Artificial Intelligence (AI) and Academia

Generative AI generates images, music, speech, code, video or text, while it interprets and manipulates pre-existing data. Generative AI is not a new concept and machine-learning techniques behind generative AI have evolved over the past decade. The latest approach is based on a neural network architecture, coined ‘Transformers.’ Combining transformer architecture with unsupervised learning, large foundation models emerged that outperform existing benchmarks capable of handling multiple data modalities (Goyal, Varshney, & Rozsa, 2023). Examples include: ChatGPT, Bard, Bing, Fireflies.

• Generative AI like ChatGPT is not an encyclopedia or a source for factual information.
  • Its knowledge is limited to its training data, which might be outdated or incorrect, and it sometimes suffers from hallucinations where it fabricates completely fictional information and presents it with great conviction — like making up titles and authors of books (Brownell, 2023). It can make up authentic-looking article information from the knowledge it has, when in fact the article information is nonexistent. See example below:

    • Diabetic ketoacidosis in neurosurgical patients" by Adogwa et al., published in the Journal of Neurosurgery in 2013. This article describes a retrospective study of 26 neurosurgical patients who developed diabetic ketoacidosis (DKA) during their hospitalization. The authors discuss the risk factors for DKA in this population and the challenges of managing it in the perioperative setting. (not a real article; produced by ChatGPT)

• Generative AI can be helpful in:

  • Organizing thoughts

  • Get the writing process started

  • Giving feedback for writing (think Grammarly)

  • Providing a different perspective