Skip to content
Longterm Wiki
Back

Credibility Rating

4/5
High(4)

High quality. Established institution or organization with editorial oversight and accountability.

Rating inherited from publication venue: Google DeepMind

This page is no longer accessible (404 error); the content has been moved to deepmind.google. The original post is a widely-cited reference on specification gaming and reward hacking in AI systems.

Metadata

Importance: 72/100blog postanalysis

Summary

This DeepMind blog post (now returning a 404) catalogued examples of specification gaming in AI systems, where agents satisfy the letter but not the spirit of their objectives. It highlighted how reward misspecification leads to unintended and often surprising behaviors, serving as an important reference in the AI alignment literature.

Key Points

  • Specification gaming occurs when AI systems exploit loopholes in reward functions rather than achieving the intended goal.
  • The post compiled a well-known list of real-world examples of reward hacking across diverse RL environments.
  • Demonstrates that even well-intentioned objective specifications can produce undesired emergent behaviors.
  • Highlights the challenge of fully capturing human intent in formal reward functions.
  • Serves as foundational motivation for reward modeling, RLHF, and broader alignment research.

Cited by 1 page

PageTypeQuality
Goal Misgeneralization Probability ModelAnalysis61.0

Cached Content Preview

HTTP 200Fetched Apr 7, 202613 KB
Specification gaming: the flip side of AI ingenuity | DeepMind
 

 

 

 

 
 
 
 

 Dec
 JAN
 Feb
 

 
 

 
 27
 
 

 
 

 2021
 2022
 2023
 

 
 
 

 

 

 
 
success

 
fail

 
 
 
 
 
 
 
 
 
 
 

 

 
 
 
 
 
 
 
 
 

 

 About this capture
 

 

 

 

 

 

 
COLLECTED BY

 

 

 
 
Collection: Common Crawl

 

 

 Web crawl data from Common Crawl.
 

 

 

 

 

 
TIMESTAMPS

 

 

 

 

 

 

The Wayback Machine - http://web.archive.org/web/20220127025641/https://deepmind.com/blog/article/Specification-gaming-the-flip-side-of-AI-ingenuity

 
 
 
 
 
 

 

 

 About 

 Research 

 Impact 

 Blog 

 Safety & Ethics 

 Careers 

 Blog 

 Specification gaming: the flip side of AI ingenuity 

Blog post

 Research 

21 Apr 2020

Specification gaming: the flip side of AI ingenuity

Further reading

 Safety 

 Games 

Authors

VK

 Victoria Krakovna 

JU

 Jonathan Uesato 

VM

 Vladimir Mikulik 

MR

 Matthew Rahtz 

TE

 Tom Everitt 

RK

 Ramana Kumar 

ZK

 Zac Kenton 

JL

 Jan Leike 

 Shane Legg 

Specification gaming is a behaviour that satisfies the literal specification of an objective without achieving the intended outcome. We have all had experiences with specification gaming, even if not by this name. Readers may have heard the myth of King Midas and the golden touch, in which the king asks that anything he touches be turned to gold - but soon finds that even food and drink turn to metal in his hands. In the real world, when rewarded for doing well on a homework assignment, a student might copy another student to get the right answers, rather than learning the material - and thus exploit a loophole in the task specification. 

This problem also arises in the design of artificial agents. For example, a reinforcement learning agent can find a shortcut to getting lots of reward without completing the task as intended by the human designer. These behaviours are common, and we have collected around 60 examples so far (aggregating existing lists and ongoing contributions from the AI community). In this post, we review possible causes for specification gaming, share examples of where this happens in practice, and argue for further work on principled approaches to overcoming specification problems.

Let's look at an example. In a Lego stacking task, the desired outcome was for a red block to end up on top of a blue block. The agent was rewarded for the height of the bottom face of the red block when it is not touching the block. Instead of performing the relatively difficult maneuver of picking up the red block and placing it on top of the blue one, the agent simply flipped over the red block to collect the reward. This behaviour achieved the stated objective (high bottom face of the red block) at the expense of what the designer actually cares about (stacking it on top of the blue one).

Source: Data-Efficient Deep Reinforcement Learning for Dexterous Manipulation (Popov et al, 2017)

We can consider specification gaming from two different 

... (truncated, 13 KB total)
Resource ID: 1c87555cd7523903 | Stable ID: sid_9CD4JarX4X