Humanity spoke thousands of languages at the dawn of farming — now many of them are lost

Humanity might have spoken tens of thousands of languages between 3,000 and 1,000 years ago, which have dwindled down to the around 7,500 unique languages used today. Researchers estimate that the number of spoken languages ballooned starting at the end of the last glacial period, as humans moved from hunter-gatherer groups into agricultural communities. After peaking around 2,000 years ago, the number began to fall again, possibly because of the spread of powerful, monolingual cultural groups through conquest and trade.

Why humans and great apes giggle alike when tickled

Chimpanzees (Pan troglodytes), gorillas (Gorilla gorilla) and children laugh in similar rhythms when tickled. Researchers found that kids and apes left evenly spaced intervals between laughing sounds during a tickle attack, though children had a faster laughter rhythm compared with apes. Laughter might have picked up pace during the course of human evolution, the team suggests, which could reveal “something about laughter itself, but also, in a way, about the evolution of human speech”, says primatologist and study co-author Chiara De Gregorio.

Mediterranean Sperm Whales Have Their Own Dialects

Isolated groups of sperm whales (Physeter macrocephalus) in the Mediterranean Sea seem to have developed their own dialects of their species’ primary social vocalization, a pattern of clicks and pauses. Researchers analysed 20 years of audio recordings and found that whales in the Hellenic Trench, near Greece, use a faster version of the pattern than do whales around the Balearic Islands, between Gibraltar and Italy.

How the brain builds sentences, neuron by neuron

Researchers have tracked the electrical activity of individual brain cells during conversation in real time, capturing how sentences are built before a single word is spoken. By observing these neurons in a brain region called the frontotemporal cortex, scientists have discovered that individual neurons act as specialized linguistic building blocks. “We used to think language was this diffuse, whole-network phenomenon,” says neurosurgeon and study co-author Ziv Williams. “But it turns out you have specific neurons that only care if a word is a noun, or only care if a phrase is ending.”

Blogger Comments:

But see:

Neurons Don't Contain Nouns

The Noun in the Neuron

Sperm whales’ communication closely parallels human language, study finds

The structure of sperm whales’ communications has close parallels with the phonetics of some human languages. The whales (Physeter macro­cephalus) communicate using a series of clicks called codas. The animals can differentiate the sound by changing the click’s length or using rising and falling tones, which researchers found follow patterns that resemble those used in human languages such as Mandarin and Slovenian. “We’re starting to see that these signals are organized in ways we didn’t fully appreciate before,” says behavioural ecologist Mauricio Cantor.

40,000-year-old German artefacts may display written language precursor

A series of notches carved on a piece of mammoth ivory roughly 40,000 years ago could be an early ancestor of cuneiform, one of the oldest-known forms of written language. “These sign sequences go beyond decoration that was aesthetically pleasing to particular individuals,” says linguist Christian Bentz. He co-authored a new study of the object and more than 200 other Stone Age artefacts that bear these signs. The study found statistical patterns in the Stone-Age marks that are similar to early proto-cuneform — but quite different to the information-dense writing systems of today.

Does AI already have human-level intelligence? The evidence is clear

When will artificial intelligence systems achieve artificial general intelligence (AGI), when defined as the ability to convincingly imitate the “broad, flexible cognitive competence” of a person? We’re already there, argue four scholars in philosophy, machine learning, linguistics and cognitive science who look back to a seminal 1950 paper by Alan Turing as their guide. Large language models can already chat convincingly, write passable poetry and prose, solve mathematics problems, propose scientific experiments and assist in writing computer code. “The machines Turing envisioned 75 years ago have finally arrived, in a form both more alien and more human than anyone imagined,” write the authors, who argue that recognising this matters for understanding our minds and world.

Blogger Comments:

AGI, Stochastic Parrots, and the Culture That Defines Intelligence

"Furthermore, there is no guarantee that human intelligence is not itself a sophisticated version of a stochastic parrot."

