Tuesday, March 25, 2025

Transmogrify My AI


AKA I Am a Man!

Leading AI chatbots show dementia-like cognitive decline in tests, raising questions about their future in medicine
Dec 2024, phys.org

Something about anthropomorphism - saying they have dementia, like saying they hallucinate, is assigning human-like qualities to a robot:

Researchers assessed the cognitive abilities of the leading, publicly available LLMs — OpenAI's ChatGPT versions 4 and 4o, Anthropic's Claude 3.5 "Sonnet", and Google's Gemini versions 1 and 1.5 — using the Montreal Cognitive Assessment test, widely used to detect cognitive impairment and early signs of dementia, usually in older adults. 

ChatGPT 4o achieved the highest score (26 out of 30), followed by ChatGPT 4 and Claude (25), with Gemini 1.0 scoring lowest (16).

The uniform failure of all large language models in tasks requiring visual abstraction and executive function highlights a significant area of weakness that could impede their use in clinical settings.

"Not only are neurologists unlikely to be replaced by large language models any time soon, but our findings suggest that they may soon find themselves treating new, virtual patients - artificial intelligence models presenting with cognitive impairment."

via Department of Neurology at Hadassah Medical Center Jerusalem, Hebrew University, Tel Aviv University: G Koplewitz: Age against the machine—susceptibility of large language models to cognitive impairment: cross sectional analysis, BMJ (2024). DOI: 10.1136/bmj-2024-081948



AI's next frontier: Selling your intentions before you know them
Dec 2024, phys.org

Sure it sounds scary, but isn't this what predictive analytics is all about? (And we've been doing that for years)

Forthcoming - "persuasive technologies" using "digital signals of intent" to predict your behavior in real time via Anthropomorphic AI agents. 

"We caution that AI tools are already being developed to elicit, infer, collect, record, understand, forecast, and ultimately manipulate and commodify human plans and purposes."

Again, "We caution that AI tools are already being developed to elicit, infer, collect, record, understand, forecast, and ultimately manipulate and commodify human plans and purposes."

via University of Cambridge's Leverhulme Center for the Future of Intelligence: Beware the Intention Economy: Collection and Commodification of Intent via Large Language Models, Harvard Data Science Review (2024). DOI: 10.1162/99608f92.21e6bbaa


Condé Nast, other news orgs say AI firm stole articles, spit out “hallucinations”
Feb 2025, Ars Technica

In February 2024, [generative AI company] Cohere announced that it would provide legal protection against intellectual property claims to its paying enterprise customers. This includes "full indemnification for any third party claims that the outputs generated by our models infringe on a third party's intellectual property rights," for Cohere "enterprise customers that comply with our guidelines and do not intentionally attempt to generate infringing content." (Note this is not a ruling like the Reuters case but just the beginning of the lawsuit.)

Release the copybot trolls!


New study identifies differences between human and AI-generated text
Feb 2025, phys.org

Just the stats ma'am: 

They show how LLMs write by prompting them with extracts of writing from various genres, such as TV scripts and academic articles. 

LLMs used present participle clauses at two to five times the rate of human text, as demonstrated in this sentence written by GPT-4o: "Bryan, leaning on his agility, dances around the ring, evading Show's heavy blows."

They also used nominalizations at 1.5 to two times the rate of humans, and GPT-4o uses the agentless passive voice at half the rate as humans. This suggests that LLMs are trained to write in an informationally dense, noun-heavy style, which limits their ability to mimic other writing styles.

The researchers also found that instruction-tuned LLMs have distinctive vocabularies, using some words much more often than humans writing in the same genre. For example, versions of ChatGPT used "camaraderie" and "tapestry" about 150 times more often than humans do, while Llama variants used "unease" 60 to 100 times more often. Both models had strong preferences for "palpable" and "intricate."

