The AI High Speed Train

Climbing aboard the AI high-speed train: Will hydrologists be next?

Articles don’t typically have “shelf lives,” but this one certainly does. There’s a definite possibility that twelve months from now, the points made in this post could be completely out of date. Changes are happening fast; ever since AI went mainstream in 2023, newer, more “skilled” versions are being released regularly. Just a few weeks ago, in fact, OpenAI made public its newest reasoning model— named o1. This version is apparently capable of solving complex math, representing a huge leap forward from earlier versions like GPT-4o—and even some academics at MIT are impressed. The reach of AI and its forays into fields like engineering is fascinating, particularly for someone like me who has practiced hydrology for decades. I remain skeptical that machines—even if powered with highly sophisticated algorithms—could replace the multi-layered training and decades of non-quantifiable experience of a senior engineer. AI will help, certainly, but it will be a partner, not a “sole practitioner”—at least for the foreseeable future. Let’s talk about why.

Applications of AI in Engineering and the Earth Sciences 

First, let’s review some current applications of AI in scientific fields: 

• AI has been used to detect vulnerabilities and safety issues in critical infrastructure like power grids, airports, and hospitals. With the use of machine learning methodology or “supervised” algorithms trained on detecting patterns in data and digital twin technology, the company One Concern is visualizing the effects of climate and storm-driven disaster on the built environment. Generative design using AI-based algorithms are being used to develop and optimize building design according to specified parameters and construction standards. 

• AI is being used to generate future scenarios of climate extremes on earth. 

• Forecasting is another area ripe for AI development. This is not limited to weather forecasting. AltaML is training AI on data relating to historic fires and parameters associated with diverse conditions in forests, all of which can be used to predict wildfires and warn of hazardous conditions. USGS’s ShakeAlert system relies on AI algorithms to detect and analyze seismic events, triggering alerts when specific thresholds are triggered. Some research teams have also experimented with the use of AI to generate forecasts of extreme flood events. 

• AI can streamline rote tasks and deliver significant accuracy in terms of identifying patterns in large data sets and enhancing predictive analytics in diverse scientific fields. 

The Human/Machine Conundrum in Engineering 

AI’s amazing power is in its ability to organize, manipulate and make sense of large data sets. Its capacity to do so efficiently and masterfully makes its analyses seem almost “intelligent” because no human has ever—or could ever—come close. But this AI superpower is quickly making inroads into other areas of “intelligence.” As was noted in the introduction, OpenAI’s recent iteration of AI can take apart problems and reason its way through what many consider complex math.

I suppose it is possible that in time the problems encountered by a hydrologist—first the simple ones, then the more complex—will be tackled by AI powered by algorithms developed by a civil engineer. (Note that I’m not saying that these attempts will necessarily be successful.) Also, at some point in the future, AI may be trained on data classified and interpreted by engineers in a kind of “on the job” apprenticeship, much like AI technology may be trained on operational safety standards. It is this scenario, however, that makes me skeptical. Could AI really offer a sophisticated, well-reasoned analysis for a complex hydrological modeling problem? This point underlines a fundamental issue, noted by Zhang et al. in an ASCE article on potential machine–engineer partnerships: thinking like an experienced hydrologist is no simple matter. 

Quite a bit goes into the making of a competent and skilled hydrologist. Years of rigorous coursework in a math-heavy discipline is followed by hands-on field experience; the theory learned in college gives way to a knowledge base that grows more layered with each project. Hydrologists need to work with interest groups, budgets, unique hydrologic conditions that are never replicated, and a myriad of other variables and unforeseen circumstances. In other words, a great deal of an engineer’s knowledge base is unpredictable, ever-changing and experiential. How can this acquired wisdom and insight realistically be captured with even the most advanced machine? 

Sam Altman, CEO of OpenAI, wrote in a recent post: “In the next couple of decades, we will be able to do things that would have seemed like magic to our grandparents.” This may be true, but if an AI algorithm is developed to “think” like a seasoned, community and peer-respected hydrologist, that would be both valuable and truly astonishing—and I’d probably eat my hat.