Is Artificial Intelligence Now Routine Engineering?

Guest post from KPMG by Ken Hardy & Bryan Clarke

Artificial Intelligence has provided scares for moviegoers for over 70 years. From Gort in The Day the Earth Stood Still to Ava in Ex Machina, the potential of artificial intelligence has long been promised but is it now starting to become realised. AI has potential for a range of applications, particularly as a tool for automating data-intensive activities and delivering novel insights in industries such as Fintech and Pharma.

JP Morgan has an AI Research team headquartered in New York that is investigating trading using image classification. They observed that traders always execute their trade orders while observing images of financial graphs on their screens. Using the images, they trained over a dozen machine-learning classification models. This progress suggested that visual recognition is beneficial for identifying trade signals.

A recent article in the Lancet, one of the world’s oldest and best-known general medical journals, discussed the role of AI in COVID-19 drug repurposing.  The paper presented guidelines on how to use AI to accelerate drug repurposing or repositioning. Not only did they discuss how to use AI models in precision medicine but also how AI approaches are now necessary and can accelerate COVID-19 drug repurposing.  The researchers expect future successful AI models for drug repurposing to be accurate.

While AI needs to be facilitated by the processing of a large amount of data, the number of tools and frameworks available to data scientists and developers has increased exponentially with the growth of AI and ML.

The major players in cloud computing all now offer machine learning (ML) tools to help computer scientists and engineers develop applications and systems to help companies remain competitive.  Microsoft Azure’s offering empowers professional and non-professional data scientists to rapidly build ML models using the no-code user interface (UI) or using a software development kit (SDK).

Amazon offers a series of ML solutions that complete tasks such as turning text into lifelike speech (Amazon Polly), deep learning-based image recognition (Amazon Rekognition) and natural language processing (Amazon Comprehend).  Google’s second-generation AI system called TensorFlow was released in February 2019. You can now certify in all three of these tools and become professional ML engineers.

With the growth of AI and ML, the number of tools and frameworks available to data scientists and developers has increased dramatically. IBM Watson, Scikit Learn, TensorFlow, Theano, Caffe, MxNet, Keras, PyTorch, CNTK, Auto ML, OpenNN, H20: Open Source AI Platform are just a few examples.

With the advent of these new tools, the question of the role of AI and ML activities in a businesses’ operations determines whether or not a company can qualify for R&D Tax Credits. If AI is now part of routine engineering the answer is no but if AI still involves R&D then the answer may be yes.

Irish Revenues research and development (R&D) Tax Credit Guidelines state that “software developments using known methodologies in standard development environments using the standard features and functions of existing tools would not typically advance technology and would not address or resolve technological uncertainty”.  This is an important definition in determining if your ML project can qualify for R&D tax credits.

The use of AI to automate data-intensive activities using known ML models to process large quantities of data in cloud environments such as Azure, AWS and Google cloud with a predictable outcome, some might argue, is becoming routine engineering and is something that you would expect a professional ML engineer to be able to complete.

However, there remain challenges in developing these AI tools, such as data heterogeneity and low quality, insufficient data sharing by pharmaceutical companies, as well as the security and interpretability of the models. The JP Morgan study suggested that visual recognition is beneficial for identifying trade signals only. Therefore, the use of AI to deliver novel insights appears to be very much at the cutting-edge of research and development.

If you are considering claiming R&D tax credits for an AI and ML project, it is important to understand if the project is routine engineering or involved research and development.  As with all new technologies that mature over time it is important to carefully consider how your project meets the five key science criteria set out in the R&D tax credits legislation. KPMG’s R&D Incentives practice assist our clients in this regard.