Trusted Autonomous Systems DCRC
Is GPT-3 a step towards creating artificial general intelligence?
Chair: Associate Professor Kate Devitt – Chief Scientist, Trusted Autonomous Systems
Panel:
• Professor David Chalmers (NYU)
• Professor Susan Schneider (NASA and Florida Atlantic University)
• Professor Marcus Hutter (ANU)
Content
A philosophical discussion on the development of artificial intelligence and specifically advances in Generative Pre-trained Transformer-3 (GPT-3).
GPT-3 is an auto-complete algorithm created by OpenAI as part of their endeavour to develop artificial general intelligence. GPT-3 is the third in a series of autocomplete tools designed by OpenAI. (GPT stands for “generative pre-trained transformer.”). GPT-3 is fed on an unimaginatively large corpus of human knowledge including all of Wikipedia, millions of books, websites and other materials including philosophy texts. In fact, any type of information uploaded to the internet is possible food for GPT-3’s artificial mind to dwell on. The result? Eerily coherent, complex and interesting thoughts about almost any topic. The sophisticated, nuanced text produced by GPT-3 seems to pass the Turing Test for many–including philosophers. Some of GPT-3’s answers are shedding new light on enduring philosophical questions.
Is GPT-3 the beginnings of an artificial general intelligence. Does it create ideas like a human mind, or even better than a human mind? Is human cognition similarly some sort of autocomplete program in our brains? Is it possible that GPT-3 one day becomes consciousness or is it already conscious?–How could we tell. If an AI passes our tests for consciousness, do we then have an obligation to accord it rights? If so, what sorts of rights might it deserve. Independently of rights, how should humans manage an AI that has access to everything that is posited and known and can trick humans into believing that another rational agent is communicating with them?
The panel considers what GPT-3 tell us about the ambition to build an artificial general intelligence, consciousness, human thought and how we should treat AI in an increasingly digital and disembodied world rife with mis- and disinformation.