AGI Study Group
Dates: various (2010-2011)
Venue: Littlegate House, Oxford
FHI are to hold ongoing AGI study groups (by invitation only). Future dates to follow in due course.
8 March 2012: Multipolar non-WBE outcomes
1 March 2012: Disperse Global Cognition (Pt. 2)
27 February 2012: The “Hail Mary” method; Dispersed Global Cognition (Pt. 1)
24 February 2012: Genetic engineering and cognitive enhancement; Solmonov induction
20 February 2012: Indirect Normativity and Finding Earth (Pt. 2)
10 February 2012: AGI general strategies (Pt. 2); Indirect Normativity (Pt. 1)
27 January 2012: AGI general strategies; Oxford Protocol start
11 November 2011: Internal lecture by Dr Eric Drexler on Nanotechnology Timelines
2 November 2011: What is it like to be AIXI?
26 October 2011: Multipolar outcomes
20 October 2011: Value-learning and "Hail Mary" solutions
28 June 2011: Value-loading and agency (3.30pm, followed by optional dinner)
15 June 2011: Agency: Internal and External Institutions (3.30pm)
8 June 2011: Agency: Internal and External Institutions (4.30pm, optional dinner afterwards). Please refer to list of questions to be disussed at meeting.
26 May 2011: Boot-strapping and Value-Loading (3.30pm, followed by optional dinner)
5 May 2011: Anthropics and decision theory
15 April 2011: Jaan Tallinn’s Ideas and Questions
1 February 2011: New argument concerning simulation hypothesis; and evaluating SIA-DA argument
26 January 2011: Multipolar scenarios
25 January 2011: Timeless decision theory and anthropics-confused AI
24 January 2011: Oracle AI safety
23 January 2011: Schema for AI motivation
26 November 2010: Don’t wait for Moore’s Law: Statistical Emulation for Computer Simulation. With introductory talk by Bela Nagy.
Abstract: The statistical principles of designing and analysing experiments have had an enormous impact on both science and business after R. A. Fisher published his seminal book in 1935 (The Design of Experiments). Nowadays no serious investigator would consider running an agricultural or industrial experiment without using these principles. But how about virtual experiments? How should one experiment in a virtual world that only exists in a computer? Ironically, in this case virtually no one would call a statistician for help in spite of the fact that such questions have been thoroughly investigated in the past 20 years or so and now are well understood in a specialised subfield of statistics (The Design and Analysis of Computer Experiments). In my talk I will show how one can build a statistical meta-model (emulator) for a computer model (simulator), resulting in an instantaneous software speedup that can span many orders of magnitudes. Furthermore, statistical emulation also provides confidence intervals that can precisely quantify the uncertainty about such approximations. In many cases, this can potentially remove the computational bottlenecks strangling the development of computationally expensive simulations because this is a software trick that doesn't require any hardware upgrades. We don't have to wait for Moore's Law.
Readings: A brief overview of the Design and Analysis of Computer Experiments can be found here
12 November 2010: Indirect normativity approaches
Readings:
1. A recent review of the CEV approach and its rationale by Nick Tarleton and Roko Mijic
2. Eliezer Yudkowsky’s original paper
26 October 2010: Subject: "The Treacherous Turn"
19 October 2010: Intelligence and motives chapter of Nick's book and The Basic AI Drives by Omohundro.
Readings:
1. The Nature of Self-Improving Artificial Intelligence by Stephen M. Omohundro
2. The Basic AI Drives by Stephen M. Omohundro
15 October 2010: Continuation of the Schmidhuber discussion.
23 September 2010: AIXItl Paper "Towards a Universal Theory of Artificial Intelligence based on Algorithmic Probability and Sequential Decisions" by Marcus Hutter
Readings:
1. Hutter's paper (the pages describing the method are 9-10)
2. Ultimate Cognition à la Gödel by Jürgen Schmidhuber
