Machine Learning Risks for CI Applications - (Part 6)

There are areas where, if not appropriately applied and managed, Machine Learning can have its problems. This is why it is important to understand the added value that ML brings to the table and not rely on this technology as a “cure-all” within your intelligence systems.

There are several challenges to ML implementation and usage, but in practice, these are the top three that need to be managed:

1)    Human Judgement Override

ML is simply one tool in the intelligence practitioner’s toolbox. There are many other tools that are used for various applications to arrive at the fastest time to insight and execution. Reliance exclusively on any one tool is not recommended. However, knowing when and how much to apply a level of human judgement to the model is also a challenge. Sometimes, an overzealous use is caused by having a preconceived determination on how the analysis or outcome should be going (sort of like guiding a mouse thru a maze, rather than letting the mouse guide itself).

2)    Replacement versus Augmentation

In theory, all ML learning should ultimately result in a “single” point of knowledge. i.e. regardless of any situation, you arrive at the same outcome as determined by the algorithms. Of course, this is highly impractical in a real-world scenario. Therefore, it is critical that ML reliance is based on an interaction of this function with other algorithms and tools, including human intelligence. The objective here is to augment your toolbox to include a highly valuable capability around ML that can help you anticipate industry and competitive behaviors within the overall intelligence modeling.

3)    Specialized Learning versus Generalized Learning

There are applications for both, but understanding where to use them is the key to success. Some ML capabilities can be confined to specific components of the intelligence model, while others can holistically provide industry and global knowledge. Sometimes users may over-apply ML to areas that might not be best suited to the design of the ML algorithm, thus polluting the findings of other tools and compromising the overall results of the analysis. Understanding how to apply specific ML functions that integrate with your overall analysis is the desired objective for the intelligence professional. 

ML can have valuable applications towards a CI practitioner's needs, but only when understanding the impacts and capabilities relative to the application in a real-world environment.

More to come...

This was part SIX in our series, based on the article “Machine Learning Implications for Intelligence and Insights”, written by Jesper Martell, Comintelli, and Paul Santilli, Hewlett Packard Enterprise.