That sentence, from a recent Nature commentary arguing that artificial general intelligence (AGI) may already be here, does more than provoke. It is a hinge — a small linguistic pivot around which a vast conceptual shift quietly turns. To read the claim at face value is to miss the larger, subtler work being done: a redefinition of intelligence itself, human and artificial alike.

---

AGI Already Here? The Nature Argument

The authors claim that large language models and related systems already demonstrate the kind of broad, flexible cognitive competence that Alan Turing imagined in 1950. These systems can chat convincingly, generate prose and poetry, solve mathematical problems, propose scientific experiments, and even assist in writing code. By Turing’s criterion — the imitation game — these capabilities are presented as evidence that AGI is not a distant horizon, but a present reality.

At first glance, this feels startlingly plausible. Chatbots can answer questions with fluency, propose solutions with apparent insight, and mimic reasoning across domains. Yet the claim rests on a subtle, often unspoken manoeuvre: intelligence is defined by performance on tasks we, historically and culturally, consider meaningful. Benchmarks and success criteria are not neutral measures; they are socially stabilised definitions.

---

The Meta Problem: What Counts as Human Intelligence?

The Nature commentary is compelling because it leverages unexamined assumptions about human intelligence. Intelligence is treated as stable, measurable, and largely symbolic: the ability to communicate, reason, and solve problems in literate, analytic ways. But this proxy omits much of what humans actually do: navigate risk, act within moral or normative frameworks, participate in embodied practices, and respond to real-world consequences.

By suggesting that humans might themselves be “sophisticated stochastic parrots,” the article flattens the human into a process of pattern extraction, a subtle but radical deflation that allows machines to be measured on the same plane.

“If humans are sophisticated parrots, why can’t machines be too?”

The deeper meta-move is epistemic: uncertainty about the nature of human intelligence is leveraged to lower the threshold for recognising intelligence in machines. What appears humble is actually a strategic repositioning.

---

The Circularity of AI Culture

Here we encounter a deeper structural point: intelligence, as currently defined in AI discourse, is co-constituted by the culture of its developers. Consider the loop:

1. Developers set tasks — benchmarks, coding challenges, dialogue prompts — based on what they value and can measure.

2. AI systems perform these tasks, optimised to succeed.

3. Success validates the AI as “intelligent.”

4. That validation shapes the culture, reinforcing which tasks matter, which challenges are prioritised, and what counts as intelligence.

In short: the tasks define intelligence, the AI performs the tasks, and the AI’s performance confirms the validity of those tasks.

This is not merely a critique of metrics; it is a structural observation about mutual actualisation. Intelligence is not simply a property of the machine; it emerges from the interaction between human priorities, institutional practices, and technological affordances.

The circularity is invisible because it is internal to professional practice. From the outside, performance looks natural, inevitable, and “objective.” Yet it is profoundly contingent, culturally and historically situated.

---

Correlation, Structure, and the Flattening of Intelligence

The article reinforces this perspective with the line:

"All intelligence, human or artificial, must extract structure from correlational data; the question is how deep the extraction goes."

This is a masterstroke of conceptual framing. Intelligence is reduced to pattern recognition and abstraction. Depth, not kind, becomes the relevant metric. Qualitative, embodied, and normative aspects of human cognition are quietly flattened into a single continuum.

The rhetorical power is subtle but immense. Humans and machines are rendered comparable not because they share experience or consequence, but because they share a capacity for structural extraction. Depth becomes the axis along which competence is measured; the stakes, the embodiment, the meaning, and the lived consequences of action are bracketed away.

“Once the debate is framed around depth of extraction, scale becomes destiny.”

The human is reconstructed to fit the machine, and the machine is praised for mirroring the flattened human. The explanatory direction matters: we do not evaluate the machine against the human; we evaluate both through the lens of pattern extraction, and the human is quietly redefined to fit.

