Tekoälytutkijan nousu

Chat-GPT-pohjaisten avatarien potentiaalin vapauttaminen virtuaalitodellisuuden digitaalisissa kaksoislaboratorioissa

Kirjoittajat

  • Mae Taylor University College London
  • Zaid Muwaffak University College London
  • Matthew Penny University College London
  • Blanka Szulc University of Kent
  • Steven Brown Scott Bader (United Kingdom)
  • Andy Merritt LifeArc
  • Stephen Hilton University College London

Avainsanat:

laboratorio, AI, chatbot, VR

Abstrakti

Artikkeli englanniksi.

Digital twin laboratories, accessible via the use of low-cost and portable virtual reality (VR) headsets, have emerged as an immensely powerful tool for chemical education and research collaboration. Having an immersive environment identical to that of a laboratory can provide scientists with a unique platform in which to plan future experiments, conduct laboratory tours, and train on specialist equipment. However, what digital twin laboratories currently lack is on-hand support of co-workers to assist with tasks such as locating chemicals, aiding with machine set-up, and issuing reminders regarding local laboratory health and safety rules Here we show how this key gap can be overcome with the use of knowledge-loaded Chat-GPT avatars in VR. We trained three different chat avatars to perform specialist functions crucial to working in a laboratory and obtained accurate and useful responses in up to 95% of cases, using a range of evaluation metrics including Human Evaluation, Set-Based F1 Scoring, and BERTScore. Our findings demonstrate the vast potential of this technology in harnessing the capabilities of AI assistants for scientists and enhancing immersive digital twin environments within VR settings.

Cite: Taylor, M., Muwaffak, Z., Penny, M., Szulc, B., Brown, S., Merritt, A., & Hilton, S. (2023). The Rise of the AI Scientist: Unleashing the Potential of Chat-GPT Powered Avatars in Virtual Reality Digital-twin Laboratories. ChemRxiv. https://doi.org/10.26434/chemrxiv-2023-t4vg7

Tiedostolataukset

Julkaistu

2023-11-28