Keynote Speakers

Tim Menzies

Title: AI software is software and we, as software engineers, have to understand how to use and refactor and modify it

Abstract: Much of SE treats AI as some kind of off-the-shelf service where SE researchers and practitioners use tools, without modification, imported from the AI community. That approach is pragmatically appealing since, that way, SE people do not need a deep understanding of those AI tools. Nevertheless, I would encourage you **not** to do that since SE needs a different kind of AI. Certainly, these new AI tools can be based on standard AI methods. But before we race to apply methods from other communities, we need to first reflect on what our community needs and knows and wants. And if we do that, then we arrive at a better kind of AI for SE. To back up this argument, I will discuss applications from NASA flight control software, software text mining and sentiment analysis, software process planning, information retrieval, software quality prediction, software issue mitigation, software configuration, search-based software engineering, hyperparameter optimization, software bias mitigation.

Michael S. Mazzola

Title: Back to the Future: How the decarbonized electric grid of 2050 will be built on a model from 1897

Abstract: In 1894 a former secretary to Thomas Edison by the name of Samuel Insull was visiting Brighton England on Christmas holiday. Insull was then President of the unprofitable Chicago Edison, a rather small electric utility in Chicago, a city crowded with other competing utilities. He was astonished to learn that the company providing electricity to Brighton had time-of-use electricity rates. This pricing model allowed Insull in 1897 to work out how to increase the load factor of his utility by selling electricity during low-demand period at reduced rates. This increased the sales of electricity at a time when the expensive capital equipment would otherwise be idle. This in turn led to the growth of his utility empire through the socio-technical model of economies of scale. But initially Insull’s model required the participation of consumers to make it work. Later, the rise of state regulation of so-called natural monopolies returned the pricing model to generally flat electricity rates (with exceptions) in return for a duty to serve the customer by the utility. For residential customers this is largely the model of 2022, and it has allowed the electrical grid to become the most critical of all infrastructure. In other words, the reliability of electrical service has enabled a way of life that can tolerate interruption of electrical service for only short periods of time. However, the need, and in North Carolina the legal requirement, to evolve the electric grid to be carbon neutral has led many planners to realize that Insull’s model of consumer participation will once again underpin the operation of the electric grid. And since the grid will become even more critical to the workings of the economy through the electrification of most aspects of the energy economy, this introduces the duty to participate by the consumer as essential to maintaining reliability. This presentation will describe how IOT technologies will allow consumers to fulfill their duty to participate by assisting a grid dependent on variable renewable energy resources with load flexibility. New software-controlled, grid-edge technologies are needed to develop and mature vast new socio-technical systems of secure and ethical automation that understands human behavior and that allows utilities to work with consumers to jointly control the electric grid of 2050.

Agus Sudjianto

Title: Machine Learning Reliability for High-Risk Applications and Regulated Industry

Abstract: Machine Learning (ML) models are quickly becoming ubiquitous and widely embedded in many banking software for mission critical applications. Testing, validating and monitoring for model failure are key activities before and after model deployment as model errors can create significant risk. As models in production will operate under constantly changing environment, model design and testing must include model stress testing evaluation including their robustness, reliability and resilience.