What can gorilla Koko teach us

In Koko's first year of life, she was separated from her mother for treatment at the zoo hospital. Here, the baby was first encountered by 24-year-old graduate student Francine Patterson. During the treatment, Patterson taught her American Sign Language. Something amazing happened, and Koko began to communicate with gestures.

Before Koko, similar experiments had already been conducted. Gorillas and chimpanzees are incredibly capable of expressing their emotions, showing joy, laughter, disappointment, sadness, and sorrow. Koko went further. For several hours a day, the gorilla studied new gestures. Soon, in addition to the usual requests for food, a solid stock of learned gestures helped Koko tell people about herself.

The gorilla was unable to have children. Koko's maternal love was given to a pet kitten. When he was hit by a car, the gorilla made the gestures "bad bad bad" and "frown cry frown sad trouble."

Koko mastered over a thousand gestures in total. Due to her excellent abilities, the gorilla became the most famous talking ape on the planet. She knew how to joke and loved to do it.

The intelligence of gorillas is similar to that of humans. In tests, Koko demonstrated a high IQ. If compared to people, Koko is smarter than the average resident of sub-Saharan Africa. This fact is not commonly discussed. When the smartest man in the world (IQ 210), Christopher Langan, repeated this fact, Facebook¹ banned his account.

Koko had a personality and was capable of lying. One day, she was left alone in a room with a potted plant. It’s not surprising that the herbivorous gorilla ate the tasty greenery. Patterson returned and asked Koko what happened. Koko replied that a graduate student had eaten the plant. When Francine pointed out that people don’t eat plants, only gorillas do, Koko claimed it was another gorilla.

Koko's last recorded message was her communication to the participants of the 2015 UN Climate Change Conference. In the video, Koko introduces herself and says, "I am a gorilla. I am flowers, I am animals, I am nature."

"Humans, Koko loves," Koko said. "Earth, Koko loves. But humans are foolish... Foolish! Koko regrets... Koko cries. Time to hurry! Fix the Earth! Help, Earth! Hurry! Protect, Earth... Nature sees you. Thank you."

On June 19, 2018, at the age of 46, gorilla Koko passed away.

How Children Learn Language

The so-called "last message" was released three years before the gorilla's death. It’s quite possible that there are newer video recordings.

Should we believe that a gorilla was taking temperature measurements and decided that global warming is an important threat to report to the UN conference? Of course not. The video above is a funny nonsense seriously presented by some group of French climate lobbyists. The obvious editing cuts and the way the gorilla looks to the right (probably repeating someone else's gestures) give it away.

In fact, Koko never knew how to speak. But let's start from the beginning.

If a person is injected with a solution of amytal sodium into one of the carotid arteries, the corresponding hemisphere of the brain will be turned off. If the hemisphere of the brain that controls speech is inactivated, the person will lose the ability to converse. This is usually checked by asking the person to count and read aloud.

This medical procedure is called the Wada test. It is not some interesting experiment or stage entertainment, but a real medical test. It is needed to determine which hemisphere of the brain of a particular patient is responsible for speech and memory before surgery. The fact is that speech functions are concentrated in the left hemisphere in only 85-90% of people, more often in right-handers than in left-handers - 90-95% versus 70-85%.

However, descriptions of speech centers in the human brain do not bother with such details and paint an average picture. The Broca's area and Wernicke's area are described. The first fragment of the cerebral cortex is believed to be responsible for the creation (reproduction) of speech and its motor organization. The second area is associated with speech comprehension.

The actual number of functions of these centers is higher. Since these areas are not isolated from the rest of the brain, their functions are deeper than such short descriptions. For example, the Broca's area is also connected with the basal ganglia and various structures responsible for movements. It is easy to trace this connection in real life by the fact that people cannot help but use gestures and facial expressions where there is no visual component — on the phone or if both interlocutors are blind.

However, these areas are weakly connected with the limbic system. That is why people often retain the ability to sing with Broca's aphasia — the loss of the ability to compose meaningful messages and sequences of words due to damage to the eponymous center. Also, with Broca's aphasia, in the absence of damage to other speech centers, a person may retain a good ability to understand other people's speech.

