thesis.toc

\contentsline {section}{\numberline {1}What is decision theory and why is it broken?}{5}
\contentsline {subsection}{\numberline {1.1}Expected utility}{7}
\contentsline {subsection}{\numberline {1.2}Decision theory paradoxes}{11}
\contentsline {subsubsection}{\numberline {1.2.1}St. Petersburg Paradox}{11}
\contentsline {subsubsection}{\numberline {1.2.2}Pascal's Wager}{12}
\contentsline {subsubsection}{\numberline {1.2.3}Pascal's Mugging}{12}
\contentsline {subsection}{\numberline {1.3}Unifying overvalued paradoxes}{13}
\contentsline {subsection}{\numberline {1.4}HULP problems}{14}
\contentsline {subsection}{\numberline {1.5}Systematically exploiting expected utility maximisers}{16}
\contentsline {subsection}{\numberline {1.6}Negative HULP problems}{18}
\contentsline {section}{\numberline {2}Exploitation}{19}
\contentsline {subsection}{\numberline {2.1}How does exploitation work?}{20}
\contentsline {subsubsection}{\numberline {2.1.1}Formalising exploitation}{20}
\contentsline {subsection}{\numberline {2.2}Should agents avoid having their desires changed?}{22}
\contentsline {section}{\numberline {3}Avoiding exploitation}{25}
\contentsline {subsection}{\numberline {3.1}Agents who don't maximise expected utility}{25}
\contentsline {subsection}{\numberline {3.2}Agents with bounded utility functions}{28}
\contentsline {subsection}{\numberline {3.3}Agents with low-probability cutoffs}{30}
\contentsline {section}{\numberline {4}RNP and principled choices of epsilon}{35}
\contentsline {subsection}{\numberline {4.1}Ideal and real agents}{37}
\contentsline {subsection}{\numberline {4.2}Presuppositions of decision theory}{38}
\contentsline {subsection}{\numberline {4.3}CSH as a lower bound on decision theory}{42}
\contentsline {subsection}{\numberline {4.4}Skepticism and dead hypotheses}{43}
\contentsline {subsection}{\numberline {4.5}Objections, microscopes and black holes}{46}
\contentsline {section}{\numberline {5}Discussion and anticipated objections}{50}
\contentsline {subsection}{\numberline {5.1}Overvaluing St. Petersburg}{50}
\contentsline {subsection}{\numberline {5.2}HULP is an empty set}{51}
\contentsline {subsection}{\numberline {5.3}Dead hypotheses contradict Bayesianism}{52}
\contentsline {subsection}{\numberline {5.4}Lower-bounded \(\epsilon \) ignores grave risks}{54}
\contentsline {subsection}{\numberline {5.5}RNP disallows strict \(\epsilon \) values}{55}
\contentsline {section}{\numberline {6}Evaluation}{57}