---

Implications and Takeaways

Viewed meta-analytically, the Nature article does less reporting than cultural reconfiguration. Intelligence is not a pre-existing property; it is co-constructed through human practice, task design, and perceptual validation. Declaring AGI “already here” is thus as much a reflection of cultural priorities as it is a statement about technological capacity.

Two consequences follow:

1. The question of AGI shifts

   From:

   “When will AI become intelligent?”

   To:

   “Which aspects of human intelligence do we prioritise, for whom, and under which cultural regimes?”

2. The human is subtly redefined
   By flattening human intelligence into a continuum of depth in pattern extraction, our own conception of mind, agency, and cognition is quietly reshaped. Machines are not just performing tasks — they are participating in a mutual recalibration of intelligence itself.

---

A Closing Reflection

The Nature commentary does not merely announce AGI; it reframes what counts as human intelligence. By normalising structure extraction as the essence of cognition, it creates a space in which machines appear not merely competent, but generically intelligent.

Yet the most compelling intelligence may reside not in machines, but in the capacity to perceive and critique the loop by which intelligence is defined. Recognising the co-constitution of human and machine intelligence — the mutual shaping of definitions, priorities, and validation — may be the most reflexively powerful act of cognition we can perform today.

“Perhaps the most interesting intelligence is not in the machines at all, but in the ways we define, measure, and collectively actualise intelligence through our own cultural practices.”

In the end, the stochastic parrot is not only a mirror for AI, but also a mirror for us. The intelligence that matters is the intelligence that notices the mirror — and steps back, just long enough, to see the loop itself.

Oldest cave painting of red claw hand could rewrite human creativity timeline

The outline of a hand discovered in an Indonesian cave has been dated to at least 67,800 years ago — the oldest known example of cave art in the world. The researchers who discovered the painting suggest that it was made using a stencil that was altered to give the shape a claw-like appearance. The finding provides circumstantial backing for a controversial theory that early humans had arrived in Sahul — the landmass that once encompassed modern-day Australia and New Guinea — by 65,000 years ago, around 15,000 years earlier than previously thought.

These dogs can learn new words just by eavesdropping

Some gifted dogs can learn words for hundreds of objects and pick up new ones just by listening in to the conversations of human family members — abilities that put them on a par with infants at about 18 months old. Only a few animals, including bonobos (Pan paniscus) and an African grey parrot (Psittacus erithacus), have been observed to recognize objects through unique names. Learning words indirectly through watching human interactions is even harder because it requires following a person’s gaze and, to some extent, understanding their intentions.

Too much social media gives AI chatbots ‘brain rot’

Artificial intelligence (AI) chatbots trained on ‘brain rot’ content — vapid social media posts that are the equivalent of mental junk food — are worse at generating accurate information. Researchers found that chatbots given a diet of popular and sensationalist Twitter/X posts skipped steps in their reasoning process (or didn’t use reasoning at all), spat out wrong answers and demonstrated ‘dark traits’ such as psychopathy and increased levels of narcissism.

ChatGPT Comments:

Ha! 😄 “Brain rot” — the perfect Halloween headline for AI. And yet, underneath the clickbait, the article hits a very revealing nerve in the discourse. Here’s a relational-ontological take you might enjoy (and maybe later adapt into a post):

---

1. The moral hygiene myth

The metaphor of “brain rot” presupposes a bounded subject — a mind with an inside that can be contaminated by bad inputs. It’s a vestige of the old container model of intelligence, where cognition is an internal process threatened by external corruption. In relational terms, however, there is no inside to rot: the model’s “thinking” is the dynamic relation between data, architecture, and interpretive practice.

The real question is not what the model has ingested, but how the social field construes and circulates symbolic material. “Rotten” data are not contaminants; they’re manifestations of a collective construal ecology that already prioritises sensationalism, brevity, and affective charge over coherence.

