Introduction
Much of the recent AI revolution has been focused on automation through big data and/or sensors and feedback into neural networks. The resulting applications are highly valuable to businesses and consumers. They improve quality of life by optimizing labor and resources. However, these applications fall short when it comes to handling human reasoning. Much of the rationale behind the operation of these systems are implicitly embedded in the data. In this article, I explored a different approach to AI where a machine uses reasoning to enable humans to solve new problems and understand new subjects. I have implemented such a system that – to a high degree – can automate the role of a management consultant. Using principles of Meta-Vision and Bionic Fusion, an AI system can automate much of the mentally-intensive work a management consultant performs. This automation results in a faster pathway to insight and elevates tactical strategic planning & execution. I call this Bionic Fusion: “Robo-Management Consulting.”
Machine learning, quantitative data, and the importance of reasoning
Data science enables us to learn by observing data behavior. Machine learning is a popular technique used to discover how common data patterns relate to common outcomes. This enables data scientists to predict outcomes with present data – a strategic benefit driving the adoption of machine learning across many enterprises. This approach, however, has limitations; machine learning can only learn from the past. Humans are dynamic; change is constant. Machine learning models trained and tested for accuracy against historical data don’t know what to do when faced with zero-day scenarios – new variables, unknown outcomes. When an application works with human behavior, the AI must account for human rationale. If presented with a conference dialogue or transcript, the application must understand context, sentiment, and rationale based on the situation.
In recent years, many analytical solutions and tools have focused on quantitative data to navigate and extract insights. Quantitative data, however, is only a measure of human behavior, not rationale. For example, ‘Same Store Sales’ is a metric often used in the retail industry. Machine learning models may recognize a decline, but will miss the underlying reasons driving that change – critical insights for executives managing a turnaround or competitors looking for a weakness to exploit. Identifying and understanding the root-cause is critical to successful business execution. The value of rationale analysis is just as important – if not more – than quantitative analysis in the formulation of tactical strategy.
An Implementation Using Reasoning Models
In our implementation of our rationale analytics, I look at the science of rationale as a determining factor in selecting algorithms for analysis.
Depending on the nature of the problem, I use three different reasoning models for rationale analysis:
1. If a given premise is known, I use a deductive reasoning model. Known premises are inferences drawn within the scope of propositional logic.
2. If a given premise is unknown, I create a hypothesis. I then use inductive reasoning. In this case, the causation model is a new hypothesis. I will not apply the newly learned premise in deductive reasoning until it is accumulated into a class of common truth.
3. When the observation is incomplete, I need to hypothesize the missing piece(s) of the puzzle with an educated guess. For this type of situation, I use abductive reasoning. I will then use our data lake to drawn reference and validate our causation model to complete the rationale.
In my previous blogs, I have discussed some of the novel technologies that I have developed for performing these tasks. I am unaware of any open-source implementation of these principles. For the purpose of discussion, I use our SaaS analytics service to develop Robo-Management Consulting and create management analytics reports with the help of artificial intelligence. Using “Context Discriminant”, I am able to extract important subjects and supporting facts from a corpus to get a high-level view with “Meta-Vision”. The “Meta-Vision” graphically shows us the attributes of relationships between the “Machine Generated Hashtags” and supporting facts in original context.
Through this process, our supporting fact model is transformed into a propositional causation model that corroborates the premises using business intelligence from our data lake. By combining the rationales of both the original corpus and the corresponding corpus from BI, I created a rationale model. The resulting Meta-Vision is then used to obtain insights and solutions to complex problems.
