CITATION — REFERENCE ENTRY
Toward Idealized Decision Theory — Machine Intelligence Research Institute
- Key
- soares2014tdt
- Authors
- Soares, Nate; Fallenstein, Benja
- Issued
- 2014
- Type
- article-journal
- Publisher
- Machine Intelligence Research Institute
Raw CSL JSON
{
"URL": "https://intelligence.org/files/TDT.pdf",
"note": "Technical Report 2014-4",
"type": "article-journal",
"title": "Toward Idealized Decision Theory",
"author": [
{
"given": "Nate",
"family": "Soares"
},
{
"given": "Benja",
"family": "Fallenstein"
}
],
"issued": {
"date-parts": [
[
2014
]
]
},
"publisher": "Machine Intelligence Research Institute"
}
Claims
-
FDT is motivated in part by the problem of building beneficial AI agents: an AI designed with CDT-style reasoning could be exploited by predictors, while one designed with FDT-style reasoning would behave robustly in a broader range of environments.
-
Soares and Fallenstein identify TDT as the starting point of the MIRI decision theory research programme, with UDT and FDT representing successive refinements that address limitations in TDT's handling of logical uncertainty and policy-level reasoning.
Available in