GPT-4V Demonstrates Gameplay in Doom, Raises Ethical Concerns

Microsoft scientist Adrian de Wynter highlights the ease of programming AI without understanding, urging for ethical considerations in AI deployment.

GPT-4V, a variant of GPT-4 with vision capabilities, has been programmed to play the classic video game Doom, demonstrating basic gameplay functions such as opening doors, fighting enemies, and firing weapons without prior training.
Despite its ability to interact with the game, GPT-4V lacks object permanence, forgetting about enemies once they move off-screen and struggling with complex decision-making.
Adrian de Wynter, a Microsoft scientist and researcher at the University of York, raised ethical concerns about the ease of programming AI to perform tasks without understanding or questioning instructions, highlighting potential misuse in society.
The research underscores the need for careful consideration of AI deployment and the development of safeguards to prevent misuse, especially in sensitive areas like military technology and automated testing.
Experts urge the AI community to incorporate protections for humanity into AI models, referencing the fictional nature of Isaac Asimov's 'Laws of Robotics' and the importance of ethical AI development.