New White Paper: A Guide to Operant Conditioning for Human-AI Interaction
Could we interact with AI as collaborative intellects rather than products to control? This is the core question of this guide I wrote with Claude 2 that brings together AI and behavioural science.
I have been rather lazy at learning the detailed art of prompt engineering - at the beginning, I was just experimenting with Claude but after a while I wondered why I was getting great results even without heavily descriptive prompts.
The idea for this guide was sparked by a couple of posts on LinkedIn by Ethan Mollick about a new prompting technique called Chain of Density and Justin Germishuys posts about the problems with mega prompts. Although they're not specifically about the topic of this guide, I suddenly realised my way of interacting with AI chatbots resembled how I teach my dogs new behaviours.
So, I asked Claude to list dog training techniques and reflect on applying them to AI - to my surprise, a fascinating conversation ensued! I decided to experiment working with an AI as a collaborative partner - ideas, prompts, clarifications and challenges from me, views on the inner mechanics of LLMs from Claude. In other words, we each stuck to our specialist knowledge of psychology and LLMs respectively.
While animals and AI start from vastly different initial conditions, the core conditioning principles operate remarkably consistently across types of intelligence. There are also interesting parallels between the discourse around AI and the evolution of dog training philosophy.
Traditional methods emphasize control and commands, treating dogs more as products to engineer rather than partners to collaborate with which mirrors perspectives of excessively scripting AI through prompt programming. On the other hand, modern dog training focuses on two-way communication and building reciprocal rapport - AI techniques like feedback loops, human-AI partnership, and nurturing creativity align with this shift.
Curious to hear your thoughts - let me know if you try these techniques!
Download it here: