Marcus Hutter
As seen in:
Medium,
Wiley Online Library,
arXiv,
MDPI,
The Conversation,
Routledge,
Tech Xplore,
AI Magazine,
Asia Times,
the decoder
Is this you? As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.
Claim your profile
Articles by Marcus Hutter
Statistics > Machine Learning
[Submitted on 28 Sep 2026 (v1), last revised 4 Nov 2026 (this version, v2)] View PDF HTML (experimental) Comments: Subjects: Machine Learning (stat.ML); Artificial Intelligence (cs.AI); Information Theory (cs.IT); Machine Learning (cs.LG); Statistics Theory (math.ST) MSC classes: 68Q32 (Primary) 62A01, 68Q30 (Secondary) ACM classes: I.2.6; E.4 Cite as: arXiv:2609.34458 [stat.ML] (or arXiv:2609.34458v2 [stat.ML] for this version) https://doi.org/10.48550/arXiv.2609.34458
[2512.17086] Value Under Ignorance in Universal Artificial Intelligence
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
[2505.17882] Formalizing Embeddedness Failures in Universal Artificial Intelligence
arXiv:2505.17882 (cs) View PDF HTML (experimental) Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2505.17882 [cs.AI] (or arXiv:2505.17882v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2505.17882 Submission history From: Cole Wyeth [ view email] [v1] Fri, 23 May 2025 13:31:28 UTC (13 KB) Bibliographic Tools Bibliographic Explorer Toggle Bibliographic Explorer () Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?)...
RL, but don't do anything I wouldn't do — LessWrong
by Michael K. Cohen, Marcus Hutter, Yoshua Bengio, Stuart Russell Abstract: In reinforcement learning, if the agent's reward differs from the designers' true utility, even only rarely, the state distribution resulting from the agent's policy can be very bad, in theory and in practice. When RL policies would devolve into undesired behavior, a common countermeasure is KL regularization to a trusted policy ("Don't do anything I wouldn't do").
Modeling the Arrows of Time with Causal Multibaker Maps
1. Introduction The “arrows of time” are natural phenomena that are asymmetric under time reversal. They are responsible for the qualitative differences between how we relate to the past and the future, governing our experience of the irreversible passage of time. To review some important examples, the thermodynamic arrow of time is the observation that entropy increases over time [1,2]. The causal arrow of time is the observation that causes precede effects [3,4].
An Introduction to Universal Artificial Intelligence
An Introduction to Universal Artificial Intelligence provides the formal underpinning of what it means for an agent to act intelligently in an unknown environment. First presented in Universal Algorithmic Intelligence (Hutter, 2000), UAI offers a framework in which virtually all AI problems can be formulated, and a theory of how to solve them.
Die Gefahr einer fortgeschrittenen KI, die ihr eigenes Feedback kontrolliert
Midjourney prompted by THE DECODER DEEP MINDS Podcast Podcast über Künstliche Intelligenz und Wissenschaft DEEP MINDS #12: Künstliche Intelligenz regulieren Verfügbar bei Youtube, Soundcloud, Spotify, Apple, Google und Amazon Wie würde eine KI entscheiden, was sie tun soll? Ein gängiger Ansatz in der KI-Forschung ist das sogenannte „Reinforcement Learning“.
The danger of AI self-reinforcement learning
How would an artificial intelligence (AI) decide what to do? One common approach in AI research is called “reinforcement learning.” Reinforcement learning gives the software a “reward” defined in some way, and lets the software figure out how to maximize the reward. This approach has produced some excellent results, such as building software agents that defeat humans at games like chess and Go, or creating new designs for nuclear fusion reactors.
The danger of advanced artificial intelligence controlling its own feedback
How would an artificial intelligence (AI) decide what to do? One common approach in AI research is called “reinforcement learning”. Reinforcement learning gives the software a “reward” defined in some way, and lets the software figure out how to maximise the reward. This approach has produced some excellent results, such as building software agents that defeat humans at games like chess and Go, or creating new designs for nuclear fusion reactors.
Advanced Artificial Agents Intervene in the Provision of Reward
License Copyright (c) 2022 Michael K. Cohen, Marcus Hutter, Michael A. Osborne This work is licensed under a Creative Commons Attribution 4.0 International License. Authors who publish with this journal agree to the following terms: 1.
Advanced artificial agents intervene in the provision of reward
INTRODUCTION We call an agent advanced to the extent that it effectively selects its output, which we call its actions, in order to achieve high expected utility in a broad set of environments. Since we will likely want advanced artificial agents to operate in environments for which we lack the source code, like the real world, we consider agents acting in an environment that is unknown to them.
[2207.09238] Formal Algorithms for Transformers
arXiv:2207.09238 (cs) Download PDF Comments: Subjects: Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Neural and Evolutionary Computing (cs.NE) Cite as: arXiv:2207.09238 [cs.LG] (or arXiv:2207.09238v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.2207.09238 Journal reference: Latest 2022 version at http://www.hutter1.net/publ/transalg.pdf Submission history From: Marcus Hutter [ view email] [v1] Tue, 19 Jul 2022 12:49:02 UTC (43 KB)...
[2112.14586] Isotuning With Applications To Scale-Free Online Learning
[Submitted on 29 Dec 2021 ( v1 ), last revised 11 Jul 2023 (this version, v2)] Title:Isotuning With Applications To Scale-Free Online Learning Download a PDF of the paper titled Isotuning With Applications To Scale-Free Online Learning, by Laurent Orseau and 1 other authors Download PDF Abstract: We extend and combine several tools of the literature to design fast, adaptive, anytime and scale-free online learning algorithms.
500'000€ Prize for Compressing Human Knowledge
Copying and distribution of this page (http://prize.hutter1.net) is permitted, provided the source is cited. The prize will be paid if the solution reflects the spirit of the contest. In particular decompressors (secretly) receiving any kind of "outside" information are forbidden. Also in order to verify your claim we need to be able to run your executable on our machines within reasonable space and time constraints. This is a privately run and funded contest.
Designing agent incentives to avoid reward tampering
By Tom Everitt, Ramana Kumar, and Marcus Hutter From an AI safety perspective, having a clear design principle and a crisp characterization of what problem it solves means that we don’t have to guess which agents are safe. In this post and paper we describe how a design principle called current-RF optimization avoids the reward function tampering problem. Reinforcement learning (RL) agents are designed to maximize reward.
[1408.3169] Indefinitely Oscillating Martingales
arXiv:1408.3169 (cs) View PDF Comments: Subjects: Machine Learning (cs.LG); Probability (math.PR); Statistics Theory (math.ST) Cite as: arXiv:1408.3169 [cs.LG] (or arXiv:1408.3169v1 [cs.LG] for this version) https://doi.org/10.48550/arXiv.1408.3169 Submission history From: Jan Leike [ view email] [v1] Thu, 14 Aug 2014 00:19:03 UTC (28 KB) Bibliographic Tools Bibliographic Explorer Toggle Bibliographic Explorer () Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps...
Universal Intelligence: A Definition of Machine Intelligence
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them. Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Show More
loading
Actions
Is this you?
As a journalist, you can create a free Muck Rack account to customize your profile, list your contact preferences, and upload a portfolio of your best work.Get in touch with Marcus
Contact Marcus, search articles and posts on X, monitor coverage, and track replies from one place.
Learn more about Muck Rack