AI for Problem Solving - Fall24 Preview PDF

© Mazzelli, 2024-2025 2 © Mazzelli, 2024-2025 3 © Mazzelli, 2024-2025 4 © Mazzelli, 2024-2025 5 © Mazzelli, 2024-2025 © Mazzelli, 2024-2025 AI-ENHANCED PROBLEM-SOLVING © Mazzelli, 2024-2025 10 © Mazzelli, 2024-2025 WHAT A PIECE OF WORK IS A MAN...

© Mazzelli, 2024-2025 2 © Mazzelli, 2024-2025 3 © Mazzelli, 2024-2025 4 © Mazzelli, 2024-2025 5 © Mazzelli, 2024-2025 © Mazzelli, 2024-2025 AI-ENHANCED PROBLEM-SOLVING © Mazzelli, 2024-2025 10 © Mazzelli, 2024-2025 WHAT A PIECE OF WORK IS A MAN! HOW NOBLE IN REASON, HOW INFINITE IN FACULTIES… THE BEAUTY OF THE WORLD, THE PARAGON OF ANIMALS…” - HAMLET, ACT 2, SCENE 2 …STILL SO SMART?  THE ORIGINS OF OUR MISTAKES Possibility #1 Our environment is designed badly Possibility #2 Our minds are designed badly THE GREAT MINDS WHO STUDIED THE LIMITS OF HUMAN REASON Dan Kahneman Amos Tversky Herbert Simon 19 20 APPROXIMATELY 80% OF THE TIMES WE END UP SOLVING THE WRONG PROBLEM! WHAT ABOUT CASES THAT ARE LESS RETRIEVABLE? HOW ABOUT CHANCE, BAD LUCK, OR RANDOMNESS? PAST PRESENT Decline in revenues (Problem Symptom) Decline in revenues (Problem Symptom) ≈ A POOR PRODUCT QUALITY B A 99% B 0.9% D C 0.1% INEFFECTIVE MARKETING STRATEGY 24 E NARROW FRAMING PITFALL When we tackle a complex problem that we superficially understand… It’s tempting to frame it narrowly to make it look like one we’ve worked on before (availability and representation bias). We can then reason by analogy to quickly identify a solution instead of investing in thoroughly understanding the problem. THE (LESS) PHENOMENAL POWER OF HUMAN MIND We are generally overconfident in our ability to estimate and so we collect fewer data than necessary (overconfidence bias) We tend to ignore or dismiss evidence that contradicts our own view, while favoring evidence that confirms our initial predisposition (confirmation bias).  COVID-19 = SEASONAL FLU Jan 22, 2020: “We have it totally under control. It is just a person coming from China. We will be fine.” Feb 10, 2020: “The virus – they are working hard – it is a seasonal flu, looks like in April, when it gets a little warmer, it will go away.” https://www.youtube.com/watch?v=2eB_xCk5ABw&t=259s  CLIMATE CHANGE = ENGINEERING PROBLEM “As a species that’s why we’re all still here: we have spent our entire existence adapting. So we will adapt to this. It’s an engineering problem, and it has engineering solutions.” - Rex Tillerson, former CEO of Exxon Mobil HOW TO AVOID SOME OF THESE PITFALLS OR, AT LEAST, MITIGATE THEIR NEGATIVE CONSEQUENCES? HOW TO MITIGATE BIAS? Modify the decision maker Make people more aware of their unconscious biases through education and the provision of tools and rules 30 RELY ON EVIDENCE TO SYSTEMATICALLY INFORM PROBLEM FORMULATION AND SOLVING… © Mazzelli, 2024-2025 EVEN THOUGH WE RELY ON EVIDENCE TO SYSTEMATICALLY INFORM PROBLEM FORMULATION AND SOLVING… © Mazzelli, 2024-2025 36 TWO APPROACHES TO MITIGATE BIASES Modify the Modify the decision environment maker Make people more aware of their unconscious Create situations in which a bias is either biases through education and the provision of irrelevant or may even be helpful tools and rules 37 AI-enhanced problem solving  1. AI CAN HELP YOU ASK BETTER QUESTIONS Question velocity Question variety Question novelty …so that questions that spark change —“catalytic” questions — can emerge. https://hbr.org/2023/05/ai-can-help-you-ask-better-questions-and-solve-bigger-problems ASK THE AI TO BE SOMEONE ELSE TO CONTEMPLATE DIFFERENT FRAMES (AND SOLUTIONS) 1. Always start a new chat to avoid bias. 2. Frame your prompt by presenting the situation without hinting at a solution. 3. Choose characters or personalities that differ from your own perspective. PROMPTING IS WEIRD. PROMPTING MATTERS.  2. AI AS A CREATIVE ENGINE In the real world, most new ideas do not come from the ether; they are based on combinations of existing concepts. Large Language Models are very good at this, acting as connection machines between unexpected concepts. ASK THE AI TO ASSIST WITH DATA COLLECTION What analytical approach to adopt? What are the relevant variables and why? How to collect the data (Sampling approach)? How to produce relevant evidence from the data (Data analysis)? How to interpret the evidence? ASK THE AI TO ASSIST WITH DATA CLEANSING When it comes to data analysis, ChatGPT's Code Interpreter is an impressive data scientist. It operates at a very advanced level. To illustrate this, let's start with a fun dataset— a public domain list of superheroes and their powers. You can download it if you want to try these steps yourself. You should include an initial prompt with the data, but it can be quite minimal. For example, I used, 'Here is some data on superhero powers, look through it and tell me what you find' 62 © Mazzelli, 2024-2025 THE PERILS OF LARGE LANGUAGE MODELS FOR PROBLEM- SOLVING LLM are still mysterious. Revealing their true nature is urgent and important. LLMs to solve problems and increase productivity, people need to better understand both the successes and failures of these tools. LLMs can confabulate or make up facts that may lead to the wrong decision. LLMs can give advice that is not specific and contextualized. LLMs’ output depends on how questions are framed – and questions can be biased. © Mazzelli, 2024-2025