‘We need the bravery to change the process’: How AI is impacting physical environments

In recent years, AI-driven workflow automation platforms have gained prominence within the artificial intelligence sector, with startups developing tools that can plan and execute tasks. Recent data suggests a notable shift towards physical and scientific applications, as experts highlight the expanding role of AI in the physical environment.

European advanced materials startups have attracted €3 billion in investments this year, a substantial increase from the €1.6 billion raised throughout all of the previous year. Similarly, drug discovery startups have raised €4 billion this year, on track to surpass the €4.7 billion collected in 2021. Many of these emerging companies are already incorporating AI, and investor interest is expected to rise as the technology progresses.

Experts emphasize that AI's true potential lies in its application to real-world environments, including science and computing. Chad Edwards, co-founder and CEO of CuspAI, explains that AI should be directed towards addressing global challenges such as climate change and disease. "A large portion of economic activity is rooted in the physical economy, and the potential for growth is far greater than we've witnessed in the language domain," he asserts.

Historically, AI's implementation in physical environments has been constrained by slow, expensive experimental processes. "Software allows for rapid code production, execution, and result observation," Edwards notes. Conversely, physical experimentation and testing take longer. To overcome this challenge, AI must not merely identify statistical patterns in data but must "understand" the fundamental laws of physics.

In materials science and biology AI modeling, this requires incorporating physical constraints into neural networks.

Anthony Bradley, co-founder and chief scientific officer at DaltonTx, emphasizes the necessity of integrating physical constraints directly into AI models. For instance, when developing generative capabilities for antibody design, the AI needs to comprehend the structure and limitations of antibodies to generate viable designs within that domain.

Experts agree that breakthroughs require a combination of strong scientific domain knowledge, robust AI capabilities, and exceptional engineering skills. Bradley stresses that the systems being developed are based on intricate scientific problems, necessitating a multi-faceted approach.

However, transferring AI from controlled lab settings to the physical world presents several hurdles. Iraia Ibarzabal, chief growth officer at quantum AI company Multiverse Computing, points to infrastructure as a primary obstacle. "A system that functions in a laboratory may not perform optimally in real-world conditions," she observes. To address this issue, organizations must establish new workflows incorporating machine intelligence from the outset.

Bradley emphasizes the importance of rethinking processes to fully harness AI's transformative potential. "If we continue with our existing practices and merely add a layer of AI, we will not witness substantial change. We must have the courage to alter our approach," he asserts.

This change involves designing AI technology around the realities of end-users. Organizations should define the desired end solution during the project's initial stages to ensure the resulting product can function effectively in real-world environments. By doing so, AI becomes a valuable tool that empowers scientists, freeing them from labor-intensive tasks such as literature scanning, data curation, and trial-and-error testing.

The value of AI is generally gauged by its ability to deliver tangible outcomes in various fields, including materials science, drug discovery, and computing. Bradley warns against overestimating AI's ability to generate genuinely novel ideas without data. While AI excels at combining existing data, its ability to generate novel insights diminishes when venturing beyond that realm. As such, AI strategies should focus on impactful problems and solutions that genuinely improve outcomes.

Looking ahead, Edwards and Ibarzabal envision a future where AI becomes an almost imperceptible layer in our physical surroundings. In computing, Ibarzabal predicts significant advancements will emerge from deploying AI on physical edge devices, especially in real-life settings devoid of constant cloud access. "This will enable AI to operate seamlessly in resource-limited environments, such as defense applications and industrial equipment," she explains.

Similarly, Edwards anticipates that over the next five years, the primary metric of AI success will no longer be server metrics but the physical impact of AI-driven devices and equipment. "Imagine standing in the world and pointing to a device and saying, 'this is powered by our materials,' knowing that it's making a positive contribution to the world," he muses. "That's the ultimate goal, whether it's enhancing solar technology or other transformative applications."