Somewhat speculative, but reviewing a lot of science, including the modelling results out of our own group. Author’s final version, from July 2020.
Read MoreThis is the first paper unifying our transparency work, our games work, and our work on cultural variation in human cooperation and anti-social punishment. Hopefully there will be a bunch more in 2020 and 2021. Camera ready from July 2019 or so. Software on github, linked from the AmonI Software Page.
Read MoreNot directly AI policy, about biological innovation. Basically, mutation really can be a factor of innovation, because compensating for things that go wrong is a lot easier than you thought. We’ve been working on this for years, it’s finally submitted somewhere — that fortunately encourages biorxiv use. See also my papers with Yifei below. Version from December 2019.
Read MoreHighly relevant to information technology policy: The more you know, the less you need to trust, though if you know nothing or don’t have a choice of who you work with, people have no reason to be trustworthy. Trust comes with PARTIAL information, AND at least some freedom. Open access because Bath+Springer
Read MoreA summary of everything biology and biological anthropology have to say on the subject, for the benefit of roboticists in particular. Open access is as of late November 2014, a lightly updated version is now available in the Handbook.
Read MoreBe sure to also look at the supplement, which gives the stimuli and shows similar results for a different corpus and word-embedding model. Meaning really is no more or less than how a word is used, so AI absorbs true meaning, including prejudice. We demonstrate this empirically. This is an extension of my research programme into semantics originally deriving from my interest in the origins of human cognition, but now with help from the awesome researchers at Princeton I’ve merged this with my AI ethics work, and also managed to pitch for cognitive systems approaches to AI. Open access version: authors’ final copy of both the main article and the supplement (pdf).
Read MoreOpen access version, authors’ final copy (pdf).
Read MoreThere’s a tradeoff between how fast gossip spreads vs problems with its potential for corruption: it can be a lot more useful than direct experience if it spreads faster than that experience and there isn’t too much false information. Actually, in the real world gossip may give you more information than your perception, but that’s not one of the things we deal with here. This work was actually done prior to (and informed) our 2015 article Value Homophily Benefits Cooperation But Motivates Employing Incorrect Social Information, but took longer to get out for a few reasons. Open access draft.
Read MoreI made unacknowledged contributions to this part introduction too. Open access version (pdf).
Read MoreI made unacknowledged contributions to this part introduction too. Open access version (pdf).
Read MoreReviews Paul’s awesome work on understanding why it’s adaptive to both have blind faith in others in your community, yet refuse to cooperate with those making unfair but mutually-beneficial offers.
Read MoreThis brings together all three threads of my research: action selection, natural cognition and collective behaviour, and the mischaracterisation of AI as an active threat. In response to the apocalyptic futurism typified by Bostrom’s Superintelligence, I frame AI as an ordinary part of human culture, which for 10,000 years has included physical artefacts that enhance our cognitive capacities, and is apocalyptic enough in its own right. Open access: here’s the post-review submitted version from September 2014, or email me for the corrected final.
Read MorePreferring to cooperate with those with a similar cooperation style supports the evolution of cooperation. Reputations spread through gossip supports this strategy. But now that you are spreading two kinds of information (reputations of others, and your own style of cooperation) you can have a conundrum when these conflict. When there is such a conflict, signalling honestly about your cooperation strategy can be more beneficial to your community than telling the truth about someone else. Free open access draft is here. Software is coming soon. Draft is from October 2014, yet the work originally derived from On the Reliability of Unreliable Information: Gossip as Cultural Memory, which came out in 2016. Such is academia.
Read MoreA commentary on Paul Smaldino’s The Cultural Evolution of Emergent Group-Level Traits. Open access version.
Read MoreThis book follows from a workshop. Here is a free version of the chapter, the revised draft from May 2013. See further our Cultural Variation in Costly Punishment project page. Note: Google Scholar managed to find a USAF white paper derived from our final report by the same title which has a lot of irrelevant detail and a couple theoretical errors we’ve since discovered. The book chapter is 15 months more recent.
Read MoreInteresting for both biology & machine learning, looks at the role of a potentially-hierarchical network representation in the genome for handling semi-predictable environmental change. Final version is open access, because computer science has archival conferences.
