ADAPT Experts Discuss ChatGPT - the Limits and the Potential

30 January 2023

OpenAI’s ChatGPT has been grabbing headlines since its launch late last year.  The latest podcast in the ADAPT Centre’s HumanAIse series looks at some of the limitations of this chatbot but also the potential for how it could change our world and access to information.  Show host, journalist and science writer, Claire O’Connell, spoke to ADAPT researchers Dr Robert Ross, Senior Lecturer in the School of Computer Science in TU Dublin, and Dr Benjamin Cowan, Associate Professor in UCD’s School of Information & Communication Studies.

The conversation explored the difference between ChatGPT and what has gone before, and what makes this conversational interface different to previous models.  Speaking about the system, Dr Robert Ross said: “Large language models are not new, they have existed for many years.  The big deal with this system is that it integrates so much information and it gives a more conversational interface”.  Access to information was a key point being made by Ross who highlighted that this system can be used by almost anyone and has the potential to change how we search for information.

The experts also spoke about trust in AI systems.  Large scale language models attempt to construct sentences that make sense but often the content they produce is wrong despite the system appearing quite confident in the delivery of the information.  Speaking about this issue, Dr Ben Cowan said:”These models need to up their game in terms of accuracy but in terms of the naturalness of the language they are creating they seem pretty accurate.  Ultimately it is just a pattern matching exercise and we need to take the human angle to this.” Cowan explained that there can be a blurring of boundaries of what is human and what is machine especially when it is incorporated into a dialogue agent.  “They are not human; they can pattern match but can’t truly understand so it becomes confusing for users”, explained Cowan.

The conversation also explored the ethical and social questions of these interfaces along with the impact of ChatGPT in an educational setting.  

As the area of generative AI becomes more common, we need to reflect and think critically about the journey we are on with it.

To listen back to this podcast and others in the HumanAIse series, check out ADAPT Radio 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.