{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "da05e5fa",
   "metadata": {},
   "source": [
    "# Intelligent Agents and Active Inference\n",
    "\n",
    "- **[1]** (##) I asked you to watch a video segment (https://www.vibby.com/watch?vib=71iPtUJxd) where Karl Friston talks about two main approaches to goal-directed acting by agents: (1) choosing actions that  maximize (the expectation of) a value function $V(s)$ of the state ($s$) of the environment; or (2) choosing actions that minimize a functional ($F[q(s)]$) of *beliefs* ($q(s)$) over environmental states ($s$). Discuss the advantage of the latter appraoch. \n",
    "\n",
    "- **[2]** (#) The *good regulator theorem* states that a \"successful and efficient\" controller of the world must contain a model of the world. But it's hard to imagine how just learning a model of the world leads to goal-directed behavior, like learning how to read or drive a car. Which other ingredient do we need to get learning agents to behave as goal-directed agents? \n",
    "\n",
    "- **[3]** (##) The figure below reflects the state of a factor graph realization of an active inference agent after having pushed action $a_t$ onto the environment and having received observation $x_t$. In this graph, the variables $x_\\bullet$, $u_\\bullet$ and $s_\\bullet$ correspond to observations, and unobserved control and internal states respectively. Copy the figure onto your sheet and draw a message passing schedule to infer a posterior belief (i.e. after observing $x_t$) over the next control state $u_{t+1}$. \n",
    "\n",
    "<img src=\"./i/fig-active-inference-model-specification.png\" style=\"width:500px;\">\n",
    "\n",
    "\n",
    "- **[4]** (##) The Free Energy Principle (FEP) is a theory about biological self-organization, in particular about how brains develop through interactions with their environment. Which of the following statements is not consistent with FEP (and explain your answer):      \n",
    "  (a) We act to fullfil our predictions about future sensory inputs.      \n",
    "  (b) Perception is inference about the environmental causes of our sensations.       \n",
    "  (c) Our actions aim to reduce the complexity of our model of the environment.       \n",
    "  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "92d9e9ea",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Julia 1.5.2",
   "language": "julia",
   "name": "julia-1.5"
  },
  "language_info": {
   "file_extension": ".jl",
   "mimetype": "application/julia",
   "name": "julia",
   "version": "1.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}