Read MoreSome simple simulations of culture and modularity showing interesting stability effects, inspired by a talk Dan Sperber gave in 2008. Open access draft from 2010. Open source netlogo model described in the paper on the AmonI Software Page.
Read MoreA chapter length description of our attempt to use social media as a source for a more accurate portrayal of the space of human emotions. Derived from Eugene Bann’s undergraduate dissertation. A more recent paper by the same authors on a related topic came out in 2013 ...
Read MoreA review article; see further our Cultural Variation in Costly Punishment project page. If you don’t have access to APA, here is the revised final version submitted to the publisher (May 2013).
Read MoreConsiders the most common “emotion” keywords on Twitter, and discovers that some concepts, e.g. sleepiness and sadness, are relatively culturally invariant, but others like “surprise” and “stressed” seem to be used quite differently in different global regions. Also, Europeans are the most positive and excited tweeters. Camera-ready from April 2013.
Read MoreCan the digital revolution lead to permanent memorials, or even a sort of Internet-based immortality? By reviewing the nature of human intelligence, this article shows that an individual’s memory and ideas cannot operate directly without the human body in which they were developed. Further, there is no meaningful way in which technology can allow indefinite access to our culture. However, our lives and actions do influence the thoughts and actions of those who are living, and the Internet can enhance this influence even after our death.
Read MoreAuthor’s final version (pdf) from August 2009 (I only found out this ever got published in 2017.)
Read MoreI’m particularly pleased about this paper because it shows clearly how models can advance even well-established social-scientific theories provided that we work directly with domain experts who really understand the theory and data. There are some very pithy, quotable text about this in our response to commentaries, From the Imaginary to the Real: The Back and Forth Between Reality and Simulation. Open access pre-proof version of the target article, and of the response to commentaries. Associated software is available from the AmonI software page, and also in the electronic appendix. Oxford Anthropology have made a web page about our simulation of religion work.
Read MorePenultimate version on arXiv. This paper shows that punishment alone can’t explain altruism, the papers that thought it could didn’t take into account the well-documented behaviour of anti-social punishment. Basically, some people punish those that contribute to the public good. This is the first article of at least five we expect to publish explaining this phenomenon, and why it varies by culture. See our Cultural Variation in Costly Punishment project page.
Read MoreThe role is not so much for immediate selection, but for updating models for future selection. In case you don’t have access, here’s the submitted draft from July 2012.
Read MoreAn invited chapter for a book written in honour of the late Brendan McGonigle. The chapter mostly takes a neuro and psychological approach, but last section is Eco Evo Devo, with some ideas I’ve been working on lately on the origins of cognition. This is the draft sent to the publisher in March 2010.
Read MoreThe discussion is updated from our 2007 PTRS-B article, though the models are not. Penultimate draft from 2010, related (including improved) software is available here.
Read MoreAn expanded and updated version of our 2007 PTRS-B special issue which was a condensed version of our 2005 conference proceedings. I think this is the last version of this work we’ll see since we’ve now run through all the primary permutations for the order of the editors (see below).
Read MoreThe paper is final from April 2011.
Read MoreThis is an update of the AISB update of the Vienna Consciousness paper. The next step should be a journal article. The title is a reference to a Dennett quote well worth knowing. The paper claims we already have conscious robots and it’s not that big of a deal. It also puts forward some cool ideas about the functional role of the action-selection-related process that we experience as consciousness.
Read MoreDiscriminates the size of a culture (how much information can be transmitted from one generation to the next) from its extent (how much useful behaviour can be generated) and argues that the vast majority of cultural ratcheting is because the size of human culture finally got large enough that cultural evolution could start increasing its extent.
Read MorePenultimate draft, from May 2009. Related software and articles.
Read MoreThe paper is a scientific consequence of the ideas put forward in “Crude, Cheesy, Second-Rate Consciousness” (see below), and the work I am doing on understanding the evolution of cognition. It concerns the tradeoffs between individual and genetic learning, and whether these may be shifted on the basis of individual experience over an agent’s life history. Evidence is derived from models of macaque task learning. Camera ready from April 2009. Associated software comes with the standard lisp distribution of BOD.
Read MoreFinal draft from March 2009. An earlier, shorter version discussing Dennett more directly is listed under 2008. The scientific ideas (without mention of robots or consciousness) are supported in my CogSci09 paper, above.
