Expertise automation – the missing link in RPA-AI based intelligent automation

It is now widely accepted that the RPA enabled digital workforce represents the catalyst or the gateway drug (phrase coined by HFS Research) to AI powered Intelligent Automation (IA). This clearly implies that the adoption of AI is critical to the success of IA. However, there is a misconception that Deep Learning is the Foundational technology that will underpin all AI. Based on my 35 years experience in applying AI in many fields, Machine / Deep Learning based AI alone will not be capable of delivering a wide range of Intelligent Automation applications and this will become a fundamental issue stopping many enterprises from unlocking AI’s full potential and implementing successful intelligent automation solutions. Machine Learning based AI need to be used in conjunction with the tried and tested expertise automation (Symbolic AI) technologies. Combining learning from data and automated human expertise is key to Intelligent Automation

There is also a misconception that the well established expertise automation (Symbolic AI) approaches of the 80s, 90s and early 2000s are outdated and have been superseded by the DL based new AI. Fortunately, we are now beginning to hear voices that are prepared to challenge this narrow view of AI and argue that far from replacing the Symbolic AI approach, the Machine Learning based AI is very complementary to Symbolic AI and that the blending of the two approaches is what is needed to deliver successful IA solutions. Below is a brief articulation of the argument for combining Symbolic & Machine Learning AI.

Deep learning has been remarkably effective is addressing the previous limitations of Symbolic AI namely; Speech, Vision, Text, Natural Language processing and complex pattern recognition. Some of these capabilities have even exceeded human abilities.

However, in many Intelligent Automation applications, the breadth and volume of the data required to apply deep learning is simply not available. Examples of such scenarios are automating the performance monitoring and fault prediction for complex machinery, generating complex treatment plans for patients, generating configuration & quotations for complex products and services and many others . In contrast, automating human expertise using Symbolic AI can overcome the data limitation problem because human experts are able to learn from a very small volume of data. This is because human learning differs drastically from current approaches to machine learning as summarised below:

  • Human experts benefit from what is called transference learning whereby they apply expertise from a previous domain to learn a new domain based on only a small number of new observations. Current Deep learning algorithms are unable to match this capability (or have limited ability in applications like image recognition)
  • Human expert can apply common sense knowledge of the world and background knowledge of the subject matter to make new predictions when faced with a new problem. Deep learning can only learn from the data presented to it as the modelling of common sense and subject matter knowledge is still not feasible.
  • Human experts are able to ‘understand’ concepts in natural language text and apply their transference learning and common sense and background knowledge to learn new concepts / expertise from the text. Deep learning cannot match this powerful human capability of understanding and is restricted to applying almost statistical word counting and pattern matching to natural language text processing.

There are a number of Symbolic AI approaches available to automate the various types of heuristic human expertise. Symbolic AI covers Decision / Business Rules automation, Expert Systems, Fuzzy logic, Case Based Reasoning, GA optimisation and other technologies. The Symbolic AI approaches had limited success / adoption historically due to three reasons and are now mostly overcome.

  • Connectivity : The limited access to enterprise data required for intelligent automation. RPA and IOT platforms have over the last few years made enterprise data accessible for Intelligent Automation.
  • Low Code development: The lack of low code platforms for the capture and authoring of the heuristic expertise of human experts.
  • Hybrid technologies: No single platform supported the multiple Symbolic AI technologies required for the development of a broad range of automation problems.


The last few years have seen the emergence of platforms, such as XpertRule, which support a low code graphical development environment and supporting multiple Symbolic AI approaches. The processing of Natural Language, text, speech and images using DL based cognitive services can provide structured outputs which can be consumed by Symbolic AI to allow effective automation in many fields.

The role of Symbolic AI in Intelligent Automation is to up-skill the digital workforce created by RPA thereby producing smart software robots! Examples of skills that are difficult to automate using DL based AI alone and therefore require expertise automation are:

  • Configure / Price/ Quote of complex products & Services in ecommerce
  • Determining benefit and parole eligibility, compliance with Building/planning regulations in Public sector
  • Medical Treatment planning
  • Validating regulatory reporting and compliance in financial services
  • Cognitive chat bots supporting complex conversations with users to make complex recommendation, advise or problem trouble-shooting
  • Complex task allocations & Sequencing
  • Complex risk assessment in financial services and plant inspection


For more information please visit