"Trust Me, I'm AI": A Deep Dive into Trust, Creativity, and Deception

AI is dominating the news. It's all we can talk about and the EU AI Act is attempting to keep society and humans at the forefront. So, what does it mean to trust? How do we trust AI and should we?

I.am.DANI is an art-making, poetry-writing, AI created for the 40th birthday of MSX. DANI, a chatbot from 1986, written in MSX BASIC by Sean Davidson, has come back to life as I.am.DANI creating art and writing poetry. What differentiates the art created by I.am.DANI from AI generated art, is that the art in I.am.DANI is created by teams of students from TU Dublin Computer Science and Game Design. I.am.DANI creates visuals and writes poems live in response to sound, movement in VR or step sequencer MIDI controller (The Arturia Beatstep) that allows the performer to control all aspects of I.am.DANI in real time.

There is a race going on between developing sophisticated Large Language Models(LLMs) which can generate text mimicking human writing and developing detection tools such as Originality.ai, Turnitin, and GPTZero, which can detect such generated text. This talk discusses who is winning and who is losing the race. I provide insight into state of the art techniques for AI-generated text detection and as a bonus also provides a few techniques to fool existing AI detectors.