By: Alixandra Wilens, MA and Brian Zaboski, PhD, ABPP
From Isaac Asimov’s rules of robotics to Arnold Schwarzenegger’s Terminator, humankind has long been captivated—and sometimes unnerved—by the idea of machines that can think like humans. Today, with tools like ChatGPT and DALL-E, this fascination has evolved into practical questions about how artificial intelligence (AI) can be applied in fields like mental health, and specifically in the treatment of and research on OCD.
Within the context of OCD practice and research, the AI Special Interest Group (SIG) intends to address these questions and more. We hope to increase AI knowledge among individuals with OCD, researchers, and clinicians. We intend to engage in discussions of clinical applications, brainstorm research projects, and consider the balance between AI’s risks and rewards. Starting with this brief introduction, our goal is to contribute to the conversation of AI and mental health. But first, we need to introduce and demystify AI.
Artificial intelligence (AI) is broadly defined as the ability of machines or computer programs to perform tasks that usually require human intelligence. This can include self-driving cars, fraud detection systems, speech recognition, and language translation. Machine learning is a subset of AI used to process and predict outcomes from data. It traditionally requires researchers to manually define data features. For instance, in facial recognition, a researcher would have to program the system to recognize the key components (or features) of a face. In contrast, deep learning, which has gained more favor in recent years, automates this process by allowing the system to discover these features on its own through multiple layers of computation.
Think of deep learning as a series of filters through which data passes. As the data move through each layer of the network, it gets more refined—similar to how our brains process visual information. If a network is learning to recognize cats, the first layer might detect basic shapes. The next might recognize textures like fur. Eventually, the network can distinguish a cat from other animals.
Using mathematics, AI as it exists today can perform incredible tasks, though these tasks tend to be narrow and within defined parameters. Some AI systems use mathematics no more complicated than what students learn in high school–like linear algebra and calculus–by learning from huge amounts of data. For example, an AI system designed for image recognition using a simple neural network can be trained on large datasets of labeled images to distinguish between cats and dogs. Despite the simplicity of this network, it can achieve high accuracy by learning from vast amounts of data. In mental health, AI’s ability to sift through large datasets is invaluable. For example, machine learning models can analyze medical records to predict a patient’s likelihood of developing depression, offering clinicians data-driven insights that help improve treatment plans. While these systems are promising, they still operate within narrow tasks, meaning their use must be carefully considered. While there are efforts to create general forms of artificial intelligence that can generalize its training from a previous task (like how to play games) to accomplish something it was never trained to do (to write a poem, for example), the possibility of this kind of general artificial intelligence is hotly debated.
AI has captured our collective imagination because it equips us to achieve goals we would be hard-pressed to attain on our own. It is a tool with great promise that must be studied, tested, and used responsibly. This starts with open dialogue. We invite you to join us in this exploration of AI’s role in OCD treatment and research. What questions or concerns do you have about AI and mental health? Share your thoughts and let’s explore the possibilities together.
Future blog posts will examine more complicated models, as well as exploring the ethical consequences of using these models within the world of OCD. We will examine where we get the input data that fuels these models and who owns that data. We will discuss how large language models work and the ways clinicians and researchers might use them in practice. We will also explore the mistakes these systems make and how to fool them.
If you are interested in joining the Artificial Intelligence SIG, please complete this interest form to receive meeting information and updates. More information about Special Interest Groups is available here.
Some Starting Points for your AI journey:
Most helpful to everyone:
“Artificial Intelligence and Machine Learning” from the APA
“The Ethics of ChatGPT – Exploring the Ethical Issues of an Emerging Technology” by Bernd Carsten Stahl and Damian Eke
“But what is a neural network? | Chapter 1, Deep learning” by 3Blue1Brown
Most helpful to clinicians:
“What psychologists need to know about the evolution of generative AI” by Zara Abrams
Most helpful to researchers:
“Choosing Prediction Over Explanation in Psychology: Lessons From Machine Learning” by Tal Yarkoni and Jacob Westfall
Leave a Reply