João Saint-Aubyn, VP Analytics Strategy
Miguel Moreira da Silva, Managing Partner
Dear readers, we propose a brief game of two truths and a lie in the application of AI to companies in the energy sector, as well as industries at large. Let us begin:
"There won't be a single company that won't be immediately impacted by the current tidal wave of AI technology."
"AI is another technological shift on top of traditional efficiency-seeking transformations, parallel to revolutions such as electrification, globalization or digitalization."
"The new LLM ("Large Language Models") technologies are changing everything, and we must jump aboard the bandwagon with the entire organization as soon as possible".
You may be surprised that the lie is #3, even more so as heads of an AI and advanced analytics consultancy… Allow us to elaborate: companies are clearly facing increasing complexity in assets, people, and resources management. For example, infrastructure networks include increasingly diversified asset classes with varied technologies (fossil generation, renewables, and storage), lives and operating envelopes that are quite different from initial designs, such as CCGTs that nowadays must operate "on demand" to cover demand peaks (instead of baseload).
The growing complexity of energy network infrastructures and, similarly, production structures in each industry, compounded with increasing regulatory and legal requirements, require optimization models that demand more external parameters with faster updates, only manageable with recent technologies and AI techniques. Although most business problems are known and addressable with traditional statistical techniques, innovating efficiency scenarios beyond current local optima requires the application of advanced analytics algorithms. Recent technological advances make it possible to work massive volumes of data through several alternative models (including evolutionary genetic models), at “market speed” without local computing restrictions.
Adding relevant information, such as from macro trends such as electrification, globalization, or digitalization, is a challenge for companies, since internal data is compartmentalized under custodial departments and business units and external data is rarely systematically internalized. In fact, companies must evolve their singular optimization models for each group or class of assets into collective / portfolio ones, which include this multitude of additional internal and external variables (e.g., meteorological data, altitude, atmospheric discharges, corrosion indices, vegetation, etc.). As such, all companies with significant datasets face challenges that can only met through the application of new AI tools, key to evolve optimization models previously intractable due to technological constraints.
Now let us deal with the lie: that new LLM technologies can be deployed "effortlessly with a subscription plan for all", without previous customization work using company-relevant data, prior user training and validation with legal and ethical guidelines. The lie comes from a misunderstanding of current technological limitations and requirements, as well as from misleading marketing from many technology companies and consultancies. It is necessary to demystify the new oracles, and for business users to understand that, for example, makes no sense to implement a generative AI tool without giving it context (that comes from incorporating relevant data and business information) and using critical reasoning to correctly interpret its results. In other words, the hasty and uncritical application of LLM technologies can be neither here nor there and even lead to catastrophic results. These tools require prior work that includes identifying, processing, and integrating relevant data, as well as training users for critical assessment (thus avoiding the so-called automation bias). Applying technology without proper consideration can be counterproductive, as illustrated by customer-service chatbots that cut human cost with appalling results. We have yet to meet someone who is enthusiastic about interacting or arguing with AI agents…
In fact, the current epidemic of LLM tools (such as ChatGPT and Google Gemini) being used as shortcuts without criteria or specific context, has interesting parallels with the initial use (2000s onwards) of web search engines by consultants and data analysts, which produced mediocre and flat results without depth / insights due to the lack of criteria in the choice of sources and data.
From our experience, we say that such models provide little to no value if the results they formulate do not consider each company’s specific context and its organic development and M&A history. For this reason, we propose an approach that focuses first on considering the prioritization of implementing an LLM-type tool with a maximum value / minimum effort logic, within the perimeter of an AI Governance and Compliance strategy, always ensuring human supervision ("human-on-the-loop"). Only after settling the scope come details such as IT infrastructure, data picking, process flows and fine-tuning. With due judgment and reasoning, yes, LLMs can be powerful tools in the service of operational efficiency.
In fact, for energy companies and industries, we appeal to AI professional and conscious rigor – of key importance at a time when the dream of easy access to LLM tools can lead to undiscerning deployment with disappointing results. And there is nothing worse than a bad experience to slow down innovation and ultimately prevent companies from reaping the full benefit of these powerful tools that have the potential to answer many of their challenges.
Article originally published in the energy section of the Spanish newspaper elEconomista, available here
Comments