Can we just pause for a minute and recognize that we're saying "more often than humans do" as if we knew what universal human speech was like. All this talk about bias in the algorithms, certainly important because it can be amplified, but what about the bias in the base sets? What are the base data we're using to say what a "human" is like?

Also, as a native English speaker, I do recognize that non-natives tend to overuse present participles (like words ending in -ing), which may or may not have anything to do with this and the joke that 'AI is just three [people from underdeveloped communities] in a trenchcoat'. If you're not sure what I'm talking about, go listen to an Excel tutorial for a few minutes. ...

via Carnegie Mellon University: Alex Reinhart et al, Do LLMs write like humans? Variation in grammatical and rhetorical styles, Proceedings of the National Academy of Sciences (2025). DOI: 10.1073/pnas.2422455122

Posted by at No comments:
Labels: , , , , , ,

Monday, March 24, 2025

Facts, Fakes, Fake Facts and Psychological Operators

 

Brainwashing an entire population is hard work!

News consumers are more influenced by political alignment than by truth, study shows
Nov 2024, phys.org

The study is based on research conducted in 2020 with a sample of the voting-age U.S. population matched to the demographics of the census, and made up news headlines whereas earlier studies relied on existing ones. They also used a cover story about memory and communication, and even included foil questions, so that participants would not be aware of the nature of the study.

Both supporters and opponents believed those that aligned with their views more than they believed true headlines that did not align with their views.

"We found that the strongest predictors of bias include extreme views of Trump, a one-sided media diet, and belief in the objectivity and lack of bias of a person's own political side relative to the other side." 

In one measure of bias, the researchers assessed people's beliefs about their political side's objectivity as compared to the other side. Ironically, those most confident in their political side's lack of bias were the most biased.

In addition, the effect of partisan bias was stronger for real news than fake news. That is, people were more likely to disbelieve true information that challenged their political worldview than to accept false information that confirmed their worldview.

"We saw it on both political sides and even among people who scored well on a reasoning test. We were a bit surprised to see how widespread this tendency was. People were engaging in a lot of resistance to inconvenient truths."

"Everyone thinks it's the other person who is the problem."

But this very important point can't be overlooked, dropped halfway through the press article:

One contributing factor for this state of affairs, the researchers suggest, is increased consumption of partisan media.

And just for good measure, let's recall that as of 2010 Citizen's United allows private industry to contribute campaign funds to those running for public office. (As in, who's paying for that partisan media? Yes, that's who.) 

via: Department of Psychology in the School of Humanities and Sciences at Stanford: Michael C. Schwalbe et al, When politics trumps truth: Political concordance versus veracity as a determinant of believing, sharing, and recalling the news., Journal of Experimental Psychology: General (2024). DOI: 10.1037/xge0001650



Brief scientific literacy interventions may quash new conspiracy theories
Dec 2024, phys.org

Americans in states with higher scientific literacy scores were less likely to believe in conspiracies and had higher COVID-19 vaccination uptake rates over the time period.

via Penn State's Smeal College of Business: Nathan Allred et al, Conspiracy Beliefs and Consumption: The Role of Scientific Literacy, Journal of Consumer Research (2024). DOI: 10.1093/jcr/ucae024


A tangled web: Social media analysis suggests coordinated messaging among fossil fuel-derived hydrocarbon industries
Jan 2025, phys.org

Power rises to the top: "Our study suggests that climate obstruction in different industries is more coordinated than is generally recognized...these different companies in different sectors are using the same strategic messaging to promote a distorted image of their environmental responsibility."

Details: 125,300 unique tweets posted from 2008–2023 by the main Twitter accounts of nine key players in the US fossil energy/plastics/agrichemical trade: ExxonMobil, Chevron, American Petroleum Institute, Dow, Dupont de Nemours, Inc, American Chemistry Council, Corteva Agriscience, FMC Corp, American Farm Bureau

via Northeastern University: PLOS Climate (2025). DOI: 10.1371/journal.pclm.0000370


New AI tool detects fake news with 99% accuracy
Jan 2025, phys.org

The method developed by the researchers uses an "ensemble voting" technique, which combines the predictions of multiple different machine learning models to give an overall score.

