AI ALIGNMENT FORUM
AF

Wikitags

Information Hazards

Edited by Ruben Bloom, Alex_Altair, et al. last updated 6th Feb 2024

An Information Hazard (or infohazard for short) is some true information that could harm people, or other sentient beings, if known. It is tricky to determine policies on information hazards. Some information might genuinely be dangerous, but excessive controls on information has its own perils. 

This tag is for discussing the phenomenon of Information Hazards and what to do with them. Not for actual Information Hazards themselves.

An example might be a formula for easily creating cold fusion in your garage, which would be very dangerous. Alternatively, it might be an idea that causes great mental harm to people.

Bostrom's Typology of Information Hazards

Nick Bostrom coined the term information hazard in a 2011 paper [1] for Review of Contemporary Philosophy. He defines it as follows:

Information hazard: A risk that arises from the dissemination or the potential dissemination of (true) information that may cause harm or enable some agent to cause harm.

Bostrom points out that this is in contrast to the generally accepted principle of information freedom and that, while rare, the possibility of information hazards needs to be considered when making information policies. He proceeds to categorize and define a large number of sub-types of information hazards. For example, he defines artificial intelligence hazard as:

Artificial intelligence hazard: There could be computer-related risks in which the threat would derive primarily from the cognitive sophistication of the program rather than the specific properties of any actuators to which the system initially has access.

The following table is reproduced from Bostrom 2011 [1].

TYPOLOGY OF INFORMATION HAZARDS
I. By information transfer mode
 Data hazard 
Idea hazard
Attention hazard
Template hazard
Signaling hazard
Evocation hazard
II. By effect
 TYPESUBTYPE
ADVERSARIAL RISKSCompetiveness hazardEnemy Hazard
Intellectual property hazard
Commitment hazard
Knowing-too-much hazard
RISKS TO SOCIAL ORGANIZATION AND MARKETSNorm hazardInformation asymmetry Hazard
Unveiling hazard
Recognition hazard
RISKS OF IRRATIONALITY AND ERRORIdeological hazard 
Distraction and temptation hazard
Role model hazard
Biasing hazard
De-biasing hazard
Neuropsychological hazard
Information-burying hazard
RISKS TO VALUABLE STATES AND ACTIVITIESPsychological reaction hazardDisappointment hazard
Spoiler hazard
Mindset hazard
Belief-constituted value hazard 
(mixed)Embarrassment hazard
RISKS FROM INFORMATION TECHNOLOGY SYSTEMSInformation system hazardInformation infrastructure failure hazard
Information infrastructure misuse hazard
Artificial intelligence hazard
RISKS FROM DEVELOPMENTDevelopment hazard 

See Also

  • Computation Hazards

References

  1. Bostrom, N. (2011). "Information Hazards: A Typology of Potential Harms from Knowledge". Review of Contemporary Philosophy 10: 44-79.
Subscribe
1
Subscribe
1
Dangerous Knowledge
Discussion0
Discussion0
Posts tagged Information Hazards
25Don't Share Information Exfohazardous on Others' AI-Risk Models
Thane Ruthenis
2y
1
53Conjecture: Internal Infohazard Policy
Connor Leahy, Sid Black, Chris Scammell, Andrea Miotti
3y
2
23Needed: AI infohazard policy
Vanessa Kosoy
5y
13
42The Fusion Power Generator Scenario
johnswentworth
5y
10
47Some background for reasoning about dual-use alignment research
Charlie Steiner
2y
6
35Principles of Privacy for Alignment Research
johnswentworth
3y
20
26Gradient Descent on the Human Brain
Arun Jose, gaspode
1y
0
18AI Safety via Luck
Arun Jose
2y
0
Add Posts