Colorectal cancer (CRC) is a leading cause of mortality, and the second most common cause of cancer related death in the United States [1]. Declines in the incidence and mortality of CRC over the past decades have been attributed mostly to colonoscopy [[2], [3], [4], [5], [6]], which allows for both the diagnosis and immediate removal of early cancerous lesion through polypectomy.

While all polyps are routinely sent for final histopathology, about 70–80 % of polyps are diminutive [7] (<5 mm), with advanced histological features found in only 0.7 %–1.7 % of polyps that size [[8], [9], [10], [11]]. While both the American Society for Gastrointestinal Endoscopy (ASGE) [12] and the European Society of Gastrointestinal Endoscopy (ESGE) [13] recommend the removal of all polyps in the interest of reducing colorectal cancer incidence and mortality, many of these polyps have little to no malignant potential. The ability to visually diagnose whether a polyp has cancerous potential and leave in-situ if able (also known as “diagnose-and-leave”) or resect-and-discard without histopathology assessment would decrease overall colonoscopy time, use of resources, unnecessary use of pathology, and the risk of complications related to polypectomy [[14], [15], [16], [17], [18]].

Technology has been developed in order to improve optical diagnosis of histology by allowing for improved visual assessment of polyp architecture, including narrow-band imaging (NBI) [19,20], iScan (Pentax) [21,22], magnifying chromoendoscopy [23], Fujinon Intelligence Color Enhancement (FICE, Fujinon Inc) [24] and confocal laser endomicroscopy [25], among others. While there is a magnitude of data supporting the use of these technologies and their meeting the ASGE Technology Committee thresholds set for this purpose [14], universal uptake of optical diagnosis and its associated strategies has been slow.

Since ChatGPT was originally released in November 2022, artificial intelligence (AI) has expanded with applications in nearly every field, including clinical medicine [26,27]. In the past years, several trials have looked at the applications of AI to optical diagnosis [[28], [29], [30]], with the hope of determining whether AI will change the landscape of the field, and lead to a paradigm shift where the diagnose-and-leave strategy is widely accepted. Here, we will explore this data and seek to determine whether optical diagnosis now seems plausible in the world of AI.