(Too bad we don't actually care; see above.)

via Keele University School of Computer Science and Mathematics: Patricia Asowo et al, An Ensemble Modelling of Feature Engineering and Predictions for Enhanced Fake News Detection, Artificial Intelligence XLI (2024). DOI: 10.1007/978-3-031-77918-3_16


Meta-analysis uncovers public's skill in detecting fake news, but skepticism towards true news persists
Mar 2025, phys.org

While their findings suggest that most people can accurately judge the veracity of news, they also showed that people are slightly better at spotting false news than true news. In other words, most of the people who took part in the studies appeared to be better skilled at rating false news as false than rating true news as true.

"A small majority of people show this trend (59%). The implication of this finding is that we should focus more on increasing the acceptance of true news. Currently, a lot of efforts are dedicated to making people skeptical of (false) news, however, our data shows that there may be more room to increase the acceptance of true news than to reduce the acceptance of false news."

I also remember the study about how mainstream news being skeptical of a science fact are more influential than a fake news site spreading obvious fake shit, for example BBC saying there may be a small risk of myocarditis for example, even if they have a million other articles that say vaccines are safe. There is a risk of myocarditis, but the risk of dying, or becoming disabled, from Covid due to lack of vaccination is far, far higher. See here or here.

via Jan Pfänder et al, Spotting false news and doubting true news: a systematic review and meta-analysis of news judgements, Nature Human Behaviour (2025). DOI: 10.1038/s41562-024-02086-1.

Posted by at No comments:
Labels: , , , , , , ,

Sunday, March 23, 2025

Digital Double Definition in Doubt


As awareness of digital twins grows, so does its definition, as well as its warnings, but so does its results. We're right here in the middle between pre-hype and post-hype:

Digital twins of the Earth: Researchers are critical of the term
Dec 2024, phys.org

FYI - The Destination Earth project was launched in 2022 as a key pillar of the European Commission's efforts towards the Green Deal, and wants a digital simulation of the Earth. The German Federal Agency for Carthography and Geodesy is working on a digital, intelligent 3D image of Germany and the German federal state of Rhineland-Palatinate is running a digital "hydro-twin" project.

These scientists are trying to add to the conversation:
They point out the lack of a clear definition of the term "digital twin of the Earth," which may be misleading. "All digital representations of our planet are model representations. As such, they will always be detached from reality—as a map can never fully replicate the land it depicts" and "suggests that we can create a digital representation that allows us to stress-test the structural properties of the Earth system with any desired degree of accuracy and precision" and as that is not the case, "as every model is a simplification of reality, and its creation requires simplifying assumptions that will unavoidably lead to uncertainties." Also, "the creation of new models with higher resolutions will not necessarily lead to an improvement in knowledge and results; and new complex models need new methodologies that will enable researchers to apply them."

via Johannes Gutenberg University Mainz: Robert Reinecke et al, How to use the impossible map – Considerations for a rigorous exploration of Digital Twins of the Earth, Socio-Environmental Systems Modelling (2024). DOI: 10.18174/sesmo.18786

Image credit: The above image is a first, I believe, where the authors of a science paper used a generative network to make a press release thumbnail image for their article, which happens to be about the union of humans and AI. The credit is titled: "Credit: By the authors" [link] and it comes from the Carnegie Mellon Robotics Institute.


Digital twin research finds colon cancer cells can be reverted to normal cells
Dec 2024, phys.org

Results already, so we're in the next phase of the digital twin hype, where it's not hype anymore. 

via The Korea Advanced Institute of Science and Technology: Jeong‐Ryeol Gong et al, Control of Cellular Differentiation Trajectories for Cancer Reversion, Advanced Science (2024). DOI: 10.1002/advs.202402132

Posted by at No comments:
Labels: , , ,

Saturday, March 22, 2025

Automatic Updates to Your Spacetime Matrix


You would think something as universal as time would be less susceptible to major changes like revised definitions or refined measurements. But no, that's not how it works. 

