Trends to watch: Top Three

Adopt

Continuous delivery for machine learning (CD4ML)

CD4ML is a software engineering approach in which cross-functional teams produce ML applications based on code, data and models in small increments that can be reproduced and released anytime, with short adaptation cycles. CD4ML allows enterprises to build the capabilities and processes needed to govern, release and update ML models in a reliable and safe way, while remaining able to respond to changes and evolutions in the ML landscape.

Analyze

XAI

Explainable AI (XAI) refers to a set of tools and approaches to understand the rationale used by an ML model in reaching a conclusion. These tools generally apply to models that are otherwise opaque in their reasoning. Explainable AI is becoming increasingly important as ML models are applied in areas that directly impact the health, rights, and economic well-being of individuals, such as healthcare, criminal justice and credit decisioning. XAI will likely play an even greater role as the implications of AI for privacy and regulations such the EU’s General Data Protection Regulation (GDPR) are better understood.



Anticipate

Brain-computer interfaces (BCI)

BCI interfaces enable users to interact with computers by means of brain activity only, generally measured by electroencephalography (EEG). Advances in the use of machine learning to interpret electrical signals from the brain have created new use cases where this could be applied to help people make split-second decisions, and vastly decreased the costs of technology in this space.

Advice for adopters

• Develop trust in your data. The AI models that support your teams will be created based on the data made available to them. That means the data must be transparent, accurate and reliable enough that you and your people can generally trust any decisions resulting from it will be sound and well-informed. 
  
  
• Understand and communicate where machine-made decisions are subject to bias and ethical concerns. Create a framework to track what sensitive decisions will be made, describe how you expect these to affect any stakeholders involved and set out how you will detect potential unintended consequences. Ensure these decisions can be sufficiently explained and traced back to the ‘source.’ 
  
  
• Create quantitative baseline measurements of your current processes. Create quantitative baseline measurements of your current processes using techniques such as value stream analysis to identify areas with significant opportunities for automation, where AI and ML may produce the biggest potential gains. Refer to and update these on a regular basis to measure the effectiveness of AI/ML rollouts, and ensure automation remains targeted rather than the default solution for every problem.
  
  
• Empathize with staff fearful that “the robots will take our jobs”. Empathize with staff fearful that “the robots will take our jobs” and support them as they learn to work alongside algorithms, making it clear the ultimate aim is freeing them up for more rewarding work. Develop clear transition and career progression plans for your employees to grow into roles enabled by automation handling repetitive tasks. 
  
  
• Consider whether human-machine collaboration can help you enhance customer value. Consider whether human-machine collaboration can help you enhance customer value, rather than seeing automation purely as a way to replace headcount and reduce costs. Existing staff have a wealth of knowledge about your business and customers, and these insights should be applied to effectively boost customer experience as they people spend less time on mundane functions. Any AI or ML solution should be judged not by the efficiencies it creates internally, but by the additional customer value it enables teams to deliver.