Accordingly, if the Wernicke's area is affected by a stroke, word deafness will develop — the inability to perceive speech. But not only perception degrades. Although speech does not disappear with Wernicke's aphasia, patients often do not realize that they are producing a meaningless stream of jargon.

Aphasia develops differently in bilinguals who learned to perceive a second language in the first 6 years of life, and in people who learned a second language at an older age. In the second case, the second language is located in more peripheral parts of the Broca's center and Wernicke's area, while in the first case, the two languages are encoded in overlapping regions. Therefore, in aphasia, bilinguals lose the ability to both languages equally. Those who learned a second language in adulthood may lose only one of the two languages in a stroke.

Physically, Broca's center and Wernicke's area are not adjacent. They are connected by an arcuate fasciculus - a group of axons that help combine speech production with its understanding. In theory, if this and only this part of the brain were damaged, a person would retain the ability to speak and understand, but would not be able to fulfill the request to repeat a word. But conduction aphasia, which is similar in symptoms, is relatively rare.

How do children learn a language? Clearly not in school, where it is recorded as rules. Native language lessons for schoolchildren are a relatively recent phenomenon, only a few centuries old. Children independently learn the rules of the language based on how those around them speak it. At the same time, they adopt the language not from their parents, but from their peers or slightly older children. If a family lives in a foreign country or locality, the children will start speaking with a local accent, not with their parents' dialect.

Language learning is universal. At the beginning of life, any person can absorb absolutely any language with any set of sounds. Later, this ability is lost, and it becomes much harder to understand the difference between two phonemes of a foreign language.

The development of speech centers in the brain begins even at the fetal stage. Already at 12–16 weeks of pregnancy, those areas in the left hemisphere begin to thicken, which will become the Broca's area and Wernicke's area. These areas are pronounced by the 30th week of pregnancy.

After birth, in the first 1–3 months of life, the response to all phonemes is approximately the same.

Although the child does not speak yet, his brain is not an amorphous vegetable, but identifies the structural features of the language. By the 3rd month of life, the child prefers to hear real spoken speech rather than a random stream of syllables. At the age of 3–6 months, children pay more attention to vowels.

At about 6 months of age, the ability to identify statistical patterns of the language is discovered. This was found as follows: statistically similar combinations of syllables were reproduced in front of the child, where one sound was usually followed by another. When the child heard a syllable from an unusual combination of sounds that had not been encountered together before, his pulse quickened.

In the first two years of a person's life, the axons of neurons are most intensively covered with a myelin sheath. It does not form everywhere at once. The Wernicke's area is myelinated in the ninth month of life — about 3 months earlier than the Broca's area. Children begin to understand language before they reproduce it. It is around 9 months that children start babbling: the child repeats sounds that they have heard somewhere.

Why does a child eventually master the language? Discussions of this process took place in the 1960s.

In the camp of behaviorists and their prophet B.F. Skinner, a theory reigned that the reader can guess without hints. It was believed that children receive a reward for reproducing the correct sequences of sounds. To simplify to the point of absurdity, saying the word "milk" would get them fed, while saying the indistinct "bobobo" would not get them food. The blood glucose level would drop, and the child would make another attempt to correctly say the cherished phrase. In the traditions of classical behaviorism, conditioning would occur.

Skinner was opposed by Noam Chomsky. At that time, behaviorists completely dominated American psychology, and the struggle was unequal. But it was Chomsky's versions of the theory that won.

As Noam pointed out, between 15 and 20 months of a child's life, a key event occurs: the child utters aloud the first combination of words that they have never heard from others before.

Creating something new, which the child could not receive from outside, cannot be explained by the theory of behaviorism. Then, for extended periods of life, the child learns about ten words a day. This also does not fit into the theory of behaviorism.

The child receives disparate and unclear fragments of empirical data, does not imitate particular cases, and does not learn the entire grammar of the language through a single induction. In the book "Rules and Representations," Chomsky introduces the principle of the poverty of the stimulus, according to which part of the linguistic abilities is innate in a person.

