Dreaming beyond electric sheep: AI and machine learning

Over the last decade, AI has made huge advances in interpreting complex information learnt from lots of data. First, this transformed image processing, e.g. recognizing photos, then it transformed speech/language processing, e.g. Siri/Alexa, and more recently it has transformed text generation, e.g. ChatGPT. However, to affect the real world directly, the AI needs to be embodied in robots. In this talk, I describe how this may be the next major advance in AI and how it could transform future robotics.

Among the uncountable galaxies in the cosmos, eruptions have been detected in X-rays coming from the centres of just five. These eruptions come from accretion on to supermassive black holes and could give more details of the strong gravity just outside the event horizon. To better understand these eruptions we need to find more of them: enter Machine Learning. Is it possible with just 14 pieces of data to find more as yet un-discovered sources of eruptions?