AI and Society Seminar Series

These seminars consider AI's social impacts in a range of areas, and discuss how AI can best be overseen to maximise positive impacts and minimise negative ones.

They are open to anyone who is interested in attending. We don't presuppose any technical background.. if we present AI systems, we do it 'from scratch', in ways that should be accessible to everyone.

Discussing AI's social impacts is a very interdisciplinary task, and our speakers and participants come from many different backgrounds, in academia, government, industry and NGOs.

The series organiser is Ali Knott ( please email Ali if you'd like to be added to the mailing list.

Details of previous years' seminars can be found here: 2023

Seminars are at 4:00-5:30pm (unless otherwise specified).

1 March (Rutherford House, RH1209/1210)
This seminar will provide an overview of the EU Artificial Intelligence Act, a landmark proposal poised to become the first comprehensive regulation of AI globally. We will discuss the Act's scope, detailing who will be affected and the exceptions that apply, such as for open-source initiatives. The presentation will further explore the Act's categorization of AI systems, highlighting the key responsibilities and considerations for each category. We will also delve into the enforcement strategies, and the anticipated challenges and complexities in implementing the Act, aiming to provide an understanding of its potential impact on the global AI landscape. Additionally, we will address the stance that New Zealand could adopt in relation to this regulation.

8 March (Rutherford House, RH1209/1210)
This seminar will provide an update on new developments in AI over the last few months. We'll cover:
- Google/DeepMind's Gemini multimodal generator (Pro 1.5, Ultra and Nano versions);
- OpenAI's Sora (which produces videos from text prompts);
- Google/DeepMind's Alpha Geometry (which solves geometry problems at Olympiad level);
- Microsoft's Copilot generator (now integrated into many Microsoft products), with some use cases for GPT;
- 1X's EVE humanoid robot, and the new Rabbit personal companion / operating system (both powered by 'large action models').

15 March (Rutherford House, RH1209/1210)
The EU's AI Act imposes some new obligations on generative AI companies, to support detection of the content their generators produce. Biden's Executive Order on AI also makes some requirements on companies in this area. In this seminar I'll introduce these new obligations, and discuss how they could be met. I'll also flag some promising-looking methods for identifying AI-generated images, which seem to work reliably (at present) even without support from companies.

22 March (Rutherford House, RH1209/1210)
  • Tom Barraclough (Brainbox Institute): Regulatory responses to AI in New Zealand
Like other countries, New Zealand is considering how it can maximise the benefits and minimise harms of AI technologies and the way they’re deployed. But New Zealand has some unique characteristics too and it’s likely we’ll require a broader regulatory response than a single piece of legislation. Tom will speak to the EU AI Act as an international example, and outline where initiatives like the NZ AI Policy Tracker can play a role in our domestic response.

29 March: Good Friday, no seminar!

5 April: Mid-semester break, no seminar!

12 April: Mid-semester break, no seminar!

19 April (Rutherford House, RH1209/1210)
  • David Talbot (Talbot Mills Research): Making myself redundant: A market researcher's experiments with gen AI

26 April (Rutherford House, RH1209/1210)
  • Jess Robertson (Chief Scientist, High Performance Computing and Data Science, NIWA): Making decisions in flux: challenges for the use of AI in policymaking and regulatory stewardship
We often approach AI in governmental decision making from an assumption of a fixed decision-making framework (eg existing legislation, an existing regulatory system). This is a useful assumption from an ML perspective because it lets us optimize our approach for nice mathematical metrics like predictive accuracy rather than fuzzy concepts like ‘fairness’. However, the reality is many of our regulatory systems are trying to achieve other aims, including transparency or timeliness of decision making, or early signalling of potentially challenging decisions to give regulated parties time to adjust and accept outcomes in which they might be worse off. These are challenges that ML and AI can help with but I will argue in this talk that it will require a better dialogue between the tools we create and the frameworks we use for decision making.

3 May (Rutherford House, RH1209/1210)
  • Sean Audain (Wellington City Council) and Jocelyn Cranefield (School of Information Management, VUW): AI for civic consultation and communication.
Sean will talk about 'Using AI to connect the city with its people: opportunities and challenges'. Jocelyn will talk about the 'Giving voice to the City' project.

10 May (**Note different venue! Rutherford House Mezzanine floor, RHMZ54**)
  • Ali Knott and Markus Luczak-Roesch (Engineering & Computer Science / Information Management, VUW): Should foundation AI models be open-source or closed-source? The case for public ownership of AI infrastructure
In this seminar, we'll start by reviewing the emerging debate between proponents of 'open-source' and 'closed-source' generative AI models, pitting the newly-founded AI Alliance against the scarcely-older Frontier Model Forum. This is largely a debate between big tech companies; we'll argue that some of the some of the dilemmas in this debate are minimised if large AI systems are owned and administered in the public domain. We'll discuss a few examples, relating to governance of AI compute power, and the stewardship of personal data.

17 May (Rutherford House, RH1209/1210)
  • Chris Cormack (Catalyst Ltd): Data sovereignty: Why it matters in Aotearoa NZ

    Royal Society Te Apārangi's recent report, Mana Raraunga Data Sovereignty, gives an overview of the concepts of data sovereignty, Indigenous data sovereignty, and Māori data sovereignty. In this digital era, when data has become hugely valuable and a source of power, these concepts are helping guide answers to questions about who owns, controls, and protects our data. The report also summarises new data practices that are emerging to create a data future that benefits us all.

24 May (Rutherford House, RH1209/1210)
  • TBA
31 May (Rutherford House, RH1209/1210)
  • TBA