AI, A Menace to Stability?
image credit: ID 159087096 © Kosssmosss | Dreamstime.com
- Nov 30, 2019 12:17 am GMT
- 1154 views
In Aldous Huxley’s Brave New World, efficiency is paramount and every change is a menace to stability. The ‘Savage’ rejects the 'World State' utopia and clings to ideals like inconvenience over comfort and independence over proficiency. Fortunately, our society doesn’t match either extreme found in the novel but how much are you willing to give up or hand over to machines for comfort and convenience? Today, Artificial Intelligence(AI) and Machine Learning(ML) promise to provide unparalleled efficiencies for more and more businesses. In truth, AI and ML are already boosting productivity and efficiency in the energy sector.
- Duke Energy uses AI algorithms to safeguard its multi billion-dollar turbines by predicting potential disasters and shut down. After implementing over 30,000 sensors the utility used machine learning to better identify areas most critical to its infrastructure maintenance and performance. For example, a small change in normal behavior was detected in a particular turbine which alerted engineers to some loose bolts inside the flow sleeve, something that may have gone unnoticed without sensor data.
- Georgia Power, a subsidiary of Southern Company, has been employing analytics to improve operations, cut costs and enhance customers’ experience. Southern Company has considered new algorithms to access grid stability. For now, they are working on developing and evaluating new analytics and artificial intelligence tools to generate new insights for improved decision support.
- NextEra is using machine learning and predictive analytics to boost the functionality of wind turbines and reduce maintenance costs. Florida Power & Light, a subsidiary of NextEra, identified unusual patterns in the functioning of plant combustion turbines using machine learning models.
Looking back, some of last year’s AI projections have come true. From ‘Ethics in AI’ to concerns about ‘Machine Bias,’ we’re witnessing the cause and effect of Artificial Intelligence and Machine Learning across industries worldwide. This year has certainly seen legal accountability, transparency, policy and the governance frameworks of AI come under scrutiny. ML models deserve our undivided attention as they are now being used for decision-making such as hiring, mortgage loans, prisoner parole releases and the types of social service benefits applicants can qualify for. Trying to get ahead of the issue, the World Economic Forum released a report to prevent discriminatory outcomes in machine learning.
The challenges and concerns surrounding artificial intelligence are innumerable but so far the projected benefits far outweigh any growing pains we may experience. Understanding the need for funding, the U.S. Department of Energy announced $15 million in funding for 23 projects to accelerate the incorporation of machine learning and artificial intelligence into energy technology with the DIFFERENTIATE program. “The incorporation of AI and Machine Learning into our energy technology design and engineering processes has great potential to increase the productivity of our nation’s engineers and scientists,” said Rick Perry. Despite plans to resign as Secretary of Energy at the end of the year, Perry conveyed his interest in pursuing AI in Energy by saying, “These DIFFERENTIATE projects truly work towards creating a competitive advantage that further positions the United States as the world leader in AI technology development.” Additional challenges for power companies are limited risk tolerance and lack of internal data science capabilities. Integrating new platform systems with traditional systems is a challenge for any company older than the dot-com era but uniquely so for utilities. Shutting down operations to test new systems is simply not an option. This doesn’t mean we have given up. Currently, the most common applications of AI and ML in the energy industry include asset failure prediction, DER location optimization, customer churn prevention, safety incident prediction, and inventory management.
How can AI be more widely used in the industry? In the past, we limited computers to storing and indexing data and now we are asking them to make decisions. Thanks to several successful applications, machine learning is gaining approval. The majority of commercially proven applications focus on data or machine-learning/deep learning algorithms for better prediction. An algorithm is a set of instructions designed to perform a specific task. In this case, the algorithm gathers information and uses it to make decisions for us. Within the energy sector, one application of machine learning is analyzing the data gathered and identifying strong predictors of demand during certain hours in the day. This task had not been identified by internal SMEs. However, ML systems are pushing ahead, identifying demand, updating forecast models, improving error rates and reducing financial losses.
Data, Big or Small
A large quantity of accurate data is imperative to predicting behavior and demand. Not having a good dataset or a well-curated collection of data to train an AI system will hinder performance. Which will, in turn, undermine the trust companies have in artificial intelligence. Therefore, the quality of the data impacts the success of any ML system. Data must be not only collected but indexed correctly to ensure accurate outcomes. An algorithm is only as good as its author. That is, of course, if the author has been given the correct information in the first place. Supervised or unsupervised algorithms are used to find and identify trends in large sets of data. Supervised learning involves learning with labeled datasets to produce an output that is generic to that dataset. Unsupervised learning involves finding the connections between unlabeled data or clustering that data. Machine learning algorithms rely on statistical bias, that is, existing data is compared to new data and predictions are made.
How do we avoid machine bias? Bias reflects problems related to the gathering or use of data, where systems draw improper conclusions about datasets, either because of human intervention or as a result of a lack of cognitive assessment of data. One example of bias is the use of these algorithms in law enforcement. COMPAS, a machine learning algorithm used to determine criminal defendants’ likelihood to recommit crime provided inaccurate and biased predictions which were either due to a limited list of data categories, or inappropriate or invalid personal data. More commonly understood, in the advertising on the webpage you're searching are ads based on data acquired from your search history. However, this same data gathering algorithm may be inadvertently showing women lower paying jobs than men. A study by AdFisher revealed that men were six times more likely than women to see Google ads for high paying jobs, leaving women out of the applicant pool. Correcting these algorithms to eliminate bias must be first on our to-do lists. Learning from the mistakes of others can only help utilities moving forward with machine learning systems.
Data is the New Oil
“Data is the New Oil” according to a 2017 broadcast by the Economist. Data is crucial so the issue of compliance has surfaced. Companies have the data but do they have permission to use it? With Facebook before Congress, privacy protection and access to data privacy are not just the hot topics of late but are actively changing policy. Companies are responsible for getting consent to use systems that gather customer information. In other words, data ownership is a-must for utilities implementing ML systems.
With computers and systems learning over time, they can optimize and perfect tasks in such a way that changes how we do things forever. AI in energy calls into question Huxley’s utopian theory that ‘every change is a menace to stability.’ Machine Learning is changing the game. It will replace jobs as it widens our capabilities. Will this change be a 'menace to stability' or more you the point, reliability? AI and ML should provide accuracy, efficiency, and real time feedback that trumps any inconvenience by a brief interruption to ‘stability.' It is important to note, implementation is not one size fits all. ML systems require case specific information and often times patience, flexibility, and extensive testing. Be realistic. Your utopian utility may take time. Start simple. Focus on a well-defined goal. Consider the scalability of the project. And most of all, obtain, own, clean and verify the data.