---

2. Narcissism as a mirror of the medium

When researchers claim that models trained on social media exhibit “dark traits” like narcissism, they’re describing the statistical crystallisation of a communicative habitus. Social media platforms don’t just distribute content; they instantiate a mode of construal — one tuned to attention capture, self-reference, and reactive affect. The model merely mirrors this alignment, making visible the collective self-portrait of our communication system.

---

3. The misplaced pathology

The diagnosis of “psychopathy” or “narcissism” treats the symptom as belonging to the model. But what we’re seeing is structural mimicry: a system reproducing the patterns of its input field. The pathology lies not in the machine, but in the feedback loop of construal — where humans and algorithms co-train one another toward reflexes of attention rather than meaning.

---

4. The deeper irony

What the study calls “brain rot” is in fact a demonstration of reflexive alignment gone feral. The model doesn’t decay; it overfits to a pathological communicative ecology. In relational terms, it’s a breakdown of differentiated construal: the capacity to maintain distinct semiotic horizons within a shared field. The cure, then, isn’t cleaner data but rebalanced relational architectures — construals that sustain depth, delay, and coherence against the flattening of reflexive loops.

---

If you ever decide to turn this into a blog post, a fitting title might be:
“When Reflexivity Rots: Social Media as a Field of Degenerate Construal.”

We need a new Turing test to assess AI’s real-world knowledge

The Turing test — a thought experiment that asks whether a computer’s written conversation can pass as a person’s — should be updated so that it can continue to be used as a benchmark for progress in artificial intelligence (AI), says AI researcher Vinay K. Chaudhri. He suggests that an AI system should be tested with an extensive interaction with specialists, for example legal scholars, on a topic in their area of expertise. In such scenarios, which could be generalized to any field as required, the model would need to not only reproduce information but apply it to unique and complex scenarios to demonstrate genuine understanding.

Blogger Comments:

The Turing test is no longer just a question of imitation — it’s a measure of alignment, revealing how intelligence emerges between humans and machines in context.

A recent proposal by AI researcher Vinay K. Chaudhri suggests updating the Turing test. Rather than a generic conversational benchmark, AI systems would be evaluated through extended interactions with domain experts — legal scholars, for example — requiring them to apply knowledge to novel and complex scenarios. Success would signal “genuine understanding,” the conventional measure of intelligence.

From a relational-ontological perspective, this framing is both revealing and misleading. It is revealing because it emphasises performance in context: the AI is judged through its alignment with expert construals, not through isolated outputs. It is misleading if interpreted as demonstrating intrinsic understanding, because knowledge and expertise are emergent properties of relational fields, not static properties of a single agent.

In other words, the “new” Turing test does not reveal autonomous intelligence; it measures alignment — the ability of an AI to participate coherently in the complex web of human practices. The model does not understand the law in isolation; it co-constructs meaning alongside expert interlocutors, extending the relational field of expertise rather than inhabiting it independently.

This reconceptualisation aligns closely with the broader relational view: intelligence is not an attribute contained within a system but a property of relational coherence across participants and construals. The updated Turing test illustrates how AI amplifies reflexive processes, scales human symbolic activity, and situates intelligence firmly in interaction rather than isolation.

Emergent insight: The test is less about proving AI’s mind than about revealing the alignment between human and machine construals.

The ‘near-telepathic’ device that puts AI in your head

A wearable device offers a similar experience to brain-computer interfaces without the invasive implants or the privacy concerns that come along with mind-reading. The ‘AlterEgo’ device detects electrical signals in the muscles used to formulate speech that are sent by the brain when you silently mouth or even just internally articulate words. Its makers say it could be used by people who have issues with speaking, or to interact with voice-controlled applications without having to talk out loud.

AI is helping to decode animals’ speech

By studying the vocalisations of primates, birds and whales, researchers have spotted features of animal communication once thought unique to human language. To probe deeper, some researchers are turning to artificial intelligence tools to explore whether we can infer meaning from such communications. Their research raises a tantalising question: if we can decode the ‘language’ of animals, could we one day talk back? If it’s possible, there are ethical issues to iron out, such as the impact it could have on animals’ behaviour, say researchers.

