Artificial intelligence has made significant strides in various creative fields, from generating poems to writing essays and even composing books. The latest development in AI has inspired researchers at TU Delft and EPFL, the Swiss technical university, to explore its capabilities further.
Their focus? Designing a robot.
In a groundbreaking study published in Nature Machine Intelligence, the researchers investigate the potential of AI, specifically the ChatGPT platform, in the design process and the associated risks.
Cosimo Della Santina, assistant professor at TU Delft, joined forces with PhD student Francesco Stella from the same school and Josie Hughes from EPFL to pose a critical issue to ChatGPT: What are humanity’s greatest future challenges? They wanted to move beyond simple chores and asked ChatGPT to create a truly helpful robot. After careful deliberation, they decided to address the issue of food supply by developing a tomato-harvesting robot.
The researchers relied on ChatGPT’s comments and direction throughout the design phase. Stella emphasises the importance of ChatGPT’s contribution during the conceptual phase when the AI expanded the designers’ knowledge into foreign realms. ChatGPT, for example, provided insights into which crop would be the most economically viable for automation. During the installation phase, the AI system also provided practical advice, such as employing a silicone or rubber gripper to reduce tomato damage and a Dynamixel motor for efficient robot movement. The end outcome of this human-AI collaboration is a highly proficient robotic arm capable of harvesting tomatoes.
While the researchers found the collaborative design process to be positive and enriching, they also noticed a shift in their roles as engineers. The study delves into the different levels of cooperation between humans and Large Language Models (LLMs), including ChatGPT. In an extreme scenario, the AI would entirely dictate the design process, with humans blindly following its instructions. In such cases, the LLM acts as the researcher and engineer, while the human assumes the role of a manager responsible for specifying design objectives.
However, it is important to consider the risks associated with unchecked AI-generated output. Della Santina warns about the potential for misinformation and bias in the robotic field if LLM outputs are not thoroughly verified and validated. Working with LLMs also raises concerns about plagiarism, traceability, and intellectual property rights.
Della Santina, Stella, and Hughes intend to continue their robotics study in the future, using the tomato-harvesting robot as a starting point. They also plan to look deeper into AI systems’ autonomy in building their own bodies. The researchers are well aware of the difficult balance that must be struck between using LLMs to aid robot creators and retaining the crucial parts of creativity and innovation required for robotics to address the challenges of the twenty-first century.
As AI advances, the researchers’ work offers light on the potential of AI in the design process while emphasising responsible use and addressing the hazards that come with it.