Ethical Considerations in the Use of Generative Artificial Intelligence in Psychological Practice

Technology continues to develop at a dizzying pace. Across the last 20 years in particular, use of technology has led to improved efficiency and reduced errors in psychological practice, from statistical analysis programs to scoring programs for assessment measures. Integration of Generative Artificial Intelligence (Generative AI) into daily practice likely represents a natural evolution of this process.

Generative AI is an artificial intelligence tool that can create new and unique content by understanding the underlying patterns and structures of data/information. It goes beyond the pattern recognition and predictive capabilities of conventional AI systems by using machine learning algorithms to produce new content based on the existing data that the system has been exposed to. Of particular importance, conventional AI most commonly relies upon supervised learning and discriminative models, whereas generative AI is structured to allow unsupervised learning based upon generative models.

Potential Opportunities

There are potentially a multitude of benefits for harnessing the power of Generative AI in the practice of psychology. From generation of scientific hypotheses to identifying complex data patterns could catapult psychological research forward. Educators could use Generative AI to pull together a course curriculum and even create lectures on a particular topic. From the hospital to the clinic, Generative AI could speed up the clinical processes by using it for writing clinical notes or for case formulation by identifying possible diagnoses given a set of symptoms. There are enumerable possibilities for streamlining processes and improving efficiency of psychological practice.

Potential Limitations

However, Generative AI does not necessarily get the job done. It is essentially an unconscious creator that excels at repackaging already produced materials based upon the parameters it is provided but lacks the ability to truly comprehend context or concepts. Generative AI can only generate content based upon the information it was trained on and therefore, if the data is biased, incomplete, or inaccurate, then the conclusions drawn may be inappropriate, biased, nonsensical, and/or inaccurate. The data may quickly become out-of-date if the system has not been re-trained with real-time updates, thus the output may no longer be accurate or relevant. With this “unconscious” unsupervised learning, Generative AI does not “know” the laws that govern information access, sharing, and disclosure. It may generate exact copies of copyrighted materials presenting it as if it is new material. If exposed to private health information, it may disclose that information to another user of that Generative AI platform leading to a breach of privacy and confidentiality. These limitations potentially make using Generative AI for psychological practice a field of landmines.

Ethical Code as Applied to Using Generative AI in the Practice of Psychology

It is incredibly important to recognize the above noted limitations and to be mindful consumers when choosing to use Generative AI technologies in the practice of psychology. Psychologists must ensure that their use adheres to laws and ethical standards. However, with the ethical principles and code (APA, 2017) in the forefront of our minds, we may be able to capitalize on this new technology in our daily practice.

 Avoiding Harm (Standard 3.04) is always at the forefront, as are related issues that will be part and parcel of figuring out where Generative AI fits into our work as psychologists. We are ultimately responsible for the “products” of our work and that those “products” contain accurate, reliable, relevant, unbiased information. The psychologist must be “the conscious oversight” for controlling and fine-tuning the generated output to ensure that the information does not harm. If a psychologist chooses to use Generative AI, the onus is on that psychologist to verify the veracity of the information generated.

 Boundaries of Competence and Maintaining Competence (Standards 2.01 and 2.03) refers to our professional knowledge base which we gained through our education, training, supervised experience, consultation, study, or professional experience. We must rely upon those areas of competence to do our work. The Bases for Scientific and Professional Judgments (Standard 2.04) is based upon established scientific and professional knowledge of psychology and Generative AI is not necessarily trained on that knowledge. How do we fact check the information generated? What do we do if we cannot verify a point and conclusion made?

 Informed Consent, Maintaining Confidentiality, and Discussing the Limits of Confidentiality (Standards 3.10, 4.01 and 4.02) will need to be considered when integrating Generative AI into practice. There are potential issues for Maintaining Confidentiality of private or protected information. There are considerations of how and under what circumstances psychologists need to seek Informed Consent before using Generative AI. How will private and confidential information be protected if Generative AI is used to analyze data, search for patterns in the information, and/or to generate a report given the information for a patient, client, or research subject? As psychologists who employ this emerging technology, we must monitor and work to decrease harmful or potentially harmful interactions that may come from our use of this technology.

Delegation of Work to Others (Standard 2.05) typically has focused on how we employ others to assist with psychological work. However, with the potential “roles” that Generative AI can play in the practice of psychology, the delegation of work to this technology as an aid, enhancement or substitution for our work must be considered.

Accuracy in Teaching (Standard 7.03), Bases for Assessments (Standard 9.01), Interpreting Assessment Results (Standard 9.06), the use of Test Scoring and Interpretation Services (Standard 9.09), and Explaining Assessment Results (Standard 9.10) are all salient to consider in how we use Generative AI. If we choose to delegate some aspect of our work to Generative AI, we must remember that the veracity of the work products whether publications, teaching curriculum, or clinical reports remain firmly our ethical responsibility.

Conclusion

The human ability to generate insights from data is limited and is well surpassed by Generative AI. There are potential uses to improve efficiency and accuracy of psychological practice, whether clinical, research, teaching, or administrative. However, there are far-reaching limitations to Generative AI, as this technology cannot assess the context and all the parameters of the situation within which the data exists. Additionally, Generative AI is limited to only the data/information it has been trained on. The bottom line is that for psychologists to ethically use Generative AI in the practice of psychology, human input and review is still required to determine when and where to apply the use of this technology and to confirm the veracity of the information produced. As board-certified psychologists engaged in specialty practice, our mission is to serve the public by providing quality psychological services and we retain the ultimate responsibility for the information used in practice no matter whether generated through their own thoughts or through using technology.