As we’ve mentioned in previous blogs, AI, despite its power, efficiencies and potential, has some decided weaknesses. Additionally, training AI for AR isn’t the relatively straight forward process of “point and shoot” the way picture and graphic pattern recognition is for neural networks. There are several challenges:
AR training data generally must be gathered from diverse sources and domain knowledge must be added to the process to even determine what data would be relevant to the task.
Relying only on their own customers and experiences, most organizations will not have a sufficiently large sample to provide a good, reliable data set for training.
Current AI, as implemented in neural nets, relies on statistics, so domain knowledge must be carefully incorporated into the decision model to prevent false correlations
Any AR/credit decision must be accountable. It’s not sufficient for the AI just to deliver a “correct” answer. We need to be able to understand and justify the decision.
One step for any business toward addressing these challenges is collaboration at two levels. At the system level, working with domain knowledge experts, who have experience injecting AR subject matter into the AI model, who have access to large industry wide data sets that can be used to ensure productive training and who understand the current limits of any AI model and can design the overall process accordingly. And at the implementation level, structuring the AI to work collaboratively with a human partner, so each can perform to their respective strengths synergistically and thereby achieve a more productive work environment.