Where does AI fit into the future of science?

14 September, 2022

In the mid-20th century, British logician and computer pioneer Alan Mathison Turing developed some of the earliest substantial work in artificial intelligence (AI). In 1935 he conceived the modern computer, an abstract computing machine consisting of a limitless memory and a scanner that moved back and forth through the memory, symbol by symbol, reading what it found and writing further symbols. Sixteen years later, in 1951, the British computer scientist Christopher Strachey wrote the first successful artificially intelligent computer program—a game-playing checkers program that ran on the Ferranti Mark I computer at the University of Manchester, U.K.
Since these pioneering days of AI research, scientists and engineers have increasingly designed computers to "think" by making decisions and finding patterns as humans do. With the continued development of computer-enhanced reasoning, learning, and perception, AI has become more powerful, facilitating discoveries across a wide range of scientific disciplines and enabling researchers to tackle challenges that were previously beyond their reach. And many believe it will fast become a core discipline of the future, akin to mathematics and physics today. The question now is: where does AI fit into the future of science?
The Global AI Summit
Within the Kingdom, many scientists, researchers, investors, policymakers, thought leaders and innovators see the use of AI as a force for positive change. A force that can elevate humanity itself and revolutionize how we approach technology and, in turn, our future. 
Many of these leaders will gather at this week's Global AI Summit in Riyadh as they explore the impact of AI on economic mobility, healthcare, human development, and smart cities. Hosted under the patronage of His Royal Highness Prince Mohammed bin Salman bin Abdulaziz Al Saud, the Crown Prince of the Kingdom of Saudi Arabia, the summit, held from 13 to 15 September 2022, is centered around the theme of "AI Now, AI Next, and AI Never."
Through its own AI Initiative, KAUST aims to become an international leader in AI research, education, and entrepreneurship. This goal aligns with the Kingdom’s vision for a digital transformation powered by AI. And within the PSE Division, several faculty members utilize AI to varying degrees when conducting their research. PSE News recently sat down with a selection of our leading faculty to discuss AI’s place in their research and the role they believe it plays, and will continue to play in humanity’s future development.
How much of a role does AI play in your research—is it essential to your research? Do you believe that you could carry out your research without AI?
Bicheng Yan, Assistant Professor, Energy Resources and Petroleum Engineering: AI plays a significant role in my research. It offers an efficient approach to solving the physics and engineering problems related to the processes of subsurface fluid flow and transport.
Specifically, in terms of computational efficiency and scalability, AI brings a new computational paradigm to solve the problems in my research area. Besides, it is capable of seamlessly integrating a multitude of data in the subsurface, so it also simplifies the collaboration process among different disciplines.
Matteo Ravasi, KAUST Assistant Professor of Earth Science and Engineering: My research is mostly concerned with creating subsurface models from remote data. You can think of this as looking inside a box from a small number of tiny little holes and trying to interpolate the missing information in between. Solving such problems involves two key ingredients: the knowledge of the underlying physical phenomena and geological concepts and rules. Whilst longtime geophysicists have resorted to (geo)statistics as a way to encode this prior geological knowledge, AI, and more specifically machine learning, represents an appealing alternative to such a paradigm. Instead of coming up with models ourselves, we can empower automated algorithms with the help of available geological models and data. 
I can confidently say that AI is a great addition to our toolbox, and we, as a community, are trying our best to find cases where AI can make an impact. However, as with every other scientific tool, it is not always the solution to every problem and should be used with care. All in all, I do not think that all problems in geoscience require AI, but I am very happy to use it as a core component of my research when it fits the problem at hand.
Hussein Hoteit, Associate Professor, Energy Resources and Petroleum Engineering; Chair, Energy Resources and Petroleum Engineering Program: My area of research is related to petroleum engineering, where we develop numerical and experimental methods to optimize oil production by improving recovery factors while reducing cost and environmental footprint.
AI has shown to be a crucial enabler and accelerator for various disciplines in petroleum engineering and geology. Currently, every project I am involved in will have some aspects of AI developments.
