Researchers design a bot with ChatGPT for the first time

about the episode

Can ChatGPT also design a bot? they asked at TU Delft. Does the model also come up with something useful? What are the problems and risks that we face if we follow the instructions?

Not only did the researchers want to design a robot, they wanted it to actually be useful to us. They chose food supplies as a challenge and in consultation with ChatGPT came up with a tomato picking bot. In collaboration with a Swiss researcher, they set out to work and monitor all design decisions for the model.

The input proved particularly valuable in the conceptual phase, where ChatGPT drew knowledge from other areas of expertise. But the model also contributed to the build, with advice on the best materials and parts. This resulted in a robotic arm that could actually pick tomatoes.

However, this does not mean that ChatGPT can completely take over this design task from now on, the researchers warn. The output can be misleading if left unchecked. AI bots are designed to generate the “most likely” answer to a question, and so risk providing false information and creating biases in the bots. Plagiarism and intellectual property issues are also something to consider.

The researchers will continue their research, which will include looking at the autonomy of AI in designing their bodies.

Read more about research here: ChatGPT is designing its first bot with TU Delft researchers.