Real world #AI will need to have object permanence: the concept that an object continues to exist even when the object is not able to be seen or heard. Psychologist Jean Piaget described this phenomenon as mental representation of objects that children develop around 18-24 months
There is a growing focus on a neuro-symbolic #AI approach that will couple vector-symbolic architectures with other disciplines such as linguistics, psychology, and neurosciences. Coupled with neural networks, this may be a strategy to achieve artificial general intelligence.
Deep learning and symbolic #AI continue to be on opposite sides, but DeepMind utilized a hybrid model for AlphaFold. This hybrid model has the best of both worlds: deep learning with learning from large volume of data and symbolic AI with representation of protein structures.
Manipulating symbols may not be familiar to some as it is not a machine or deep learning concept. Symbols (basically codes) are patterns that represent things and manipulating is processing the symbols in a specific manner (by algebra or logic); this is the essence of symbolic AI
The #programminglanguage #Julia, a high-level, high-performance and dynamic language, is becoming more popular for data science and machine learning. Started in 2009, Julia is a general purpose language that can accommodate complex computational problems in healthcare.
More and more emphasis is being placed on #healthcare data and its conundrum when discussion about data science and #AI in healthcare takes place. High quality of data is absolutely the essential foundation of data-information-knowledge-intelligence “pyramid”.
Ryan Baker, who teaches feature engineering, describes #featureengineering succinctly as “the art of creating predictor variables”. Feature engineering process is akin to a composer composing his/her musical composition using musical notes (data).
There are different types of drifts in data science. #Modeldrift can be broken down into #conceptdrift where the outcome changes over time, whereas #datadrift is when the predictors or features change over time. One needs to think of ways to keep models in deployment “fresh”.
Much of the high-cost administrative work can be executed by robotic process automation and the cost savings can be put back into the health system for other artificial intelligence-related projects as well as #preventivehealth measures.
With the advent of transformer-type #NLP tools, the portfolio of chatbots/virtual assistants can be more sophisticated and reduce the burden of common medical situations that can be safely redirected.
The individual #healthvirtualtwins can be coupled to the entire cohort of the population to better predict #healthcare interventions and outcomes in real time for the region. We can have regional population health with predictive modeling.
The health system will be following individual health with a screening program (a #healthvirtualtwin) so that #healthcare is entirely proactive to diagnose diseases such as cancer, heart disease, and diabetes.
The testing processes in the health system will be automated from beginning to end with results prioritized for intervention and will require human oversight only when necessary. Laboratory testing and imaging workflow and interpretation will be fully automated with #AI
Hospitals will evolve into health systems so that the health of individuals will be much more monitored when they are away from the health system. #Wearabledevices will have primitive #ArtificialIntelligence capabilities.
Artificial intelligence can help organize and collate real-world data from a myriad of sources to create evidence generation and render randomized controlled trials nearly obsolete.
By the end of the decade, we will most likely see lawsuits pertaining to not using the appropriate #AI tool in clinical decision or medical image interpretation.
There is a surge in use of #extendedreality in healthcare but this is not yet coupled with artificial intelligence to render a “visual” experience into an “learning” experience with ambient intelligence for patients and caretakers.
Most #healthsystems are still performing analytics in the descriptive category with outdated software and the future hospital analytics will be performed with cognitive intelligence with deep reinforcement learning.
The lack of #objectpermanence is an issue for #AI in autonomous driving and other tasks. This understanding or appreciation that an object still exist when the image of that object is no longer visible can have limitations for medical imaging as well, especially moving images.