It is after the first independent phrase that the exponential explosion of language acquisition begins: the child starts constructing their own combinations of words, rather than repeating someone else's. At a certain stage of development, children generalize the rules of the language based on imperfect examples. This was tested through real experiments: in the first months of life, they followed children and recorded absolutely everything they said.

Deaf and Language

Deaf and hard-of-hearing people communicate in sign language. This is the only thing that the layman knows.

There is no single sign language. The page with the list of sign languages in "Wikipedia" claims that there are about three hundred of them in the world.

It is important to understand that although the dactylology of sounds for proper names is present, sign language is not a translation of individual words of the sound language into a gesture, but a completely separate phenomenon with its own grammar and lexicon.

Sign languages are culturally complete:

• Accents and dialects were formed due to regional remoteness and segregation of population groups. One of the most striking examples of this is the Black American Sign Language, which originates from the regular ASL. Among other things, the variation differs in vocabulary, speed, overall style, and the use of both hands.

• Rhyme exists in sign languages. Some signs are similar to others and resonate with them, so songs and poetry are possible (albeit weakly developed).

• Puns are also based on the similarity of signs. For example, in the first few lessons of American Sign Language, a beginner will be introduced to the following amusing pattern.

  The sign for "milk" in ASL looks similar to the process of milking cattle.

  At the same time, the sign for "pasteurized milk" clearly plays on the similarity of the English "pasteurized" with "past your eyes." Here, the sign for "milk" is made along the eyes.

Is it necessary to explain that the process of learning sign languages from a neurobiological perspective looks exactly the same? The same areas of the cerebral cortex are activated. A child born deaf, who regularly sees communication in sign language, begins to babble at the same age of 9 months, but with their hands.

The brain of those who know sign language as their first language works the same way as that of a person speaking with sounds. The same Broca's area and Wernicke's area are responsible for sign language proficiency. Their damage after a stroke or injury will lead to the same effects of aphasia. During communication in sign language, the auditory cortex is activated. Knowledge in sign language is stored in the same regions of the brain.

Sound language is not exhalations and not precise movements of the lips and tongue, but the content of the transmitted message. Similarly, sign languages focus on meaning. If a speaker of a sound language is asked to pronounce a certain sequence of words from memory and simultaneously perform some distracting movement (move their legs in a certain order), speed and accuracy will suffer. The same will happen in communication in sign language with someone who is proficient in sign language as their primary language. At the same time, the intensity of this effect will coincide.

Everyone knows that it is easiest to learn a second language as a child, and if you start after the age of 12, a person will forever have an accent of their native language in their spoken foreign speech. It is less known that new languages are also primarily invented by children.

Of course, the rules of the language are fixed by some official organization, for example, the Institute of the Russian Language for the Russian language or Gallaudet University for ASL. In the latter case, the organization even invents new words.

But it was children who invented the modern Nicaraguan sign language. Until the 1970s, there was no unified sign language in Nicaragua. In 1980, after the Sandinista revolution, a vocational school for deaf teenagers was opened in the Villa Libertad area of Managua, where children were taught Spanish and lip reading. They were not taught sign language. Teachers communicated with teenagers using dactylology - transcription of the alphabet with gestures. At the same time, the students themselves communicated with each other in sign language in their free time.

As a result, a pidgin Lenguaje de Signos Nicaragua (LSN) was formed, which is still used by those who attended the school at that time. Moreover, LSN continued to evolve. The next generations of students took this pidgin and increased its grammatical complexity, giving birth to the language Idioma de Señas de Nicaragua (ISN). In just three generations, children at the school created a full-fledged language that is now taught to deaf children in Nicaragua. The process is well described in a famous 1994 article

How monkeys learn language

It is easy to think that only humans have Broca's area and Wernicke's area in the brain. And so it will be. Of the creatures living on Earth today, only humans can speak.

The brain is a set of soft tissues, its remains are not preserved. However, the endocast, the relief on the inner side of the skull, of our australopithecine ancestors reveals hemispheric asymmetry similar to that found in modern humans.

The speech apparatus developed a long time ago, its primitive rudiments are found in some primates. Modern monkeys also have small thickenings of neurons in these areas of the brain in the left hemisphere, but not as pronounced as in humans. The prototype of the Broca's area in monkeys is associated with areas of the brain responsible for facial expressions — just like in humans.

