Reimagining AI: Investing in a Physically Connected Future

The next wave of AI innovation will focus on bridging the gap between digital and physical realms. By investing in AI that incorporates physical grounding, we can develop accurate digital twins that enhance design cycles and reduce R&D costs. This approach promises transformative advancements across sectors like manufacturing and healthcare. Focusing on AI that understands and simulates the physical world can unlock new opportunities across industries, making it a strategic priority for future investments in AI advancements.

The rapid advancement of artificial intelligence (AI) over the past few years, particularly with the emergence of generative models like GPTs, has transformed various industries and sparked significant interest in future applications. However, a critical gap remains in AI's understanding of the physical world. While previous iterations of AI have thrived on digital data, the next phase of innovation must pivot towards real-world applications that require a deeper comprehension of physical phenomena. This shift is essential for developing more accurate digital twins—virtual representations that can simulate real-world entities and processes based on the laws of physics.

As we explore how to strategically allocate investments in AI, it becomes evident that integrating physical grounding into AI technologies can lead to transformative advancements across multiple sectors, including manufacturing, healthcare, and environmental science. By leveraging robust physics simulations and real-time data from the Internet of Things (IoT), businesses can enhance their R&D processes, reduce costs, and foster sustainable practices. This article delves into the importance of investing in AI that understands and interacts with the physical world, highlighting groundbreaking research and potential applications that can reshape industries.

There has been a surge of interest in AI over the past two years since the release of GPTs. How should we strategically allocate our investments in AI to maximize future growth and innovation?

A major gap in AI has been its limited grasp of the physical world. While the first wave of generative AI was largely digital, fuelled by abundant online data, the next transformative wave will shift towards real-world applications and a deeper understanding of the physical environment.

Language models lack a true connection to the physical world; they possess only an abstract understanding and can’t accurately model physical phenomena. Similarly, image and video models prioritize visual appeal over physical accuracy, focusing more on aesthetics than on true physical validity.This indicates that digital twins require an evolution; they need to be built upon the laws of physics to establish a more accurate and reliable foundation.Physically accurate digital twins accelerate design cycles, resulting in substantial reductions in R&D costs.

The next wave of AI innovation is poised to bridge the gap between the digital and physical realms. While the initial explosion of generative AI focused on creating content and engaging in dialogue, the real potential lies in its ability to interact with and understand the physical world.

Investing in AI that incorporates physical grounding can lead to transformative advancements across various sectors, such as manufacturing, healthcare, and environmental science. By integrating AI with robust physics simulations, we can develop digital twins that not only replicate the appearance of physical entities but also predict their behaviour under real-world conditions. This level of fidelity allows for more accurate modelling of complex systems, which is essential for tasks like optimizing supply chains, enhancing medical devices, or predicting climate impacts.

Moreover, these physically accurate digital twins can drastically reduce design cycles. By running simulations that adhere to the laws of physics, engineers and researchers can iterate on their designs in a virtual space before committing resources to physical prototypes. This capability not only accelerates the R&D process but also significantly lowers costs, enabling businesses to innovate more freely and responsibly.

Furthermore, integrating AI with IoT can create systems that continuously learn from real-time data, refining their models and predictions. This synergy could lead to breakthroughs in smart cities, autonomous vehicles, and energy-efficient systems, ultimately contributing to a more sustainable future.

We would like to put some of the groundbreakingresearch work happening in the intersection of AI + Physics.

• Stanislas Pamela and co-authors just published an article in the scientific journal Computer Physics Communications dealing with the subject. They test the application of a parallel-in-time algorithm for non-linear magneto-hydrodynamic JOREK simulations of the plasma dynamics. To obtain good convergence, AI based surrogate models are employed to precondition the implicit system of equations and reduce computational costs.For all the pretty involved details, refer to the full article online.
• Researchers from Caltech's Center for Autonomous Systems and Technologies and Nvidia have developed a control strategy called FALCON (Fourier Adaptive Learning and CONtrol) for unmanned aerial vehicles (UAVs). This strategy employs reinforcement learning, enabling UAVs to adaptively learn and respond to changing turbulent wind conditions in real time. The goal is to enhance aircraft safety by allowing them to predict and react to disturbances, potentially preventing incidents like the recent turbulence-related injuries on a Singapore Airlines flight. Read more.
• How do we bring AI to scientific modelling? The standard approach has been AI to augment existing numerical simulations. In a new work in CALTECH they show this approach is fundamentally limited. In contrast, using the end-to-end AI approach of Neural Operators to completely replace numerical solvers helps overcome this limitation both in theory and in practice. Read the research paper.
• Yann LeCun has always taken a rational stance on emerging AI models, pointing out their limitations and reassuring the public that fears of AGI and AI overtaking humans are largely unwarranted. In February 2022, he outlined his vision for achieving human-level reasoning in AI, positioning the Joint Embedding Predictive Architecture (JEPA) as a key element of that vision.The fundamental part of LeCun’s vision is the concept of "world models," which are internal representations of how the world functions. He argues that giving the model a context of the world around it could improve its results. Learn more about his work in meta link, followed by an wonderful blog by Turing post where they have explained the architecture and What JEPA can do.

In summary, focusing on AI that understands and simulates the physical world can unlock new opportunities and efficiencies across industries. As we look to the future, investing in this area could yield significant returns, making it a strategic priority for those looking to capitalize on the next wave of AI advancements.

Keywords: AI, Artificial Intelligence, Generative AI, GenAI, Digital Twins, Physical Grounding, Physics Simulation, JEPA

Disclaimer: This article is an opinion of the authors.