To allow AI to help experts make better decisions and answer critical questions about engineering specifications, the AI solution first needs to learn how to leverage the knowledge coming from the expert. And that knowledge is usually not just book knowledge.
Businesses often ask many questions like given our set of specifications, can we reduce or combine them and still meet our customers' engineering needs? and can we do so while optimizing our supply chain by saving time? Can we simultaneously minimize the diversity of activities we might need to support in the future?
This example represents an important use case I've encountered in many places where artificial intelligence can provide quantifiable value. Experts typically capture their knowledge and reasoning about complex knowledge like specifications in unstructured text to draw upon later. However, standard search technologies and data-mining solutions tend to fail when required to retrieve that knowledge. Search technologies struggle to account for relevant factors that are most important for specific events, like which factors are relevant in different engineering specifications.
To allow AI to help experts make better decisions and answer critical questions about engineering specifications, the AI solution first needs to learn how to leverage the knowledge coming from the expert. And that knowledge is usually not just book knowledge it is heuristic and experienced-based knowledge gained from many years on the job. Below are three ways using expert knowledge in our AI system helped answer questions and optimize the business:
1. AI Converts Text Into Useable Data Points:
AI applications are challenging to build because businesses and users rely heavily on experience-based knowledge obtained through years of solving similar challenges in different situations. AI applications must achieve something functionally similar, even if they do so in a different way. Experience-based knowledge is similar to our visual system's ability to identify and understand complex patterns in pictures: We quickly identify concepts and patterns in a picture and, through our prior experiences of similar patterns, infer what might have led to the image and simultaneously predict what might happen next to fully understand the picture.
Our brain can look across experience-based knowledge captured in the text in a similar way to associate prior experiences, identify the most important factors in the current situation and suggest the best solutions to problems. However, with expertise captured in the text in business, we first must identify meaningful concepts, relationships, and patterns from the text and documentation. Natural language processing is an AI technique that experts can train to recognize from documents important entities, events and relationships between those events and entities. By allowing the expert to label important entities and their relationships, NLP technology helps turn knowledge from these documents into something computers can process useable data points that represent experience-based knowledge.
2. AI Organizes Data Into Meaningful Graphs Or Networks:
Engineering specifications can contain hundreds of characteristics, described by various measurement types and attributes. Some are standard and other customer- or project-dependent. The formats of specifications can vary between business units and over time, making normalization into structured databases resource-expensive and error-prone. Specification management tools exist, but even when organizations use them well consistently, variations still arise due to how different parts of the business use them and how to use over time changes.
Knowledge representation in some form semantic technology, knowledge graphs, association networks provides a richer understanding of the domain and the normalization of relevant data. By allowing the expert to easily organize concepts and their relationships into a knowledge representation, which is then mapped to underlying data, the AI application can perform inference that spans things like different measurement units, data formats or data hierarchies.
3. Expert Knowledge Improves AI Techniques:
In a prior project, we used a logic-based approach for answering questions about engineering specifications formal ontologies modeled both the domain as well as the specification, machine learning extracted specifications from documents and logic-based rules were captured from experts to represent heuristics and domain knowledge. The approach encountered difficulties for two reasons: 1) It was difficult to avoid errors while capturing specifications into formal logic, and 2) the logic used to compare and understand specifications was often hard to capture due to its fuzziness and inconsistencies.
After that experience, we tried a clustering approach. Clustering works by grouping "like" things with other "like" things. But we realized clustering could suffice with a standard notion of similarity. Instead, we allowed the expert to help the AI system learn that measure of similarity. Because each instance this task required determining many possible different measures of similarity, it would be too high a resource cost to approach as a traditional machine-learning problem with a team of data scientists executing a multi-week project. We needed an AI to help experts capture scientists' notion of specification "likeness" their expert knowledge.
For our solution, we decided to learn the likeness model that encapsulated the experience-based knowledge about the domain and the most important factors about the engineering specifications for the question at hand. By giving the expert the option to select a few examples of what "likeness" means for their specific question, leveraging the knowledge representation and extracted concepts and relationships, the AI system was able to learn a custom approach of likeness to use in clustering.
4. AI Leverages Expert Knowledge To Answer Valuable Business Questions:
AI systems can help propel businesses forward to make better decisions, but they must be designed to incorporate expert knowledge and use it effectively to answer business questions. Most AI technology requires significant engineering work to customize it and make it perform in a specific domain, and only by putting the expert into the middle of that customization will AI systems be able to be created and perform efficiently.