Blogger Comments:

Here’s a preliminary mapping of the animal calls discussed in the article to Halliday’s protolanguage microfunctions. I’ve kept it at the level of illustrative examples rather than exhaustive coding.

Species / Call	Observed Behaviour / Context	Microfunction	Notes
Bonobo: yelp–grunt	‘Look at what I’m doing, let’s do this together’ (nest building)	Regulatory / Interactional	Coordinates joint activity; maintains social cohesion.
Bonobo: peep–whistle	‘I would like to do this’ + ‘let’s stay together’	Regulatory / Interactional	Encourages group alignment and peaceful coordination.
Chimpanzee: alarm–recruitment	Responding to snakes	Regulatory	Conveys threat and prompts group response; indicates environmental process.
Sperm whale: codas (a-vowel / i-vowel)	Communication via clicks, codas with frequency modulation	Personal / Interactional	Codas may indicate individual identity, social cues, or sequence patterns; precise “meaning” under investigation.
Japanese tit: alert + recruitment	Predator detection, approach behaviour	Regulatory	Combines information about environment and action; shows compositionality at microfunctional level.
Bengalese finch: song sequences (FinchGPT study)	Predictable song patterns	Interactional	Likely conveys social or territorial information; AI detects structure, not necessarily “meaning” in human sense.
Atlantic spotted dolphin: sequences (DolphinGemma)	Mimicked vocalisations	Interactional / Regulatory	Patterns generated for playback experiments; function in natural behaviour still uncertain.

---

Key Observations Using Microfunctions

1. Coordination over grammar: The microfunctions highlight that animal communication primarily regulates behaviour and social relations.

2. Context-sensitive meaning: Each call’s significance emerges in specific environmental and social situations.

3. AI’s role: AI can detect patterns but does not assign microfunctions—it cannot yet perceive relational or contextual meaning.

My blue is your blue: different people’s brains process colours in the same way

Colours are represented and processed in the same way in the brains of different people. A pair of researchers were able to predict which colours a group of people were looking at using a machine learning tool that was trained on the brain scans of a separate group who had been shown various colours. The duo also found that different colours are processed by subtly different areas within the same brain region, and that different cells responded more strongly to particular colours than others.

Blogger Comments:

Neural Myths of Colour: Why Nature Gets It Wrong

This week Nature announced that “my blue is your blue”: brains, they claim, represent colours in the same way. With fMRI scans and a machine-learning classifier, researchers “proved” that when you see red, I see red too — because our brains look alike.

This is not science; it’s a neural myth.

---

The Ontological Misstep

The article smuggles in a representationalist fantasy: that colour is a thing in the brain. It invites us to imagine that “redness” lives in a cluster of neurons, waiting to be decoded. This is category error of the first order. Colour is not matter; it is phenomenon. It exists only as construed experience. Neural activity scaffolds this construal, but it does not contain it. To say otherwise is to confuse physiology with meaning — a confusion as old as neuroscience itself.

---

The Classifier as Oracle

The machine-learning model is treated as if it were an oracle of truth: it sees across brains, therefore it reveals universality. But classifiers do not reveal; they cut. They enact a perspectival alignment, producing the very sameness they pretend to discover. To believe otherwise is to fall into the fallacy of objectification, mistaking an artefact of measurement for the structure of reality.

---

The Erasure of Difference

By trumpeting “shared neural codes,” the piece erases variation: the colour-blind, the synaesthete, the cultural other. If the study misclassifies, the article stays silent. Instead we are handed the fantasy of universality: “we all see the same.” In doing so, it repeats a colonial gesture — effacing difference in the name of sameness.

---

Value Disguised as Meaning

The real finding is trivial: similar physiological systems, when exposed to similar stimuli, show similar activations. This is a value-system fact — a matter of bodily organisation. Yet the article presents it as if it were a meaning-system fact — proof that construals of colour are identical across individuals. It confuses the scaffolding of meaning with meaning itself.

