Top 5 Facets of the AI Gem
Blog: Jim Sinur
AI will be the gem of digital going forward for a long time. It is a co-driver of smarts in both automation and customer excellence efforts along with static algorithms. AI can learn, handle fuzzy problems, and help with increasing the probability of success in decisions, assist humans in interacting with traditional rule-based organizational systems and reaching shifting goals. There are five facets of AI that are shining bright now and for the future. There could be more down the road as AI progresses over time, but these are the top five right now.
Right now, ML is the brightest facet of AI as organizations deal with ever-growing big and fast data sources. ML learns for the data to get better and speed up responses to interesting patterns. ML is good at handling rich and complex data for incremental learning and thus assisting decisions and actions. The machines do most of the heavy lifting here, but the quality and control of the data is a key factor for success. The learning can get better with the addition of facets of neural nets to create deep learning opportunities to speed up the evolution. Keep in mind that ML can learn from bad data too and the maintenance of data can be costly.
While neural nets are popular in the deep learning portions of ML, they also have an identity of their own. They are great at interpolating between several taught patterns for classification and categorization. They pay attention to differences and emerging patterns. They are also strong at self-training and learning, particularly for unstructured data often found in natural language problems. Their strength is that no expert is needed, just training data. Keep in mind that ANN requires a significant number of patterns for better results and retention of patterns becomes a management issue. Retraining is also a factor to consider.
Fuzzy Logic is helpful when there is not a precise truth as it handles degrees of truth. It is good where there are grey situations. This is often the case with human and machine dialog where there might be linguistic uncertainties. FL is difficult to explain in some situations because it handles linguistic uncertainty. Because of this uncertainty, there will be situations that a subject to interpretation.
Bayesian Belief Networks:
Bayesian Networks are applicable to cause and effect problem domains. BBN’s are aimed at probabilities of the relationship between symptoms and situations or outcomes. This is accomplished by mapping the casual probabilistic relationship among a set of random variables, the conditional dependencies, and joint probability distribution. This is mapping is often represented in a visual model that represents a set of variables and their conditional dependencies via a directed acyclic graph (DAG). Keep in mind BBNs are dependent on having good statistics to drive results.
Reverse chaining is good at moving towards goals where all the underlying data may not be complete, but there are available inputs to leverage. Also, ARC can be sued to figure out the typical paths that brought an organization to their existing state. The inherent strength is that it handles missing information and data. Keep in mind that since there is a lot of trial and error, ARC is not the best at real-time control.
While AI is still an evolving gem in terms of application to an organizations problems, there are some bright spots that can deliver benefits. It is important to match the AI approach to the problem at hand even if there is a combination of AI facets used in the solution set. As more case studies emerge, the clarity of the application of these facets will increase and reduce the amount of pioneering. Vendors and solution providers will a real source of wisdom here over time.