These fragments in the left hemisphere with thickenings of neurons in monkeys do not lie dead weight:

• It is known that the left hemisphere controls the movements of the right side of the body, and the right hemisphere controls the left. When a human child starts babbling at the age of 9 months, the facial expressions are more pronounced on the right side of the face.

• Similarly, when monkeys scream, the facial expressions are more pronounced on the right side of their faces. Brain scans during monkey screams and the perception of screams from their kin reveal the activation of brain areas in the left hemisphere.

Moreover, animal communication is multimodal: facial expressions are inseparable from sound signals. A rhesus macaque will be extremely discouraged (as determined by heart rate) if shown an image of a kin with a friendly facial expression accompanied by a danger scream.

Linguist Charles Hockett identified 16 elements of human language: semantics (the connection of sound signals with specific meanings), arbitrariness, discreteness, reflexivity, displacement (the ability to talk about the physically non-existent, describe the past, and dream about the future), and so on.

Yes, only human communication is characterized by displacement. But some animal species also demonstrate the semantics of communication: one scream can mean danger from above, another from below.

Why not teach animals to communicate in human language? What experiments have been conducted on this topic?

It is easier and simpler in such complex matters to rely on university materials. The famous neuroendocrinologist and primatologist Robert Sapolsky is known for his work with monkeys. Robert gave a series of lectures on the biology of human behavior at Stanford University. In 2011, they were posted on YouTube. In one of the less popular videos, Sapolsky goes into detail about language and monkeys.

One of the first talking monkeys was the chimpanzee Vicki. In the early 1950s, she was raised as an ordinary human child. On film, she reproduces four words ("mama", "papa", "up" and "cup").

The training looked rough. Based on the then-breakthrough principles of behaviorism, Vicki received food and water for reproducing the correct sounds. If there were no cherished sounds, the chimpanzee was not fed. As Sapolsky says, as a result of such upbringing, Vicki turned out to be a nervous monkey. Fortunately, after a few years, the experiments were stopped.

There were no clear attempts to analyze the phonetics of Vicki's words until recently. According to a 2023 scientific article (doi:10.1007/s10764-023-00350-1), even the babbling of this individual is more like playful muttering.

Similar experiments were conducted in the 1930s by Winthrop Kellogg. Inspired by the Indian feral children Amala and Kamala, he decided to humanize a 7.5-month-old chimpanzee named Gua.

Let's remember: people learn language from their peers and slightly older children, not from their parents. Kellogg went further than Vicki's adoptive "papa" and "mama" would later do. He not only raised Gua as a human, but also raised her together with his own 10-month-old child Donald as brother and sister.

At about one year old, Gua outperformed Donald in some simple tasks, but even here characteristic differences were noticeable. Donald recognized people by their faces, Gua by their smells and clothes. When it came to language acquisition, the human child confidently outperformed the chimpanzee child.

Ultimately, this experiment did not lead to anything productive, except that Donald began to reproduce chimpanzee sounds. Humanization did not work. The experiment was interrupted nine months after it began. At the end of the experiment, Gua was sent to a laboratory where she developed pneumonia, and less than a year later she died.

In general, Kellogg's tests are striking in their unethical nature and complete lack of safety measures. In the video below at the 4:30 mark, Gua and Donald are sitting outdoors. Then the father takes out a gun and fires (hopefully blanks) half a meter from their ears. The experiment was dedicated to the reaction speed of the monkey and the human.

The larynx of monkeys is different from that of humans. Perhaps they cannot reproduce language because they do not have the appropriate anatomy? The solution is much simpler: teach them sign language!

This idea was born to Allen and Beatrice Gardner, and they tested it on the chimpanzee Washoe in the 1960s. It was claimed that the monkey mastered hundreds of gestures - 250 or 350 depending on the source. With these gestures, Washoe communicated with the Gardners.

Literature about Washoe was captivating. The articles stated that Washoe demonstrated one of the signs of human language: the invention of new words. Upon seeing a duck, the chimpanzee combined two gestures "water" and "bird" to denote it.

