ADAPT’s Prof. Anya Belz, Professor of Computer Science at DCU, and Prof. Benjamin Cowan, Assistant Professor at UCD’s School of Information & Communication Studies, featured recently on Newstalk podcast Futureproof with Jonathan McCrea to discuss the topic of AI sentience in light of Google’s LaMDA programme. LaMDA is an AI system that has featured prominently in the news recently over questions raised regarding claims from Google engineer Blake Lemoine that it had achieved sentience.
Whether AI is capable of reaching the point of sentience has been a heated topic of discussion in recent years. Lemoine’s claims give rise to the question of at what point can we consider something sentient and does it even matter if a programme achieves sentience if it appears thoughtful, reflective and self-aware?
During the podcast, Prof. Belz provides insight into the two phases state-of-the-art language models, like LaMDA, exist in: the training phase where the model ‘reads’ various texts (e.g. transcripts of film dialogue, news articles, etc.), and the interactive phase where questions can be fed to it as input and it computes a response as the output on the basis of the texts it was trained on. Responses produced by chatbots like LaMDA have now reached a level of sophistication that can effectively mimic human language behaviours giving the illusion of thought.
Prof. Cowan highlights the most impressive part of LaMDA which is its apparent ability to pattern match from vast amounts of data to provide responses that would seem like they could come from a human. He also provides an interesting perspective as to what sentience and consciousness actually are.
Listen to the full podcast here.
Futureproof with Jonathan McCrea is Newstalk’s flagship science programme.
