Discovery visiting professor Stuart Russell sparks debate with provocative keynote at AI and Society event

Before a packed room in Dublin’s National Concert Hall for ‘AI and Society’ on November 8th, AI visionary Stuart Russell likened the current competencies of artificial intelligence to only the very earliest days of aviation.

“I would be of the opinion that we are at the hot air balloon phase; a lot of hot air,” he added.

Conceding that the emergence of large language models (LLMs) such as ChatGPT are “an important step forward”, the professor of Computer Science at UC Berkeley explained that we have “no idea” how such models derive their reasoning and generalisations.

“We are making progress. There are interesting findings every day. Digital neuroscientists, as I call them, are taking giant black boxes and then running experiments trying to figure out how they work.”

But in Russell’s view, “the systems lack a coherent internal flow, which is essential for intelligence”. 

He complained that this unintelligible AI is now “polluting” the information ecosystem, “both in terms of disinformation and manipulation by recommender systems that are changing people's perceptions and preferences”.

Recommender systems are algorithms feeding content to particular internet users. 

On the plus side, he points to AI having “dramatically accelerated the rates of scientific discovery”. He also believes that, over the next decade, it has the potential to “double or triple the rate of learning for almost all children compared to the typical classroom”.

The UCD Discovery visiting fellow’s 50-minute lecture was a provocative jumping off point for the lively panel discussion that followed. 

Moderator and event organiser Dr Susan Leavy of UCD School of Information and Communication Systems was joined on stage by Niamh Hodnett, Online Safety Commissioner of Ireland at Coimisiún na Meán, Prof Barry O’Sullivan, Director, Insight SFI Centre for Data Analytics and Prof Debbie Ging, Professor of Digital Media at Dublin City University.

Hodnett agreed that recommender systems amplify harmful content online, including what she called “toxic beauty”, including unhealthy eating and feeding content, which can be detrimental for children. 

Ging spoke of the proliferation of misogyny online, her research having found that test accounts looking for this type of material were “literally firehosed with quite extreme male supremacist content”. Even male test accounts not actively seeking it were “bombarded”.

The panel discussed how best to put guardrails around harmful online content and whether social media organisations should moderate their own content. 

They debated algorithmic transparency and the possible strengths and weaknesses of the current Digital Services Act, imminent EU AI Act, US President Biden’s lengthy Executive Order on AI Safety and the UK’s recent AI Safety Summit. 

On the issue of military AI, O’Sullivan proposed the creation of an organisation with oversight of what he called “geopolitical, technological peacekeeping”, adding that Ireland “has a great history in that respect”.

AI and Society was organised by UCD Institute for Discovery, UCD Centre for Digital Policy and the Insight SFI Research Centre for Data Analytics. Elizabeth Farries, director of UCD Centre for Digital Policy, opened proceedings, welcoming Ireland’s AI Ambassador Patricia Scanlon who spoke about the challenges of regulating AI ahead of Stuart Russell’s keynote.