Awareness without time? A deep look into timelessness in deep meditative states
Dec 2024, phys.org

Some good distinctions on "coherent conceptualization of deep meditative states, focusing in particular on phenomenal temporality during meditation", dubbed the Extended Now Interpretation.

On Duration - According to Extended Now Interpretation, deep meditation occurs over an interval that the person perceives as an extended present, with no awareness of change, succession, temporal order, beginning, or end. Thus, "timelessness" refers to a lack of temporal structure but not a lack of duration.

On Alertness - Since the option of staying alert or not presents itself anew for each passing moment, the meditator is aware of time passing, and thus aware of a period of time, Dr. Frischhut argues. This does not require the meditator to consciously keep track of a succession of moments, as maintaining alertness during meditation may be enough to give them an awareness of duration even without an awareness of succession.

On Memory - Meditators report having no access to memory during deep meditative states, which would present a problem to her argument. She proposes that the experience of being alert at any given moment has a particular phenomenological character that would be different had it not been preceded by an essentially identical experience of alertness - a second knock on a door wouldn't be experienced the same without the first knock. Thus, even without engaging memory, a meditator still experiences duration in the form of instances of momentary alertness that each flows seamlessly from a previous one.

via Akiko Frischhut at Sophia University in Japan: Akiko Frischhut, Awareness without time, The Philosophical Quarterly (2024). DOI: 10.1093/pq/pqae081



Study claims all observables in nature can be measured with a single constant: The second
Dec 2024, phys.org

The group argues that the number of fundamental constants depends on the type of space-time in which the theories are formulated; and that in a relativistic space-time, this number can be reduced to a single constant, which is used to define the standard of time.

Okun stated that three basic units—meter (length), kilogram (mass), and second (time)—were necessary to measure all physical quantities. In other words, he reaffirmed the so-called MKS system (M, for meter; K, for kilogram; S, for second), which was later incorporated into the International System of Units (SI). Veneziano, for his part, argued that in certain contexts two units would suffice: one for time and one for length. Duff was equivocal, stating that the number of constants could vary depending on the theory in question.

According to their criteria, the number of fundamental constants is related to the minimum number of independent standards needed to express all physical quantities. To repeat, in Galileo's space-time, all observables can be expressed in terms of units of time and space, which are usually the "second" and the "meter." In relativistic space-time, the unit of time—that is, the "second"—is sufficient to express any observable.

And the definition of "second" is currently based on a natural constant: the energy difference between two specific levels of the electronic layer of caesium-133. One second (1s) corresponds to the time of 9,192,631,770 oscillations of the radiation emitted when an electron passes between these two states of caesium-133.

via Institute of Theoretical Physics at São Paulo State University, Institute of Mathematics Statistics and Scientific Computing at the State University of Campinas, São Carlos Institute of Physics at the University of São Paulo: George E. A. Matsas et al, The number of fundamental constants from a spacetime-based perspective, Scientific Reports (2024). DOI: 10.1038/s41598-024-71907-0


Scientists observe 'negative time' in quantum experiments
Dec 2024, phys.org

Don't get too excited: The researchers emphasize that these perplexing results highlight a peculiar quirk of quantum mechanics rather than a radical shift in our understanding of time.

via University of Toronto: Daniela Angulo et al, Experimental evidence that a photon can spend a negative amount of time in an atom cloud, arXiv (2024). DOI: 10.48550/arxiv.2409.03680 , arxiv.org/abs/2409.03680

Posted by at No comments:
Labels: , , , , ,

Friday, March 21, 2025

Levy Things and Data Management

 

There's a whole bag of words referring to this one phenomenon we call foraging behavior. Animals poke around for food, poke, poke, poke, and then move to another spot, and poke, poke, poke. A big jump followed by a bunch of little jumps, followed by a big jump, etc. The explore-exploit heuristic it's also called. This behavior is found in many other places, from the way people look for a job to the way atoms move. That's called Brownian motion, or the Drunkard's Walk. It's found in the unconscious  uncontrollable way our eyes move when we look at things, and it's kind of even related to the way we choose baby names over decades. 