Washoe could lie. As Sapolsky recounts, she lived in a trailer on the university campus, where her gestures were filmed in the evenings. Among other interior items in the trailer, there was also a refrigerator, which Washoe was forbidden to open. Sometimes something amusing happened: in the silence, one could see the chimpanzee sneaking towards the refrigerator and repeating with gestures "quiet Washoe quiet Washoe".

Washoe was given a fellow chimpanzee, who, it was claimed, learned her language in an equally interesting way and began communicating with her after a couple of months. How interesting did this achievement look in real life?

Robert Sapolsky recalls that the first polished sentence was exactly what every chimpanzee loves: "tickle me". All chimpanzees love to be tickled. Working with chimpanzees is fun: you can tickle them all day, and nothing will happen, as they do not reciprocate. By the way, the gesture "tickle" resembles pointing to a wristwatch, and it is this gesture that is so often found in Koko's "message" to the UN.

This is how this first conversation looked: Washoe sat opposite the fellow chimpanzee and declared: "Tickle me". The fellow chimpanzee responded no less meaningfully: "Tickle me". Washoe said: "Tickle me me me me me". In response, the fellow chimpanzee hinted: "Tickle me tickle me tickle me tickle me". Finally, the chimpanzees stood up and walked in opposite directions.

However, the scientific community was ecstatic about the monkeys communicating in ASL. It should be immediately noted that the primate language had several differences from human sign language. The human hand has more precise movements, so some gestures are beyond the monkeys' control and were simplified to what is called Gorilla Sign Language in English. At the same time, one cannot say that a person is fluent in a foreign language simply because they have mastered a thousand words in it.

But everyone said that the monkeys spoke ASL.

In the seventies, the best of the talking monkeys appeared. Graduate student Francine (Penny) Patterson decided that gorillas should be taught human languages. Indeed, if you compare the temperament of monkeys, chimpanzees are fidgety children with ADHD. For her purposes, Penny obtained a female named Koko from the San Francisco Zoo.

Patterson's work with Koko was full of achievements. It was said that Koko talked about her dreams and gossiped. The latter means that if Koko saw two lab assistants quarrel, she could later tell about it. Koko was allegedly capable of thinking and lying. However, all these talking monkeys had the latter trait.

Koko's fame extended far beyond scientific circles. In 1978, Koko's popularity peaked. On the cover of National Geographic magazine appeared a selfie of the sensational gorilla, which Koko took in the mirror. (The issue of monkey selfie copyright was still more than 35 years away.)

However, in the early eighties, it all came to naught.

A few years earlier, Herbert Terrace began his own experiment with a chimpanzee named Nim Chimpsky. The chimpanzee's name is a pun on the English abbreviation "chimp" (chimpanzee) and the name of Noam Chomsky (Chomsky).

Chomsky's theory stated that only humans are capable of learning to speak. With Nim's help, Terrace wanted to refute this claim.

Like all previous apes, Nim studied a variant of American Sign Language. While Washoe lived on the West Coast of the United States, Chimpsky lived in a house in Manhattan, New York. Another difference, for which the experiment would later be frequently criticized, was the rotation of trainers. At the age of two, the chimpanzee was taken from his familiar family of surrogate human parents and then subjected to training by different instructors. Sapolsky recalls that he himself spent an entire summer tickling Nim.

After three years of experiments, the achievements of the animal language world were overthrown. Terrace stated that primates do not speak and that Chomsky is right.

In the first serious analysis of this field (doi:10.1126/science.504995), it was indicated that most of Nim's gesture sequences were responses to gestures from humans. Nim interrupted the instructors' gestures much more often than a language-learning child interrupts their parents' speech.

In the 19 months of training preceding the publication of the 1979 article, Nim's lexicon size nearly tripled from 42 to 125 gestures. However, the length of the combinations did not grow by even 50%: the average indicator of 1.1 gestures increased only to 1.6. While some two-gesture sets were similar, Nim's three-gesture combinations showed no signs of lexical patterns.

This is all statistics. Real word sequences are pathetic and primitive. The tables clearly show that the chimpanzee was simply brute-forcing to get food and asking to play with him.

