The now and future of ChatGPT and GPT in psychiatry

Abstract

ChatGPT has sparked extensive discussions within the healthcare community since its November 2022 release. However, potential applications in the field of psychiatry have received limited attention. Deep learning has proven beneficial to psychiatry, and GPT is a powerful deep learning-based language model with immense potential for this field. Despite the convenience of ChatGPT, this advanced chatbot currently has limited practical applications in psychiatry. It may be used to support psychiatrists in routine tasks such as completing medical records, facilitating communications between clinicians and with patients, polishing academic writings and presentations, and programming and performing analyses for research. The current training and application of ChatGPT require using appropriate prompts to maximize appropriate outputs and minimize deleterious inaccuracies and phantom errors. Moreover, future GPT advances that incorporate empathy, emotion recognition, personality assessment, and detection of mental health warning signs are essential for its effective integration into psychiatric care. In the near future, developing a fully-automated psychotherapy system trained for expert communication (such as psychotherapy verbatim) is conceivable by building on foundational GPT technology. This dream system should integrate practical 'real world' inputs and friendly AI user and patient interfaces via clinically validated algorithms, voice comprehension/generation modules, and emotion discrimination algorithms based on facial expressions and physiological inputs from wearable devices. In addition to the technology challenges, we believe it is critical to establish generally accepted ethical standards for applying ChatGPT-related tools in all mental healthcare environments, including telemedicine and academic/training settings.

Keywords: ChatGPT; GPT; artificial intelligence; deep learning; informatics and telecommunications in psychiatry.

Fig. 1

Simplified structure of the transformer. The encoder (green box) receives inputs for processing through 6 identical layers (red box) into a sequence of continuous representations. Each layer has a sublayer of a multi‐head self‐attention and a sublayer of a feed forward network. The decoder (orange box) receives and processes these representations through another 6 identical layers (purple box) into outputs. The layers in the decoder are similar to the ones in the encoder but have an additional masked multi‐head self‐attention sublayer to receive encoder‐generated representations. This sublayer grants the model auto‐regressive property: the model only depends on prior words in a sentence to predict words at a specific position. Positional information of words are encoded and passed separately in the model (not shown in the figure). The GPT model only utilized the decoder structure of the Transformer (Transformer‐decoder‐only structure).

Fig. 2

Simplified training flowchart for GPT. (a) Building the Model: Engineers in OpenAI built the basic structure of GPT called the Transformer, and set the hyperparameters (the number of layers and parameters in the Transformer), which cannot be changed by the model itself later after trained with data. (b) Pre‐Training: The model was put into the unlabeled, unsupervised pre‐training with huge amounts of data, where the model learned the patterns and presentations in languages. The learned knowledge was stored as data generated by the model, or “weights,” which could be changed after further training. (c) Fine‐Tuning: The pre‐trained model was then fine‐tuned for natural language processing tasks. Novel techniques other than fine‐tuning were also employed to acquire better performance. In the process, weights in the model were altered to better match specific tasks. The end products were the core builds of GPT called GPT‐1 to 4. Different core builds varied as their hyperparameters and the quantity of training data differed. (d) Further Training: These core builds can be further trained into even more specialized models like the chatbot ChatGPT.