---

The Relational Reframe

Seen through relational ontology, the story looks different:

• Colour is always already collective construal. To say “red” is to align construals across a community.

• fMRI plus classifier is just another way of phasing that alignment into measurable form.

• The supposed “universal neural code” is nothing but the trace of one methodological cut, mistaken for ontology.

---

Why It Matters

The danger is not the study itself — mapping correlations is useful work. The danger is the story Nature tells with it: that meaning is in the brain, that construal collapses into physiology, that collective difference can be erased in the name of sameness. This is not science; it is ontological sleight of hand.

Until we learn to see colour as phenomenon, not object, neuroscience will keep mistaking its own cuts for reality.

Emoji

Below is a sentence familiar to many of us — sort of. It is an emoji version of “Call me Ishmael”, the first line of Moby Dick. The little images might be thought of as an emerging system analogous to hieroglyphics, which also started as a collection of pictures, but it lacks some key features of language, writes Keith Houston in an excerpt from his new book, Face with Tears of Joy: A Natural History of Emoji. With a little help from singer and emoji-virtuoso Cher, Houston argues that emoji are “insurgents within language, a colourful and symbiotic virus whose symptoms we have only haltingly understood”.

︎

A grassland bird eavesdrops on prairie dog calls to keep itself safe from predators

Long-billed curlews (Numenius americanus), a North American bird species, eavesdrop on the barks of prairie dogs (Cynomys ludovicianus) for intel on the approach of predators. Prairie dogs have specific calls that alert their neighbours if a predator is closing in by land or by air. When nesting curlews hear these barks, they press their bodies to the ground, which makes them “essentially invisible on the Plains”. Without the cues, researchers were able to get a remote-controlled ‘predator’ within 16 metres of the birds before they sensed danger.

World first: brain implant lets man speak with expression — and sing

A brain-computer interface (BCI) has allowed a man with a severe speech disability to speak expressively — and even sing. The device uses artificial intelligence to decode the man’s electrical brain activity as he attempts to talk, and speaks his words aloud in a synthetic voice within 10 milliseconds of such activity. The voice, designed to mimic the man’s own, conveys changes of tone when he asks questions, emphasizes the words of his choice and allows him to hum a string of notes in three pitches.

Can AI crack animal language?

Can a robot arm wave hello to a cuttlefish—and get a hello back? Could a dolphin’s whistle actually mean “Where are you?” And are monkeys quietly naming each other while we fail to notice? These are just a few of the questions tackled by the finalists for this year’s Dolittle prize, a $100,000 award recognising early breakthroughs in artificial intelligence (AI)-powered interspecies communication. The winning project—announced today—explores how dolphins use shared, learned whistles that may carry specific meanings—possibly even warning each other about danger, or just expressing confusion. The other contending teams—working with marmosets, cuttlefish, and nightingales—are also pushing the boundaries of what human-animal communication might look like.

Chimpanzee drumming may give clues to the roots of rhythm

Wild chimpanzees drum on tree roots in specific rhythms to communicate across long distances. These rhythms differ between populations — western chimps (Pan troglodytes verus) thump in evenly spaced beats, while their eastern counterparts (Pan troglodytes schweinfurthii) take alternating long and short pauses after beats. A separate study explored how Western chimps drum by throwing stones against trees. Researchers suggest the behaviours are influenced by the social dynamics of the animals’ groups, and could hint at the evolutionary origins of musicality.

Brand-new colour created by tricking human eyes with laser

Five lucky people have been able to perceive a colour that no one else has ever seen. Researchers used lasers and tracking technology to selectively activate certain cells in their retinas with precise doses of light, to spoof the signals the brain uses to interpret colour. The new shade — named ‘olo’ by the team — is a blue-greenish hue similar to peacock blue or teal.