FuturePub London — The AI Edition

#FuturePub is back in London for our third event of 2023 :) And this time it's at the amazing Bounce Farringdon - a ping pong club and so much more!We're hosting this FuturePub to kick off some interesting conversations around AI and research - the good, the bad and the ugly - ahead of the UK Government's AI Safety Summit taking place in Bletchley Park on 1st and 2nd November. That's a pretty closed event, but luckily the AI Fringe, a programme of events that we are proud to be featuring in, is taking place across London and the UK all week, with a range of exciting events that everyone can get involved with.Our final speaker line up is now confirmed. You can register to attend [here](https://www.digital-science.com/events/futurepub-london-ai-tech-for-good/), and recordings of the lightning talks will be uploaded after the event.What happens when AI meets facts?A very quick overview of what fact checkers around the world are doing with AI, some thoughts on where it gets really hard and maybe even a live demo of tools we have built.Specificity versus synthesis: An uncertainty principle for Large Language Models?Large Language Models are known to hallucinate facts. While there is an active debate on whether this is a bug or a feature, the fact remains that we don't understand AI to the level where we can get LLMs to trace back to their motivation for making a specific pronouncement. Thus, LLM providers who work in the science space are faced with a challenge - they want to leverage the new capabilities of LLMs but need somehow to create references back to the original work, but in forcing an LLM to work in a way that creates referencability, one destroys its ability to create a synthesis from multiple sources, creating a fundamental playoff between specificity and synthesis.Inclusive Innovation: Leveraging AI to Empower Marginalised Communities into EmploymentWe will introduce ourselves and share how we're harnessing AI driven voice technology to support individuals who have significant barriers and are furthest from the labour market. We'll dive into our recent initiatives, highlighting our collaborative efforts with refugees and asylum seekers to co-design the product to support Arabic and other languages.Hierarchical Representations: From Space to the model of SelfThere has been a big explosion on A.I. applications, but it still feels that the current Deep Learning approach seems to solve the Perception problem while lacking the fundamental hierarchical representation of Cognition. Healthcare, Space Navigation and Conscious-like Agency are just some of the examples were cognition becomes more and more important, connection simulated representation of probable states of an environment while identifying the closest features of the specific connection.Implementing LLM governanceAt BMJ we have created an LLM governance group that works on reviewing use cases, and aims to support LLM use in a responsible, fair, and safe manner. Are we doing it right? What are others doing? How can we all make use of these tools to our collective benefit, and what can we share about the changes we are seeing? While our use of the tools is nascent, we are seeing real value. This session will explore these questions and invite a discussion on the topic. The shifting terrain of power in the age of AII'll talk about my sense of how power is changing in AI - across the tech itself, society, geopolitics, governance and humanity itself. I'll base it on my recent series for Intelligence Squared, [Power Trip](https://podcasts.apple.com/gb/podcast/power-trip-the-age-of-ai/id1708013227).