Superimposing Natural Intelligence on Artificial Intelligence: Optimizing Value

摘要

Computers are number-crunching machines. Humans cannot - and, perhaps, need not - compete with computers when it comes to handling big data volume and velocity. Machines are good at data-driven predictions that are difficult for humans to make. Humans are better at handling smaller sets of data, inspecting the relevant aspects (contexts), and arriving at decisions appropriate to specific situations. As opposed to computers, humans are fuzzy, knowledge-driven analyzers. Their NI (human or natural intelligence) includes not only knowledge and experience, but also intuition and insights. Values and their dynamics are beyond the grasp of algorithms. Perhaps machines may never be fully learned, not because of a shortcoming in ML systems but because values in decision making are inherently unlearnable. This unlearnable aspect poses a challenge to automation and to the optimization of ML. AI models are extremely complex because they incorporate a wide variety of data with high volume, velocity, and variety. Added to the complexity of the data is the complexity of ML algorithms. Finally, AI models are not singular, standalone systems, but rather a combination of multiple, collaborative systems typically interacting with each other in the cloud. Thus, when the systems are totally automated and are, in effect, black boxes, any breakdown in a system may result in total chaos across all business functions. The results can be uncontrolled and dangerous, as it is typically very difficult to determine the source of the fault and fix it. Therefore, for a business decision-making or analytic task, human experience and an intuition-based, fuzzy way of thinking should complement the data-driven techniques of machines. In this article, we have argued for the need to superimpose NI on the predictions made by AI and have examined the likely risks without such superimposition as well as the advantages resulting from it.