Ensuring the success of the Content side of Smart Algorithms
Smart Algorithms encompass all kind of algorithms using Artificial Intelligence in Digital Assistants, Robotic Process Automation, and Machine Learning.
Smart Algorithms are on the verge of pervading all aspects of business and our daily lives. At the moment, most efforts seem to be focused on the “low hanging fruit” such as replacing cumbersome text interfaces with Natural Language Processing interaction, performing pattern recognition in huge data sets, or automating repetitive or highly regulated processes requiring low human intelligence (even driverless cars fall into this category).
Current developments seek to push automation into higher value-added tasks. Although machines can certainly imitate and reproduce human behaviors, we believe that machines cannot, by their very nature, reproduce human thought (despite claims to the contrary).
Therefore, instead of attempting to automate actual human expertise, and replacing human beings at work, we believe that to be successful, automation should seek rather to enhance human work performance.
The next threshold in automation will need to pay significant attention to content-based algorithms, which requires: (1) clarifying and narrowing the scope of the intended domain, and (2) successfully organizing and running the automation project.
- Clarifying and narrowing the scope of the domain: what makes a human expert so powerful are the “distinctions” that they apply in complex situations, often subconsciously:
- By eliciting and capturing these distinctions in a usable form for others, including those who will be programming the automation algorithms, we provide explicit knowledge boundaries and related human points of view.
- In addition, certain traps need to be avoided when automating, including two extremes:
- Cognitive biases / prejudices that can creep into the algorithms: These are “false distinctions”, i.e. divisions / categorizations being made where none actually (or no longer) exist;
- Lack of sophistication in automated rules: When an underlying distinction goes undetected and therefore is not implemented when it ought to be.
- We promote functional exception handling, consistency and completeness techniques within algorithms for verifying on the spot the knowledge domain boundaries.
- Successfully organizing and running the automation project: successful projects share certain characteristics:
- Well-defined scope (including at least automation target scope, technology scope, and human/organizational scope) in the context of human-machine systems;
- well-defined objectives and clear success criteria;
- Frequent coherent deliverables enabling rapid feedback.