Moving forward with the new era of R&D in petroleum energy will require AI, which is expected to push various boundaries and subject-related bottlenecks that have been persisting for a long time. 
Why, how and when do you use AI to help with your research? 
Yan: The application of AI in my research mainly lies in two aspects: (1) It can be used as a surrogate model to predict the fluid flow and transport in the subsurface. Traditional physics-based numerical solvers provide high-fidelity prediction, but their computational cost is often prohibitively expensive. In addition to maintaining decent fidelity, our developed AI surrogate models can offer a computational speedup of 100 to 10000 compared to traditional solvers.
(2) It can be used as an inversion solver to quantify the uncertainties in subsurface formations and to obtain the optimum engineering design. Such inverse modeling tasks require a remarkable number of expensive simulation runs. From our experience, it is feasible to use AI to directly infer the uncertainty parameters and design optimum engineering design, and maximally reduce the number of simulation runs. This makes large-scale real-time reservoir management tangible.
Ravasi: We empower AI when we try to estimate models of the subsurface, embedding our prior geological knowledge into a special family of neural networks called generative networks. But this is not the only way we are currently empowering AI in my group. Another opportunity comes from situations where deep learning has been shown to outperform previously developed algorithms: one such example is denoising—a problem we are interested in removing unwanted contamination from images. 
Over the last year or so, we have been experimenting with great success in applying self-supervised denoising to the so-called seismic deblending problem, the geophysical equivalent of the famous cocktail party problem. Simply put, we are interested in separating the sound waves produced by multiple sources acting simultaneously to reduce the time and environmental impact of geophysical data acquisition campaigns. In this case, we have shown that AI-based denoising can be a very powerful alternative to industry standard approaches based on so-called "signal sparsity." We believe that our approach could become the new standard for this geoscientific application.
Hoteit: Why: AI has been successful in many disciplines, and it has major potential in petroleum engineering and geology, which are about 100 years old. The potential of AI is tremendous.
How: We have developed various AI methods (deep neural networks, LSTM, GANS, and others) to tackle various problems, such as reservoir simulation and optimization using surrogate models, super-resolution related to digital rock analysis, well log interpretation, solving partial differential equations using PINNS, among others.
When: We use AI when needed, i.e., when we see the potential for improvement and impact. AI may not be needed in many problems where we have satisfactory solutions, or the buffer of impact is low. We carefully select the problems that are amiable for AI. We often plan to obtain the final solution, such as the optimizer, in the form of AI, where AI is used in the deployed solution. We work closely with our industry partners (Saudi Aramco), where we deploy AI-powered solutions that have been proven to be very useful (such as reservoir characterization).
What role do you feel AI will play in humanity’s future development (societal, technology-wise, etc.)?
Yan: I believe AI will be an important part of human activity, just like many other revolutionary technologies in history, like electricity, planes, and the internet, to name but a few. As an energy engineer and simulation scientist, I think such technology gradually makes a difference for our world, as it starts being applied to solve global problems related to energy, climate crisis, food etc. 
Ravasi: At its current stage, I am skeptical that AI could replace humans in highly complex tasks requiring multiple reasoning levels. On the other hand, I am sure that when it comes to mundane, highly repetitive tasks, AI will soon become mainstream in any industry (from energy to medicine) as it is already in the IT sector. 
I am very curious to see how much AI will influence our future lifestyle, especially that of younger generations that are used to relying on technology in every aspect of their life. My hope is that humanity will not become lazy just because everything we want can be obtained by touching the screen of our phones; instead, we will use this additional spare time to be more creative and solve long-standing problems in our society.
Hoteit: It is very difficult to predict the limits of AI and how far it can go. However, I am a strong believer that AI alone cannot sail alone. Fundamental science and engineering will be critical to guide AI. In other words, AI cannot discover a law of physics, for example, but it can help to find it.
Also, the success of AI depends on the presence of quality data; without data, AI will not be effective. Therefore, success will require combining fundamental science/engineering with AI.