decisions.
AI processed a massive data set and made statistical inferences of what the next string of words would be in response to your questions. It is up to us to decide if it is prudent to set the oven to 800° or if it is moral to cheat on taxes.
The below is a portion of a visual representation of a large language model. When you see it from this perspective it becomes more clear that this code does not have the capacity to discern truth, beauty, and goodness. But this is the way of ideologies, how we talk about something determines how we think about it.
Why We Don't Use AI in Content Generation
At Grace and Nature we do not use AI to generate our informational content. We believe that human to human contact is deeply important. Whether you are in our offices face to face with us or you are reading one of our articles or posts, we want you to trust that you are encountering another person, not an algorithm.
Here, you don’t have to ask ‘Is this AI?’:
Prior to becoming a therapist I worked off and on as a graphic designer and a web designer. I designed and continue to maintain our website without assistance (and it’s one of my favorite behind-the-scenes things to do in group practice). Even for me it’s becoming more and more difficult to spot when something has been generated by a program. I know I’ve believed something to be real online in the past and I’m pretty sure it’s going to happen again, it’s just getting way too good at mimicking real photos, language, and art.
Here, in our space, we want you to trust that the information provided by us comes from our expertise, expertise earned from consistent study and years of experience. We don’t want you to be reading something we have written or created and wonder “yeah, but did AI make this?” Here you are going to encounter only human, therapist-generated content. If it’s about human mental health, a human wrote it!
Why We Don't Use AI in the Therapeutic Pross
We also want you to trust that at any point in the therapeutic process, you are encountering and being provided real human thought, ingenuity, and intellect. That means you will not find AI being used in any of the following levels of clinical care:
Client Communication: At Grace & Nature we do not use AI to communicate with you. We do not have chat bot features turned on. We do not have automatic AI responses. Our therapists do not use AI to write emails back to you. You will be speaking to a real human being, either one of our clinical team or to our admin support team members. Real human interactions not algorithmic responses.
Case Conceptualization: Case conceptualization is where we work out our understanding of symptoms, a person’s history, environmental factors, strengths, etc. As Grace & Nature therapists we have all agreed that we do not use AI to help us think through, conceptualize, or understand what is going on for you. We learned to do this in graduate school, from trained therapists, and we consult with other human therapists.
Diagnosis: Diagnosis differs from case conceptualization. With diagnosis we are giving a name to what we call a ‘symptom cluster’. If case conceptualization explains the how and why, diagnosis tells us the what. We do not use AI models to diagnose. Again this was a skill learned in graduate school and refined through supervision, consultation, and years of experience. No data model can take into account the real human factors of the infinitely unique individual sitting across from us.
Treatment Planning: In treatment planning we take our understanding of your case and potential diagnosis and use theory to develop a plan. This plan includes things like short, mid, and long term goals, what interventions a therapist will use to work towards those goals, a timeline for therapy, and how you and the therapist will know that you no longer need therapy. Our therapists do not use AI at any point of this process.
Clinical Decision Making: Throughout the counseling process, therapists need to make in-real-time clinical decisions. This can be things like “Can this client benefit from telehealth?” “Do I need to contact this client’s emergency contact?” “Will this potential intervention cause harm to my client?” We all have moments where we are not sure what to do and as therapists we are not immune to this. But in graduate school we are taught to “Consult! Consult! Consult!” (thank you Dr. Bertram for this evergreen reminder). When we make clinical decisions and we need assistance, we consult with other professionals.
Therapeutic Judgment: Therapeutic Judgment differs from clinical decision making. In decision making we are having to decide on real world actions, generally about safety. Therapeutic judgement is typically taking place in case conceptualization. “Does this client need me to focus more on their overwhelming work stress or on the challenges in their relationship?” “Today am I going to be more directive with this client or am I going to give them space to voice the fear they have been experiencing.” This is one of the most individual and deeply human aspects of being a therapist, and it is often where you will feel most seen, heard, and understood. AI cannot replace this.
Risk and Crisis Assessment/Intervention: Risk assessment happens when a client, minor, individual with a disability, the elderly, or vulnerable individual may be at risk. This could be anything from a client experiencing suicidal ideation to a therapist becoming aware that a minor is being neglected or abused. We are trained in school and beyond how to respond in an empathic and effective manner. At Grace & Nature we rely on our training and consultation with other professionals, not AI, to respond in prudent and immediate ways.
Note Taking / Note Generation: After we have done all of the above we write our notes. More and more therapy note systems are including the ability to have AI automatically transcribe sessions, listening in as you speak with your therapist. Other systems offer the ability for therapists to write a few sentences and the system will write the progress note for the therapist. Grace & Nature therapists do not utilize either of these integrations of AI into the therapy space. There are multiple places where a therapist is using their God-given intellect to conceptualize and understand your unique story. And note taking is one of the places we process, synthesize, and plan ahead for you. At Grace & Nature, we do not offload this intellectual work to a mathematical algorithm.
Where AI Has Been Utilized with Grace & Nature
There are a few places you may see AI utilized and in the spirit of transparency we want to let you know where you may see this.
Image editing (not generation): We will not be using AI to create entire images. There is too much human creativity in the arts, both traditional and photography. We prioritize using those images in our content. However, there may be times we use the AI features in Canva to clean up things like headshots.
Unintentionally: As I said above, there have been times I myself have been tricked into thinking what I am looking at is human generated. I’ve noticed when I’m looking on Pinterest, shopping on Etsy, and even listening to songs on Spotify. There may be times when an image or a link may be AI. We’re human too and if we discover something is AI generated or a tool had integrated we were unaware of, we will amend it as soon as possible.
Comments