{"doi":"10.1111/add.16341","title":"The ChatGPT therapist will see you now: Navigating generative artificial intelligence's potential in addiction medicine research and patient care","abstract":"Generative AI offers potential for enhancing addiction medicine research and practice by analyzing medical literature, improving research efficiency, streamlining clinical workflows, and even providing counseling support. Addressing challenges such as confabulation, biases, and patient acceptance and adoption is crucial for responsible integration and to improve care in substance use disorders. Artificial intelligence (AI), primarily in decision support and predictive modeling, has increasingly been used in medicine, including the field of addiction [1]. With the widespread release of ChatGPT, there is now active interest and need for rigorous evaluation of the potential for generative AI to enhance addiction medicine research and practice. ChatGPT is a chatbot powered by an underlying large language model (LLM), named Generative Pre-trained Transformer (GPT), which can simulate human conversation. LLMs are trained on extensive text datasets to predict the next word in a given sequence [2]. As LLMs have grown, they have demonstrated emergent properties including question answering, summarization and even reasoning. Applying these models to tasks requiring medical knowledge has yielded impressive outcomes, with GPT-4, the model behind ChatGPT+, correctly answering 90% of presented United States Medical Licensing Exam questions [3]. Generative AI holds the potential to transform medical research and one notable application is using LLMs' ability to analyze vast amounts of literature to identify prior work that informs current research. However, researchers using LLMs for this purpose still need a deep understanding of the subject matter as LLMs are prone to convincingly present incorrect information, also known as confabulation. Elicit.org is an example of a tool that uses LLMs to automate literature review by producing a list of relevant literature, with a concise summary, in response to a user's question. Notably LLMs used in this tool have to be constrained using additional AI tools to limit confabulated information in the results [4]. At their core, LLMs are transformer models that seek to understand relationships between data. Fouladvand et al. [5, 6] used a transformer model to predict opioid use disorder from multiple data sources, relying on the transformer model to extract associations within and between data sources. Notably, however, this model did not include unstructured text based data, which remains challenging to work with. Our group is working on developing machine learning models to predict retention in treatment for patients with opioid use disorder with the goal of determining if unstructured electronic health record (EHR) data, unlocked with LLMs and combined with structured EHR data, will improve model performance [7]. Similar techniques of combining LLMs with structured data have already been deployed to more efficiently identify clinical trial cohorts [8]. These techniques of using LLMs alongside, and not in replacement of, more traditional approaches may limit the risk of confabulation. The text generation capabilities of LLMs offer promising possibilities in academic publishing. Researchers have hypothesized that LLMs could contribute to everything from automated manuscript editing to composing complete manuscripts. Nevertheless, without addressing their shortfalls, LLMs may not be trusted to perform such tasks. The limitations of generative AI were central to a debate that ensued after ChatGPT was listed as an author on a preprint manuscript uploaded to medRxiv in December 2022 [9]. Some argued that the term “author” was misleading because ChatGPT could not take responsibility for the works' validity. Although tools like ChatGPT will likely improve in accuracy overtime, the responsibility for ensuring that accuracy ultimately rests with humans. To harness the full potential of generative AI in medical research, it is crucial that we develop robust guidelines and validation processes to ensure the respo","journal":"Addiction","year":2023,"id":341654,"datarank":0.0,"base_score":0.0,"endowment":0.0,"self_citation_contribution":0.0,"citation_network_contribution":0.0,"self_endowment_contribution":0.0,"citer_contribution":0.0,"corpus_percentile":null,"corpus_rank":null,"citation_count":15,"citer_count":0,"citers_with_citation_signal":0,"citers_with_endowment":0,"datacite_reuse_total":0,"is_dataset":false,"is_dataset_confidence":0.9548,"is_data_producer":false,"deposit_databanks":null,"is_oa":true,"file_count":0,"downloads":0,"has_version_chain":false,"published_date":"2023-01-01","fair_score":null,"fair_percentile":null,"algorithm_id":"datarank_citation_only_1hop_v6","ranking_scope":"data_only","authors":[{"id":513828,"name":"Sajjad Fouladvand","orcid":"0000-0002-9869-1836","position":1,"is_corresponding":false},{"id":227523,"name":"Jonathan H. Chen","orcid":"0000-0002-4387-8740","position":2,"is_corresponding":false},{"id":1077640,"name":"Chwen‐Yuen Angie Chen","orcid":"0000-0002-7207-598X","position":3,"is_corresponding":false},{"id":1077639,"name":"Steven Tate","orcid":"0000-0001-9544-594X","position":0,"is_corresponding":true}],"reference_count":12,"raw_metadata":null,"created_at":"2026-07-19T01:11:08.077724Z","pmid":"37735091","pmcid":null,"fwci":null,"citation_percentile":null,"influential_citations":0,"oa_status":null,"license":null,"views":0,"total_file_size_bytes":0,"version_count":0,"fair_f":null,"fair_a":null,"fair_i":null,"fair_r":null,"fair_zscore":null,"fair_rationale":null,"fair_model":null,"fair_agent_version":null,"fair_fulltext_source":null,"fair_has_llm":null,"fair_computed_at":null,"clinical_trials":[],"software_tools":[],"db_accessions":[],"linked_datasets":[],"topics":[]}