In the 5th and final post of our series on the future of the Digital Worker (Read Post 1: Knowledge & Insight; Post 2: Visual Perception; Post 3: Planning and Sequencing; and Post 4: Collaboration, as well) we are going to explore two more of the required digital enterprise skills needed by the Digital Worker of the future – Learning and Problem Solving.
We’ve chosen to discuss these two Blue Prism Intelligent Automation Skills together because they are closely connected and relate to how the Digital Worker will be able to absorb information and adapt to its environment (Learning) and how it will be able to use this data to make both simple, deterministic decisions and probabilistic decision-making using Machine Learning (Problem Solving) .
Pre-trained AI and the Digital Worker
Now that pre-trained, narrow AI is readily available and consumable via existing APIs or Intelligent Services – such as the pre-built integrations shipped with Blue Prism that enable the use of intelligence in Google, IBM Watson, Microsoft Azure or AWS – providing Digital Workers with the ability to engage in Problem Solving has become as simple as dragging and dropping an action into a Blue Prism process.
Many tasks that require one of the six Blue Prism Intelligent Automation skills we have already explored can be achieved this way – for example the ability to translate or understand natural language, as described within the Knowledge and Insight Blog.
Pre-trained AI that is consumed via an API ecosystem is not a sub-standard approach to a built in, closed ecosystem and should not be underestimated for its capabilities. Indeed, we believe that these capabilities will very quickly surpass human levels of accuracy in many areas, as demonstrated in certain fields with the advent of Google’s AutoML.
Why have a Partner first approach to AI?
If you are involved in evaluating RPA vendors, you will not be short on opinions or viewpoints on AI within RPA, closed ecosystems, open ecosystems and grand claims about the AI that is ingrained within various products. Blue Prism has a partner centric approach to our AI strategy.
Let’s be clear – we are investing in building AI into our product, but it has to be appropriate – there are many opportunities we are exploring with our research division to create additional intelligence around autonomous workload scheduling, surface automation and other Blue Prism RPA firsts. The simple reason behind our partner first AI strategy is that the highest demand and business benefit for Intelligent Automation lies within the realm of business contextual, highly customized algorithms that will require highly specified GPU systems and advanced analytics tools to train, or within domains that will become highly commoditized by the vendors with the billion-dollar R&D budgets. By building strong, strategic differentiated partnerships with these vendors, we enable our customers the choice and flexibility to leverage existing investments and relationships and achieve the highest possible ROI from AI within RPA.
An often-overlooked fact in the marketing hype from many RPA vendors is that business-contextual AI and ML is far more complex to achieve in an RPA process, because the problem and the data involved are almost certainly unique to you and your business.
This is the inconvenient truth about AI and the claims made about the built-in abilities of some products. For a Digital Worker to be truly capable of learning and adapting, it must have the ability to exist within the entire lifecycle of a Machine Learning workflow:
- The Data Science Challenge and Defining a Model – The first challenge with integrating custom Machine Learning into an automation is creating the model in the first place. The approach varies, and we will not go into the technicalities and differences between supervised and unsupervised learning here, but generally this involves gathering (and labelling) large amounts of data and using high powered GPU based devices to train a model that is effective (at least as effective as a human) in solving the problem at hand. In most cases, this problem is best solved using the cloud, so that these high-powered systems can be scaled more efficiently and cost-effectively.
- Consuming a model – Once built, the Digital Worker must be able to obtain a prediction or outcome from the pre-trained model in such a way that it can then make decisions and progress through the process. As a side note: this will involve a shift in your attitude towards risk—from a simple, explainable, and predictable Boolean outcome, to one that may be based on machine intelligence and confidence levels. Think probabilistic vs deterministic. There are now many ways to achieve this, with many of the cloud vendors enabling a way of exposing these models via API.
- Continually optimizing the model – No matter how good your model is once it is trained, it will need to be constantly refined and optimised. This is, in many ways, the hardest part of the workflow. All data must be constantly captured, labelled (except in unsupervised approaches), separated into training and testing groups and then fed back into the model optimisation methodology where the new model can be tested to see if there is a direct improvement. This whole process generally involves a human in the loop.
We believe that the largest organisations will rise to the fore by making huge advances in AI due to their research budgets and focus. This is why Blue Prism has a “cloud first” strategy around this space and we have invested heavily in building strong partnerships with Google, Microsoft and IBM. We are now starting to release even tighter integrations with these platforms, building the most comprehensive ecosystem of pre-trained AI, but also integrating the Blue Prism Digital Workforce into the entire Machine Learning lifecycle. This will enable a streamlined workflow that allows you to build datasets, train a model on your platform of choice and consume that model from within a business process.
The strength of our strategic partnerships in this space is starting to yield fruit. With Version 6.3 of Blue Prism, you are able to integrate Blue Prism into the Machine Learning workflow using the Google GCP ML stack.
We are very proud that Blue Prism was selected by Microsoft as one of 100 partners globally to work with their AI product team on tighter integration of their AI and advanced analytics capabilities, with a strong focus on driving partner and customer adoption and case studies.
Final Thought – What about General AI and the Singularity?
What does the future hold? Well, even the luminaries at some of the
biggest companies in the AI space believe the advent of true, general AI
is still a long way off – anything between 10 and 50 years, depending
on who you ask. Our emerging technology team is focused on more
adaptive, narrow AI in the areas that make sense for RPA and the Digital
Worker – the use of Computer Vision for adaptively interpreting screens
and documents for example. Thanks to the strength of the partnerships
we have with the industry heavyweights such as Google, IBM and
Microsoft, and those we are establishing with more specialist ISVs we
believe that we are well positioned to benefit from the latest advances
in this space.