Dutch scientists have collaborated with the AI platform ChatGPT to design a robot capable of harvesting tomatoes. The team, consisting of researchers from Delft University of Technology and the Swiss technical university EPFL, engaged in a chat with ChatGPT on future challenges for humanity before settling on food supply as their topic of choice. This led to the decision to create a robot capable of harvesting tomatoes. During the design phase, ChatGPT provided valuable insights, including what crop would be most economically valuable to automate and the best type of motor to drive the robot. During the implementation phase, ChatGPT also suggested beneficial ideas, leading to the creation of a robotic arm capable of harvesting tomatoes.
In their report, published in the journal ‘Nature Machine Intelligence’, the scientists highlight the positive and enriching nature of the collaborative design process with AI. One potential risk of using AI output, however, is that it could be misleading if it is not verified or validated. Large Language Models (LLMs) such as ChatGPT generate the ‘most probable’ answer to a question, meaning there is a potential for misinformation or bias in the robotic field. Other issues that arise from using LLMs include plagiarism, intellectual property and traceability.
Despite these challenges, the researchers are continuing their study of LLMs to design new robots. They are particularly interested in studying how AI can be used to help robot developers without limiting the creativity and innovation needed to develop robots capable of meeting the challenges of the 21st century.
Ultimately, the LLM-driven collaboration with ChatGPT has provided a wealth of insights that the team will continue to utilize in their future research.
