Prediction: The era of cognitive neuroscience in AI will begin this year and will be an important dimension of AI in medicine this decade. Cognitive architecture elements such as attention, consciousness, metacognition, abstraction, rules, concepts, and memory will be increasingly more incorporated into AI.
Outcome: There are more discussions of substance this year compared to last year. This is particularly relevant as the deep learning gurus continue to feel that deep learning alone will not push the momentum, particularly in domains like healthcare. This was especially obvious in the early period of the COVID-19 pandemic.
Prediction: Data from wearable devices will demand an AI strategy as this data will be otherwise much less effective in healthcare. The data “tsunami” from the increasing number of wearable devices will need to have an AI solution such as embedded or edge AI (machine learning + internet of things) to be relevant in the overall disease management in the future.
Outcome: The wearable device domain has really accelerated as a vital part of healthcare delivery due to the pandemic that required close surveillance either before or after hospitalization. In addition, there is robust discussion about embedded and edge AI for IoT and wearable devices in healthcare, especially with relevance to post-COVID care in some patients.
Prediction: Generative methodologies such as generative adversarial networks (GANs) will help neutralize the problem of inadequate healthcare data. A major issue is lack of healthcare data (including access) for AI projects, but this deficiency can be partly neutralized by generative AI methods that will create synthetic data (although heterogeneity of data may be an issue).
Outcome: As the editor-in-chief of the nascent Intelligence-Based Medicine journal that was launched this year, I was surprised by the number of manuscripts that had deployed generative adversarial networks (GANs) in various forms in projects. This methodology is especially useful in situations with insufficient number of cases.
Prediction: AI in the form of robotic process automation (RPA) will become more appreciated as a useful tool in healthcare administration. As many tasks in healthcare administration are repetitive and susceptible to human errors, these tasks are amenable to AI tools such as robotic process automation and machine learning.
Outcome: The methodology of robotic process automation has really flourished this past year and perhaps especially with the advent of the pandemic. The advantages of this RPA methodology is that it is a good support for human labor that may be affected by the pandemic.
Prediction: Conversational AI will be increasingly more common and sophisticated in healthcare. With advances and improvements in natural language processing as well as in machine and deep learning, conversational AI (messaging apps, speech-based assistants, and chatbots) is a much needed AI tool in disease management in healthcare.
Outcome: We saw the arrival of more powerful natural language processing tools like GPT-3 and this tool will have significant positive impact on healthcare and its continual need for conversational tools and for mining of data in electronic health records.
Prediction: There will be less hype about deep learning and its ability to predict with superior results and more focus on patient outcome and
Outcome: There were manuscripts that did elaborate on the importance of patient outcome in addition to performance in the AUROC. Higher performance often does not translate to better patient outcome so this is a key development.
Prediction: There will be more application of AI in altered realities of virtual reality, augmented reality, and mixed reality.
Outcome: With clinical training and education no longer always possible on site at a healthcare facility, this domain of AI in extended reality (“intelligent reality”) remains relatively unexplored but will mature in the next few years.
Prediction: Certain surgical and procedure-based subspecialties will incorporate image interpretation and deep learning more during procedures.
Outcome: There is an increase in using deep learning in biopsies, both surgical and procedural, by surgeons and gastroenterologists, respectively.
Prediction: Important issues in AI in medicine such as bias, inequity, and particularly data privacy will be even more in the forefront.
Outcome: Perhaps in parallel with the protests on racism and gender and sexual inequity, there is more discussion on the role of AI in these very powerful social issues that has really been brought to the forefront during the past year.
Prediction: Regulatory policies regarding AI technology in medicine and healthcare will start to reflect the exponential rise of AI capabilities.
Outcome: Again, due to the pandemic and the push for exponential trajectory of acceptance of technologies like artificial intelligence and other emerging technologies, the FDA has advanced an approach that will expedite the traditional medical device approval process.
So that was 2020’s crystal ball. Look out for my predictions for 2021!