The more words in a human's message stream, the more meaning is embedded. But Nim's long phrases were not more meaningful.

The scientific article gives an example of a phrase of 16 gestures: "give orange me give eat orange me eat orange give me eat orange give me you". To convey "give orange", two gestures would have sufficed. It is easy to imagine that the chimpanzee was simply very hungry.

Nim's communication was not spontaneous, his phrases were a reaction to "if you answer, you get a piece of Fruit Loops cereal". By the way, many project participants were quietly chewing these Fruit Loops intended for the subject.

More importantly, the article talked not only about its own failure - video recordings of other subjects were analyzed. Terrace's conclusion stated: all monkeys master visual symbols, but they do not have the ability to combine them into new meanings. In the statements of the monkeys, a similar lack of a creative process and dependence on the gestures of the trainers was found.

Washoe did not invent a word for duck. The explanation for the two gestures is much simpler: these are two separate gestures "water" and "bird", caused by the fact that the chimpanzee saw a duck swimming in the water.

Researchers critically reviewed the videos with Washoe. It was noted that the chimpanzee simply repeats Gardner's gestures. All of the multi-gesture phrases that Washoe produced were immediately encountered in the instructor's sequence of gestures before the monkey used them. The same was true for Koko.

Finally, a few words need to be added about Koko the gorilla's IQ testing. This refers to the statement for which Facebook¹ banned Christopher Langan. The animal did not receive the usual test for adults, but the Cattell children's test.

As Patterson herself writes, the IQ data obtained are not suitable for comparison with humans. An example of the futility of this test is given next to the text: once an orangutan scored an IQ of 200 in such a test, and this largely signals the good development of its motor functions. Comparing the IQ of a gorilla and the inhabitants of Africa is simply impossible.

Researchers were more interested in the equivalent of Koko's age in human years. It is indicated that by May 1977, at the age of 5.5 years, the gorilla showed the equivalent intelligence of a human child who was 4 years and 8 months old.

“The smartest man in the world” Christopher Langan just scores high on IQ tests. This is a curious example of the futility of IQ tests, as this person does not engage in intellectual work in everyday life. Langan was a construction and farm worker, and also worked as a bouncer in a club for 20 years. Christopher has no expertise in human intelligence.

There is nothing new in the meme. Comparing the IQ of monkeys and some Somalis is something that political extremism enthusiasts love. The only thing to argue about here is Facebook's moderation policy¹.

What to learn from Koko

Francine Patterson continued to work with Koko, formally stealing the monkey. Initially, she was promised the animal for only a few years, but Penny simply refused to return the gorilla, arguing that Koko now possesses human values. The matter smelled of a lawsuit for the emancipation of the gorilla. The zoo anticipated the impending storm of public opinion and did not demand the return of the animal. Probably, the $60,000 cost of the animal was cheaper to write off.

Soon conflicts began between Patterson and Terres, which quickly escalated to the level of personal remarks. Terres claimed that Patterson had no idea what a controlled experiment was. Penny responded that Terres was a cold fish who drove Nim to autism with a chaotic series of 20-30 instructors a year. If there was a grain of bitter truth in this, then in the end Francine generally stated that Terres was simply unlucky - Nim was supposedly stupid.

Francine managed to continue being considered a scientific researcher without clear scientific publications. In 1976, she founded the Gorilla Foundation, a fund that sponsored her research further. The organization's website contains a list of various works. However, Sapolsky claims that instead of solid scientific data, the public was presented with a few touching documentaries about the gorilla such as "Koko and Global Warming" or "Koko and the Trade Deficit". Full videos of communication remained for internal use only.

The way Patterson communicated with Koko in these films looked funny. Sapolsky recalls that Penny showed the gorilla an object and asked, "What is this called?" The ape made an absolutely incorrect gesture. Francine responded, "Come on, you're joking." The question was repeated with another object, and the incorrect answer only made the researcher smile: "Oh, you're a funny gorilla." Patterson explained Koko's incorrect answers as jokes.