The following articles are mostly tangentially related to this phenomenon; they're about navigation and vision, and information theory in general. 
 
A new mechanism for animal food caching behavior discovered
Aug 2024, phys.org

Contrary to the long-held belief that scatter-hoarding animals rely on memory to retrieve cached food items, the researchers propose a static mechanism similar to hash functions used in computing. Hash functions in computing are algorithms that convert input data of any size into a fixed-size string of characters, which typically represents the data in a unique and efficient manner.

The researchers' mathematical model aligns with the activity of hippocampal spatial cells, which respond to an animal's positional attention. The remapping ensures that these cells activate consistently across subsequent visits to the same area but differ between areas.

This remapping, combined with unique cognitive maps, generates persistent hash functions that can aid both food caching and retrieval.

In other words: Both layers are arranged in a two-dimensional grid, with each cell corresponding to a specific location. The cache site is determined by the activity level of the output neurons, known as the cache score.

via Department of Cognition and Brain Sciences and The Department of Animal Sciences at Hebrew University of Jerusalem: Sharon Mordechay et al, A non-memory-based functional neural framework for animal caching behavior, Scientific Reports (2024). DOI: 10.1038/s41598-024-68003-8



Grid cells' rhythmic sweeps reshape understanding of brain's spatial navigation
Feb 2025, phys.org

Grid cells alternate between tracking an animal's real-time position and scanning the environment ahead in a highly regular pattern—sweeping 30 degrees to the right, then 30 degrees to the left—at a rapid pace of ten times per second.

These rhythmic sweeps create a more efficient way to anchor locations relative to one another, providing a richer and more adaptable navigation system than previously imagined.

So it's not exactly the eyes sweeping but the brain?

via Kavli Institute at the Norwegian University of Science and Technology: Abraham Z. Vollan et al, Left–right-alternating theta sweeps in entorhinal–hippocampal maps of space, Nature (2025). DOI: 10.1038/s41586-024-08527-1


How eye saccades enable mammals to simultaneously chase prey and navigate through complex environments
Feb 2025, phys.org

It's almost like we're not designed to look at static images:

Researchers reconstructed the visual fields of freely moving ferrets that were chasing a fleeing target and discovered that eye saccades (very rapid coordinated eye movements) align the world motion—and not the actual thing they are chasing—to the retina and retinal specializations used for high-acuity vision.

Saccades achieve this by countering head rotations to align the area of the sharpest vision with the direction of intended travel and the area of the least motion-induced blur. 

via Max Planck Institute for Neurobiology of Behavior and Max Planck Florida Institute for Neuroscience: Eye saccades align optic flow with retinal specializations during object pursuit in freely moving ferrets, Current Biology (2025). DOI: 10.1016/j.cub.2024.12.032.


Posted by at No comments:
Labels: , , ,

Thursday, March 20, 2025

Hide Your Brains


The era of the omnicortex is upon us, and by upon us I mean it's coming for you, and your brain.
 
Mini-brains reveal how mitochondrial mutations affect brain cells
Dec 2024, phys.org

Mini-brains, aka Brain Organoids. That is all.

via University of Bergen: Anbin Chen et al, Hallmark Molecular and Pathological Features of POLG Disease are Recapitulated in Cerebral Organoids, Advanced Science (2024). DOI: 10.1002/advs.202307136



Researchers achieve success in allowing a patient to 'speak' using only the power of thought
Jul 2024, phys.org

