ADAPT Radio’s HumanAIse series continues this month with a discussion on conversational artificial intelligent (AI) interfaces, including ChaptGPT, with ADAPT Co-founder Professor Vincent Wade and Professor John Kelleher.
Since ChatGPT burst into the mainstream late last year, conversational AI interfaces have been prompting huge discussion about how AI will shape our futures. This month’s podcast provides a deep dive into the world of Generative AI and Large language Models (LLMs) as well as focusing on how these AI chatbots work and where they can be used to the human advantage. Joining this month’s podcast to discuss is Professor Vinny Wade, co-founder of the ADAPT Centre and who holds the Professorial Chair of Computer Science and a Personal Chair in Artificial Intelligence at Trinity College Dublin and Professor John Kelleher, who is a Professor of Computer Science at Maynooth University.
Kicking off the podcast, Prof. Kelleher provides some background on LLMs as well as the training process involved in training these language models. Prof. Wade also highlights the issue of the quality of language being generated by artificial intelligence (AI). He notes that the language generated by these models is more eloquent and fluent compared to what he has seen before. Potential issues are raised when it comes to the content generated by these models as often, while very convincing, the information generation is fatally wrong. Prof. Wade includes that this type of issue is called a “hallucination” and one such example he gives is asking the model to generate the biography of a well-known individual but the content generated includes incorrect facts about this person resulting in untrustworthy content generation.
Prof. Wade continues the discussion with a look into the potential impacts within the education sector and highlights the importance of viewing ChatGPT, and similar models, as tools rather than something to copy directly from and treat as a fact source. An example he gives is teachers starting to utilise ChatGPT to generate lesson plans on particular topics showing its potential use as a productivity tool. Prof. Wade also raises the importance of being “AI Aware” and the importance of understanding how to use these tools which allows you to understand how best to utilise them while avoiding potential pitfalls. Within the education sphere, he also points out the benefits of ensuring students are transparent when they use ChaptGPT in assignments. Ultimately, it is possible to utilise these tools to do the heavy lifting on gathering information but it is up to the human to critique the content generating and identify the hallucinations.
Following on from this, Prof. Kelleher discusses the discourse around existential threats from AI. He does not necessarily see the potential threat of AI against humanity but, instead, he highlights the challenges and issues that arise where AI is used as an accelerator for people who want to generate disinformation. Both Prof. Wade and Prof. Kelleher highlight the importance of transparency when considering the data sets these models are trained on and being aware of potential bias.
Finally, to close out this month’s discussion, Prof Kelleher looks to the future of generative AI and the research topics he is most excited about, namely developing tools to tackle the spread of disinformation. He is also eager to work with colleagues in ADAPT to engage with stakeholders and policy development so as to support the development of good data governance strategies around the use and deployment of AI in the education sector. From Prof. Wade’s perspective, he aims to focus on how he can utilise personalisation and AI to empower learning for individuals. LLMs provide him with an extra tool in this regard and Prof Wade is also interested in utilising these technologies from a societal perspective which will have the biggest impact on empowering individuals.
For further insight into the interactions of humans and AI, catch HumanAIse on SoundCloud, iTunes, Spotify, and Google Podcasts.
ADAPT Radio: HumanAIse is ADAPT’s newest podcast series providing an in-depth look at the future of AI, automation and the implications of entrusting machines with our most sensitive information and decisions.