It is important to remember that Patterson could have been sincerely mistaken. It is quite possible that Penny herself believed that Koko understood her and responded with gestures—after all, the trainer of Clever Hans also believed that the horse could add and multiply in its mind. In reality, the horse was simply observing the trainer's body language and the audience's emotions.

Similar "communication" is observed in any public appearance of Koko. For example, here is a fragment of a chat on April 27, 1998, with Koko on AOL, the humor of which was noted by the media. At that time, Internet chats were still a novelty, and here they were also streaming video via webcam.

DrPPatrsn: Koko, do you like talking to people?

LiveKOKO: Good nipple.

DrPPatrsn: Yes, that's her response. "Nipple" rhymes with "people," okay? She doesn't like showing the sign for "people," so she might be trying to replace it with a similarly sounding sign, but she is indicating that communicating is "good."

By the way, the appearance of the word "nipple" is not random. For some unknown reason, as an adult, the gorilla developed a strange fixation on human nipples. She loved looking at and touching other people's chests, whether male or female. This applied both to the staff and to visiting celebrities. Robin Williams even talked about being grabbed by the nipples by the gorilla in one of his stand-up routines.

In 2005, the Gorilla Foundation had to settle a lawsuit with two former employees. Two new female staff members were asked to expose their breasts to Koko. Patterson explained that Koko was constantly seeing Francine's breasts, and she was bored of her nipples, so she needed to see new ones. Similar complaints came from other staff members of the foundation.

Returning to the chat: As is known, rhyming in sign languages is based on the similarity of signs, not the pronunciation of words in spoken languages. The signs for "nipple" and "people" in American Sign Language have nothing in common (1, 2). If a person (or any other being) thinks in ASL, it's unlikely that these two words would rhyme in their mind.

If you read through the entire chat log, it becomes noticeable that Patterson simply imagines the meanings of Koko's gestures. The monkey sits and scratches, Penny entertains the chat with interpretations of a few short combinations of gestures. In some cases, the person — accidentally or deliberately — prompts the monkey what to say: "Shall we say goodbye?"

During the pre-training stage, the large language model (LLM) receives all the texts available to researchers of humanity, based on which it independently learns the rules of human written speech. However, this text is not about the similarity of training neural networks and human children.

Can large language models reason? Adherents of universal artificial intelligence promise that they definitely can or will be able to in the near future.

It is easy to test the ability to reason. You need to take the same task and replace some words and concepts in it, for example, the names of variables or concepts. If the LLM solves the task equally well in both cases, it really reasons.

No modern language model can correctly solve the modified task about the wolf, goat, and cabbage. Usually, in this most famous logical puzzle, you need to transport two animals and cabbage to the other side of the river. In the version proposed below, the wolf and cabbage were removed. The farmer just needs to get into the boat and cross to the other side.

Nevertheless, the task is beyond the power of both Claude Sonnet 3.5 and GPT-4o. They all claim that you need to cross the river more than once.

Yes, variables — in some other tasks, things are so bad that the task cannot be solved by changing the numbers.

It seems that including multiple chains of reasoning (Chain of Thought, CoT) will help achieve the desired goal. However, the preview version of o1 also cannot solve the modified version of this task. If both the goat and the wolf are made herbivores, it starts quoting the standard solution to this task, which leaves the wolf alone with the cabbage.

As Subbarao Kambhampati jokes about CoT, if you give a LLM a fish, it will be fed for one day, if you give two fish, it will be fed for two days, if you give three, then three days, and so on. In a scientific article published in May this year with his participation (arXiv:2405.04776), naive reliance on CoT is criticized. It is argued that the positive effect of the chain of reasoning is quickly lost.

Later, Kambhampati gave a lecture where he spent almost an hour repeatedly asking not to confuse information extraction with planning. The researcher claims that LLMs can neither reason nor plan. He recommends using them as idea generators or approximate sources of information.

But what difference does it make if LLMs can think? Even if this is the wrong path to universal artificial intelligence, the industry's attention will be enough for the next few years. Even if the o1 model leads nowhere, OpenAI made the right bet.

As the example of Francine Patterson shows, if a research direction is criticized as a dead end, it can still squeeze out several years of world fame.