(Epilepsy patients) In the first stage of the experiment, with the depth electrodes already implanted in the patient's brain, researchers asked him to say two syllables out loud: /a/ and /e/ while they recorded brain activity. Researchers then trained AI models to identify the specific brain cells whose electrical activity indicated the desire to say /a/ or /e/. Once the computer learned, he was asked to only imagine... The computer then translated the electrical signals...

via Tel-Aviv University Sourasky Medical Center, Ichilov Hospital: Ariel Tankus et al, A Speech Neuroprosthesis in the Frontal Lobe and Hippocampus: Decoding High-Frequency Activity into Phonemes, Neurosurgery (2024). DOI: 10.1227/neu.0000000000003068


Artificial imagination with the 'exocortex:' Researcher proposes software to aid scientific inspiration and imagination
Jan 2025, phys.org

The conceptualized exocortex will be an extension of a scientist's brain.

Maybe you just want to know that Brookhaven is doing this, that's all.

via U.S. Department of Energy's Brookhaven National Laboratory and Electronic Nanomaterials Group leader at the Center for Functional Nanomaterials: Kevin G. Yager, Towards a science exocortex, Digital Discovery (2024). DOI: 10.1039/D4DD00178H


'DeepFocus' offers minimally invasive brain stimulation through the nose
Feb 2025, phys.org

"By going through the nose, we can place electrodes as close to the brain as possible without opening the skull."

And that's all you need to know.

via Carnegie Mellon University Electrical and Computer Engineering: Yuxin Guo et al, DeepFocus: a transnasal approach for optimized deep brain stimulation of reward circuit nodes, Journal of Neural Engineering (2025). DOI: 10.1088/1741-2552/adac0c


Direct translation of brain imaging to text with MindLLM
Feb 2025, phys.org

fMRI to text. You're fucked. 

  • AKA decoding brain activity into natural language
  • Improves on prior models like UMBRAE, BrainChat, and UniBrain

via Yale University, Dartmouth College, and the University of Cambridge: Weikang Qiu et al, MindLLM: A Subject-Agnostic and Versatile Model for fMRI-to-Text Decoding, arXiv (2025). DOI: 10.48550/arxiv.2502.15786

Posted by at No comments:
Labels: , , , , , ,

Wednesday, March 19, 2025

Who Am I


You may need to channel you inner Dan Dennett for this one:

Navigating space: Dual maps discovered in the brain
Sep 2024, phys.org

Multidimensional navigation - basically you have two navigation centers, one of them in an area called the secondary motor cortex.

"Imagine being asked where the nearest coffee shop is. You could say 'walk forward and turn left' (self-centered directions) or 'walk north, then east' (world-centered directions). We want to understand how the brain transitions between these reference frames and transforms them into action."

via Sainsbury Wellcome Centre at University of College London: Liujunli Li et al, Encoding of 2D Self-Centered Plans and World-Centered Positions in the Rat Frontal Orienting Field, The Journal of Neuroscience (2024). DOI: 10.1523/JNEUROSCI.0018-24.2024



Memories are not only in the brain, human cell study finds
Nov 2024, phys.org

Isn't this embodied cognition?

"In the future, we will need to treat our body more like the brain - for example, consider what our pancreas remembers about the pattern of our past meals to maintain healthy levels of blood glucose or consider what a cancer cell remembers about the pattern of chemotherapy."

In the research, the scientists replicated learning over time by studying two types of non-brain human cells in a laboratory (one from nerve tissue and one from kidney tissue) and exposing them to different patterns of chemical signals - just like brain cells are exposed to patterns of neurotransmitters when we learn new information.

In response, the non-brain cells turned on a "memory gene" - the same gene that brain cells turn on when they detect a pattern in the information and restructure their connections in order to form memories.

via New York University Center for Neural Science: N. V. Kukushkin et al, The massed-spaced learning effect in non-neural human cells, Nature Communications (2024). DOI: 10.1038/s41467-024-53922-x


Posted by at No comments:
Labels: , , ,
Subscribe to: Posts (Atom)