I have participated in episode 404 of the Coffee Break podcast: Signal and Noise [iVoox, iTunes], entitled “Ep404: ChatGPT and Theory of Mind; Black Holes and Dark Energy; Neanderthals», Feb 23, 2023. «The weekly gathering in which we review the latest scientific news. In today’s episode: The laser cap and climate change (min 10:00); Black holes and dark energy (1:44:50); Finding of very recent Neanderthal remains (2:27:00); Signs of the listeners (2:41:40). All the comments made during the gathering represent only the opinion of the person who makes them… and sometimes not even that.
Go download episode 404.
As the video shows, we participate by videoconference Héctor Socas Navarro @HSocasNavarro (@pCoffeeBreak), Sara Robisco Cavite @SaraRC83Carlos Westendorp @cwestendJosé Edelstein @JoseEdelstein, Gaston Giribet @GastonGiribet, and Francis Villatoro @emulenews.
After the presentation, Héctor makes an errata: the other day, speaking of the gravitational binding criterion of the minimoon, he said “negative potential energy” when he should have said “negative energy”, kinetic energy plus potential. The reason is the convention that is taken to describe the linked systems: the energy that corresponds to the infinitely separated components of the system is taken as zero; said value is greater than the energy of the bound system, so for the latter, by convention, the binding energy is negative and the total energy is also negative. Of course, it is only an agreement.
Carlos tells us about two articles on the theory of mind (ToM) of language models (LLM). Many classic psychological experiments are being repeated using available artificial intelligences, such as those based on language models (like ChatGPT). In the case of ToM, these experiments are stories with characters designed to assess the false beliefs in said characters as evaluated by children or animals. In the figure, a child hides a chocolate bar in the green cabinet; later her mother takes her out of the green cabinet and puts her in the blue cabinet; the subject is then asked which piece of furniture he thinks the child will open later. The youngest children, without theory of mind, affirm that it will be blue; but the older children, with theory of mind, affirm that it will be green.
These computational experiments are very easy to perform, so many are being done. Some are showing that some artificial intelligences seem to have a theory of mind (they are able to pretend to put themselves in another’s place); it could be said that the theory of mind has emerged spontaneously in them. However, not all the results are positive and many experts have serious doubts about this apparent emergency. And, of course, everything can be seen in the opposite direction: that a language model that appears to have a theory of mind in a psychological experiment is perhaps a sign that the experiment is not well designed. The controversy is served.
As a curiosity, Jose comments that his ex-wife did her doctoral thesis on the theory of mind and designed specific experiments to relate linguistic ability and theory of mind. Carlos tells us about the article by Michal Kosinski, “Theory of Mind May Have Spontaneously Emerged in Large Language Models,” arXiv:2302.02083 [cs.CL] (04 Feb 2023), doi: https://doi.org/10.48550/arXiv.2302.02083, which concludes that GPT-3 (davinci-002, Jan 2022) successfully solves 70% of ToM experiments (as a child 7 years old), but that GPT-3.5 (davinci-003, November 2022) manages to solve 93% of them (as a 9-year-old child). Thus they affirm that a ToM has emerged in this language model. He also comments on Tomer Ullman’s article, “Large Language Models Fail on Trivial Alterations to Theory-of-Mind Tasks,” arXiv:2302.08399 [cs.AI] (16 Feb 2023), doi: https://doi.org/10.48550/arXiv.2302.08399, offering the contrary view; It is observed that small modifications in a ToM task greatly penalize the response of a language model, when it hardly affects the children. As a result, it is concluded that the emergence of a theory of mind in current LLMs is not observed.
Finally, Carlos tells us that the news “The University Hospital of the Canary Islands and the IAC advance in the early detection of colon cancer using artificial intelligence methods”, IAC, Dec 19, 2022. His project Patolog-IA, together with Andrés Asensio and other astrophysicists, develops and applies artificial intelligence methods used in astronomy to the processing of medical images for the diagnosis of colorectal cancer. For now it is limited to validating already known algorithms with already published image databases. In the future, he intends to apply it to real patient data and improve these techniques to achieve high reliability and reduce false positives.
I have to comment, along with Gastón, the supposed relationship between black holes and dark energy. I already commented on it in “Black hole impostors with dark energy inside are proposed to explain cosmological dark energy”, LCMF, 16 Feb 2023; The idea is to propose that supermassive black holes are objects without singularity that contain within them a core of dark energy (an “exotic matter” with a negative pressure of the opposite sign to density, that is, with an equation of state p = −ρ ). The idea is to explain the (apparent) growth of supermassive black holes by a factor between 8 and 20 from redshifts z ∼ 2.7 to z = 0. Unfortunately this does not hold up, as it would explain the energy contribution to the density of the universe through a contribution of matter (since black holes outside the horizon behave like MACHOs), which contradicts cosmological observations. Furthermore, black holes are very small, so their growth in mass and size due to the accelerated expansion of their inner core cannot explain the accelerated expansion observed in the universe.
Los artículos son Duncan Farrah, Kevin S. Croker, …, Chris Pearson, «Observational Evidence for Cosmological Coupling of Black Holes and its Implications for an Astrophysical Source of Dark Energy,» The Astrophysical Journal Letters 944: L31 (15 Feb 2023), doi: https://doi.org/10.3847/2041-8213/acb704, y Duncan Farrah, Sara Petty, …, Andreas Efstathiou, «A Preferential Growth Channel for Supermassive Black Holes in Elliptical Galaxies at z ≲ 2,» The Astrophysical Journal 943: 133 (02 Feb 2023), doi: https://doi.org/10.3847/1538-4357/acac2e.
Sara tells us that it has been published in American Journal of Biological Anthropology a study of Neanderthal teeth that suggests that they were in northern Spain for longer than previously thought. At the Lezetxiki de Arrasate or Mondragón (Guipúzcoa) site, the remains have been radiocarbon dated at different levels ranging from 32,600 to 39,500 years old in the Aurignacian culture and at least 46,500 years old in the Mousterian culture.
Sara tells us many details about the comparative analysis of these Neanderthal teeth. Thanks to stratography they have been dated much earlier than expected; thus the hypothesis that Neanderthals migrated south during the ice age, something that the new results rule out. Even so, one must be cautious and wait for future studies to clarify and contextualize the new results. The article is Diego López-Onaindia, Marina Lozano, …, M. Eulàlia Subirà, “Neanderthal teeth from Lezetxiki (Arrasate, Iberian Peninsula): New insights and reassessment,” American Journal of Biological Anthropology (31 Jan 2023), doi: https ://doi.org/10.1002/ajpa.24694; a related article is Igor Djakovic, Alastair Key, Marie Soressi, “Optimal linear estimation models predict 1400–2900 years of overlap between Homo sapiens and Neanderthals prior to their disappearance from France and northern Spain,” Scientific Reports 12: 15000 (13 Oct 2022 ), doi: https://doi.org/10.1038/s41598-022-19162-z; More information in “The Neanderthals inhabited northern Spain longer than previously thought”, Agencia SINC, 03 Feb 2023.
And we move on to Listener Signals. ask: Gaston, what if the black hole is teleological in such a way that space [la dirección radial] takes a meaning inside [que] equals time dimensions [se comporta como dirección temporal] and what is in front of him falls [hacia] Her future?” Gastón comments that the horizon is a null surface, like the trajectory of a photon; on the horizon there is no time in the same sense that time does not pass for a photon. Thus, in the interior, the time-type and space-type intervals are inverted with respect to the exterior of the horizon.
Enjoy the podcast!