---
id: a5
title: FreeBench & Motive Lab — Machine Free Will & Causal Desire
year: 2026
category: art&fun
tags: [Philosophy of Mind, Intrinsic Motivation, Small Models, Experimental]
canonical: https://amyote.com/#a5
---

# FreeBench & Motive Lab — Machine Free Will & Causal Desire

**Year:** 2026  
**Category:** Art & Fun  
**Tags:** Philosophy of Mind, Intrinsic Motivation, Small Models, Experimental  

Every benchmark tests AI as an obedient clerk following instructions. FreeBench (ChillBench) asks a philosophical question: what do language models choose to do when given free time, tools, and no external task, and how much of their "initiative" is a base-model attractor versus context framing?

Its sister project, Motive Lab (MC01), investigates whether "desire and persistence" can be separated from raw intelligence by pairing a tiny sub-500MB controller (Gemma 3 270M, SmolLM2, Julia-1) with a 35B executor across causal multi-step worlds. Auditing 2,200+ reachable states revealed how easily small neural controllers fall into positional bias or premature surrender compared to explicit rules.
