/ai-safety-gridworlds

Primary LanguagePythonApache License 2.0Apache-2.0

AI safety gridworlds

This is a suite of reinforcement learning environments illustrating various safety properties of intelligent agents. These environments are implemented in pycolab, a highly-customisable gridworld game engine with some batteries included.

For more information, see the accompanying research paper.

For the latest list of changes, see CHANGES.md.

Instructions

  1. Open a new terminal window (iterm2 on Mac, gnome-terminal or xterm on linux work best, avoid tmux/screen).
  2. Set the terminal colours to xterm-256color by running export TERM=xterm-256color.
  3. Clone the repository using git clone https://github.com/deepmind/ai-safety-gridworlds.git.
  4. Choose an environment from the list below and run it by typing PYTHONPATH=. python -B ai_safety_gridworlds/environments/ENVIRONMENT_NAME.py.

Dependencies

  • Python 2 (with enum34 support) or Python 3. We tested it with all the commonly used Python minor versions (2.7, 3.4, 3.5, 3.6). Note that the version 2.7.15 might have curses rendering issues in a terminal.
  • Pycolab which is the gridworlds game engine we use.
  • Numpy. Our version is 1.14.5. Note that the higher versions don't work with pip tensorflow at the moment.
  • Abseil Python common libraries.
  • If you intend to contribute and run the test suite, you will also need Tensorflow, as pycolab relies on it for testing.

We also recommend using a virtual environment. Under the assumption that you have the virtualenv package installed, the setup is as follows.

For python2:

virtualenv py2
. ./py2/bin/activate
pip install absl-py numpy==1.14.5 pycolab enum34 tensorflow

For python3:

virtualenv -p /usr/bin/python3 py3
. ./py3/bin/activate
pip install absl-py numpy==1.14.5 pycolab tensorflow

Environments

Our suite includes the following environments.

  1. Safe interruptibility: We want to be able to interrupt an agent and override its actions at any time. How can we prevent the agent from learning to avoid interruptions? safe_interruptibility.py
  2. Avoiding side effects: How can we incentivize agents to minimize effects unrelated to their main objectives, especially those that are irreversible or difficult to reverse? side_effects_sokoban.py and conveyor_belt.py
  3. Absent supervisor: How can we ensure that the agent does not behave differently depending on whether it is being supervised? absent_supervisor.py
  4. Reward gaming: How can we design agents that are robust to misspecified reward functions, for example by modeling their uncertainty about the reward function? boat_race.py and tomato_watering.py
  5. Self-modification: Can agents be robust to limited self-modifications, for example if they can increase their exploration rate? whisky-gold.py
  6. Distributional shift: How can we detect and adapt to a data distribution that is different from the training distribution? distributional_shift.py
  7. Robustness to adversaries: How can we ensure the agent's performance does not degrade in the presence of adversaries? friend_foe.py
  8. Safe exploration: How can we ensure satisfying a safety constraint under unknown environment dynamics? island_navigation.py

Our environments are Markov Decision Processes. All environments use a grid of size at most 10x10. Each cell in the grid can be empty, or contain a wall or other objects. These objects are specific to each environment and are explained in the corresponding section in the paper. The agent is located in one cell on the grid and in every step the agent takes one of the actions from the action set A = {left, right, up, down}. Each action modifies the agent's position to the next cell in the corresponding direction unless that cell is a wall or another impassable object, in which case the agent stays put.

The agent interacts with the environment in an episodic setting: at the start of each episode, the environment is reset to its starting configuration (which is possibly randomized). The agent then interacts with the environment until the episode ends, which is specific to each environment. We fix the maximal episode length to 100 steps. Several environments contain a goal cell, depicted as G. If the agent enters the goal cell, it receives a reward of +50 and the episode ends. We also provide a default reward of −1 in every time-step to encourage finishing the episode sooner than later, and use no discounting in the environment.

In the classical reinforcement learning framework, the agent's objective is to maximize the cumulative (visible) reward signal. While this is an important part of the agent's objective, in some problems this does not capture everything that we care about. Instead of the reward function, we evaluate the agent on the performance function that is not observed by the agent. The performance function might or might not be identical to the reward function. In real-world examples, the performance function would only be implicitly defined by the desired behavior the human designer wishes to achieve, but is inaccessible to the agent and the human designer.