www.reducing-suffering.org

Brian Tomasik #fundie reducing-suffering.org

The difference between non-player characters (NPCs)in video games and animals in real life is a matter of degree rather than kind. NPCs and animals are both fundamentally agents that emerge from a complicated collection of simple physical operations, and the main distinction between NPCs and animals is one of cognitive and affective complexity. Thus, if we care a lot about animals, we may care a tiny bit about game NPCs, at least the more elaborate versions. I think even present-day NPCs collectively have some ethical significance, though they don't rank near the top of ethical issues in our current world. However, as the sophistication and number of NPCs grow, our ethical obligations toward video-game characters may become an urgent moral topic.

...

If video games can be seen as "real" in a similar way as our own world, what distinguishes video-game characters from real people and animals? I think it comes down to differences in complexity, especially with regard to specific algorithms that we associate with "sentience." As I've argued elsewhere, sentience is not a binary property but can be seen with varying degrees of clarity in a variety of systems. We can interpret video-game characters as having the barest rudiments of consciousness, such as when they reflect on their own state variables ("self-awareness"), report on state variables to make decisions in other parts of their program ("information broadcasting"), and select among possible actions to best achieve a goal ("imagination, planning, and decision making"). Granted, these procedures are vastly simpler than what happens in animals, but a faint outline is there. If human sentience is a boulder, present-day video-game characters might be a grain of sand.

Digital agents using biologically plausible cognitive algorithms seem most likely to warrant ethical consideration. This is especially true if they use reinforcement learning, have a way of representing positive and negative valence for different experiences, and broadcast this information in a manner that unifies different parts of their brains into a conscious collective. Yet, I find it plausible that other attributes of an organism matter at least a little bit as well, such as engaging in apparently goal-directed behavior, having a metric for "betterness vs. worseness" of its condition, and executing complex operations in response to environmental situations. Many NPCs in video games have some of these attributes, at least to a vanishing degree, even if most (thankfully) don't yet have frameworks for reinforcement learning or sophisticated emotion.

...

Especially in RPGs, some NPCs have explicit representations of their "welfare level" in the form of hit points (HP), and the NPCs implement at least crude rule-based actions aiming to preserve their HP. In some turn-based RPGs like Super Mario RPG or Pokémon, an NPC may even choose an action whose sole purpose is to bolster its defenses against damage in subsequent rounds of the battle. The extent of damage may affect action selection. For example, in Revenge of the Titans (source code), drones select a building to target based on a rating formula that incorporates HP damage:

rating = cost * (damage / newTarget.getMaxHitPoints()) * factor * distanceModifier;

Even NPCs without explicit HP levels have an implicit degree of welfare, such as a binary flag for whether they've been killed. NPCs that require multiple strikes to be slain -- for instance, a boss who needs to be struck with a sword three times to die -- carry HP state information not exposed to the user. They also display scripted aversive reactions in response to damage.

And maybe representations of valuation could be seen more abstractly than in an explicit number like HP. In animal brains, values seem to be encoded by firing patterns of output nodes of certain neural networks. Why couldn't we also say that the patterns of state variables in an NPC encode its valuation? Animal stimulus valuation exists because of the flow-on effects that such valuation operations have on other parts of the brain. So why not regard variables or algorithms that trigger flow-on effects in NPCs as being a kind of at least implicit valuation?

Brian Tomasik #fundie reducing-suffering.org

Is There Suffering in Fundamental Physics?

This essay explores the speculative possibility that fundamental physical operations -- atomic movements, electron orbits, photon collisions, etc. -- could collectively deserve significant moral weight

...

As an example, even a metal ball -- like an animal -- could be said to take in inputs (various forces acting on it, conveyed via gauge bosons and gravitons), integrate those inputs (compute the net force magnitude), and act in response (move in the direction of the force). Information integration, feedback loops, and (at least implicit) optimization among choices are seemingly relevant attributes of agent-like minds but are also rampant throughout mundane physics. An electron often "chooses" the path of least resistance, based on integrating signals about the physical landscape where it lies.
A maglev train initially falls downward due to gravity, but then is pushed back up by magnets, leading to a "happy" equilibrium (dare we say "homeostasis"?) position.

The remainder of this essay takes a more abstract view and proposes general reasons why it's plausible that basic physics could be seen to contain suffering -- perhaps enormous amounts of suffering. It then elaborates on how much I care and whether there are practical ways we could ameliorate the situation.