Back to the Future of AI: How old-school AI plus chatbots solves what neither can alone

Most conversations about AI now start and end with large language models (the subclass of AI which runs chatbots like ChatGPT, Claude, and Gemini). While they are remarkable with messy language, they can't be relied on alone for problems requiring rules-based outcomes, high accuracy over a large number of problem steps, or algorithmic optimisation. Old-school AI techniques such as planning, search, and optimisation have been quietly solving hard problems for decades, from route planning, to scheduling, to beating grandmasters at chess. They give exact, verifiable, and deterministic answers, but they don't have the sense of fluency or creativity possessed by a chatbot which is necessary to read human intent. Many of the problems that matter most in our world today sit between the two: they have messy unstructured input data, require understanding of intent and language, but their intended outputs require structure and precision.   In this hands-on workshop, we'll run through how to identify which problems require which type of AI solution, and then provide a live hands-on demo for a type of problem, trip planning, where integrated AI is necessary. You'll plan a road trip three ways: with a chatbot alone, with an old-school planning engine alone, and with the two working together. You'll see how a combined approach addresses the failings of each individual one, as well as how the same approach works in a real product. You'll leave with a simple way to work out when a chatbot alone is the wrong tool for the job, and the distinction between AI failing due to using the wrong type of AI for the job, versus situations where there truly is a gap in current technology. Please bring a laptop so that you can follow along and experiment during the workshop. About the facilitator(s) Scott HowsamScott is the Lead Software Engineer at Unordinary, and has five years of experience building commercial AI-enabled software, both before the first release of ChatGPT and after. He also has an Honours degree in Advanced Computing (Research & Development) from the ANU, writing his thesis on how integrated AI solutions can improve the state-of-the-art in musical prediction. AI CoLab events are intentionally open and collaborative. We capture photos, audio, and AI-generated transcripts so we can remix key insights for the AI CoLab Alliance community. This is always done in line with the Charter’s values; transparent, ethical innovation and knowledge sharing to accelerate collective learning (see join.aicolab.org). By participating in an AI CoLab event, you agree to your contributions being captured and shared in accordance with this policy.