Read MoreThe version linked from the title is the penultimate draft from December 2008.
Read MoreAssociated software is on the AmonI software page.
Read MoreDiscusses the importance of the discovery that human-like semantics can be learned simply from observing large corpora, with ramifications for the evolution of language. The final version is from November 2007, here is a penultimate draft (pdf) from August for those who do not subscribe, although it has a couple gaffs in it.
Read MoreThis is a somewhat pedantic overview of the improvements we’ve made to BOD, POSH and of course AI action selection in general in the last three years, with an eye to pleasing the EPSRC since my grant with the same title just ran out. Final version from January 2008.
Read MoreFinal version from 25 June 2007. Associated software.
Read MoreThis paper talks about how ABM fits in as a part of scientific methodology, and in particular analyzes as a case study the macaque social structure simulation in Hemelrijk’s DomWorld. The DomWorld link includes the associated software. An earlier version of this paper with the predictions and not the full analysis appears under 2005 below (Lehmann et al). Green open access draft.
Read MoreThis is actually quite a substantial article which covers the concept of action selection far more thoroughly than any other article in the two issues. I strongly recommend reading it. Written in April 2006.
Read MoreThis paper expands on the COIL system (presented first and more completely in the IEEE journal article) for imitation learning, showing how adding solid Bayesian representations improves both performance and extendibility. Final version from October 2006.
Read MoreThis paper presents an imitation learning system called COIL (inspired by Deb Roy’s CELL) capable of learning tasks in a dynamic real-time environment (Unreal Tournament). If you don’t subscribe to IEEE, here is a draft from May 2006. An even older version is Bath Technical report CSBU-2005-16.
Read MoreA model of transitive inference as the implicit learning of relationships between context-action pairs. The first version of this work appeared in my dissertation in 2001. Associated software. I also wrote a bibtex style for this article / journal: animalcog.bst.
Read MoreShows that the tendency to communicate information can be adaptive even though it has immediate costs to the communicators and there are free riders / information hoarders around the place. A chapter-length extension of Cace & Bryson ’05. Version from early March 2006.
Read MoreJon Dorbolo wanted someone to write about Dan Dennett and Cog, and apparently Dennett suggested me! This isn’t the usual kind of thing I write — it’s mostly anecdotes and amateur philosophy — but people like it. Talks about Cog’s first years, modular AI, philosophy of science, being a grad student, and a bit about memetics. Written in May 2005.
Read MoreShows that what picture is on a VR face has very significant impact on how the emotions are perceived. Final version is copyright Springer. Updated 23 June 2005
Read MoreFinal: May 2004.
Read MoreDiscusses when to use communication between agents in a multi-agent system vs. when to use behavior arbitration between modules in a modular single agent. Shows code from the primate colony simulation I’m working on with Jessica Flack. Final version is © Springer-Verlag. Updated 3 July 2002.
Read MoreUpdated 10 Dec, 2002. BBS called the issue this got printed in ‘Dec 2002’ but it must have come out in 2003! Well, I try to keep this page in sync with the actual publication dates, maybe that’s crazy … HTML
Read MoreA position paper, describes the importance of constraints in learning in artificial and natural agents. Final revision, 21 January 2002.
Read MoreUpdated Nov 13, 2001. HTML
Read MoreThat is the 173-page 11-point 1.5-spaced PDF version, with a little source code my examiners asked for. There’s also a 94-page 10-point single-spaced compressed postscript version with no source code. Complete source code is available at the bottom of this page. Summary: gives evidence for the need for structured control from three sources: the history of AI agent architectures, my experiments in two domains (robotics and artificial life), and a review of the neurological / behavioral literature. Also discusses the dialect differences between Psychology and AI, and AI as a research tool for Psychology. Final corrections, January 2000.
Read MoreA commentary on Dana H. Ballard, Mary M. Hayhoe, Polly K. Pook, and Rajesh P. N. Rao, Deictic Codes for the Embodiment of Cognition; both articles appeared in Behavioral & Brain Sciences (BBS) in late 1997.
HTML
Describes my MSc project, is also the first place I suggest adaptive requirements can serve as a key to determining how to decompose intelligence into modularized behaviors (key point in BOD).
Read More