A Transcript Filed Twenty-One Years Early
The Bureau’s method is to take a claim, find the record it bears on, and set the two side by side. The file opened this month in Geneva defeats that method at the first step, and does so in a manner the Bureau has not met before. There is no record. The events described have not happened.
Our correspondent gives his position as Recording Secretary of the Global Forum for Advanced Thought and encloses four documents. There is a stenographic transcript, three hundred and forty pages, of a debate held over four hours on a Thursday afternoon in September 2045 between eleven philosophers and fourteen genetically enhanced grey parrots, on the question of what constitutes meaningful communication. There is a seating plan. There is a roster of the avian delegation giving names, and neither hatch dates nor ring numbers. And there is an invoice from a stenography bureau on the Rue du Rhône, marked paid, terms thirty days net.
The invoice is the only paper in the envelope that behaves the way a document behaves. It carries a reference number, a rate per folio, a tax line and a settlement date. It is also dated twenty-one years after the postmark on the envelope that brought it, which is a difficulty no rate per folio can absorb.
The Bureau has therefore filed the claim in the only direction open to it. A statement about 2045 remains a statement about now, because everything it presupposes must be assembled out of materials that exist at the time of writing: birds that have to be hatched, an editing method that has to work in a parrot, a hall that has to be booked, and a question somebody has to have thought worth putting. Every one of those is checkable today.
The Roster, and the Question of Where the Birds Came From
Fourteen names, and nothing else: no hatch dates, no ring numbers, no studbook, no laboratory. Since the correspondent will not work forward from the birds, the Bureau has worked backward from the method.
A bird cannot be edited the way a mouse is edited, and the reason is anatomical rather than technical. Han and Park’s 2018 review sets it out: the avian zygote divides meroblastically as a small germinal disc lying on an enormous mass of yolk, and there is no practical way to introduce anything into it at the one-cell stage. What is done instead runs through the primordial germ cells, which in birds migrate to the gonad by way of the bloodstream. They are collected from an early embryo, cultured, edited in the dish, and injected into the blood of a second embryo. That second bird hatches, is reared, grows up and is bred, and the modified animal is its offspring. A whole generation of latency is designed into the method. It has been reported in chicken and quail, and by interspecies transfer in pheasant, houbara bustard, guinea fowl, turkey and zebra finch. It has not been reported in any parrot.
Then the generation itself, which is where the correspondent’s programme runs out of years. A grey parrot is incubated for thirty days, leaves the nest at twelve weeks, is fed by its parents for a further month, and does not breed until it is three to five years old. The chimera has to come of age before it can pass anything on, so three to five years go by before the first edited chick draws breath, on a species whose germ cells nobody has yet cultured. That is one turn of the wheel and one bird. Then rearing. Then teaching, which the correspondent has not costed at all.
Invited to give a date on which the programme began, the Recording Secretary replied that the Forum does not ask its delegates their age.
The Instrument in the Throat
A parrot has no larynx doing the work a larynx does in us. It has a syrinx, which sits at the base of the trachea where it forks into the two bronchi, and which has no vocal folds: the sound comes from membranes vibrating in the airflow. Because the organ straddles both branches, the left and right sides can be modulated independently, and some songbirds accordingly produce two sounds at once — a thing one human larynx cannot do.
Above it there are no lips, no teeth and no soft palate. There is a thick muscular tongue, and in 2004 Beckers, Nelson and Suthers established what that tongue is for. They took five monk parakeets, replaced the syrinx as a sound source with a small speaker playing broad-band noise, and moved the tongue. Tongue position on its own shifted two of the formants and altered the amplitude of all four present between 0.5 and 10 kHz. That is vocal-tract filtering: the operation a human mouth performs on the buzz from the vocal folds, carried out with different furniture.
The correspondent has, whether he knows it or not, picked the right animal. Vocal learning — acquiring new sounds by imitation instead of being born with them — has been found in eight groups only: humans, bats, cetaceans, pinnipeds, elephants, and, three separate times over, songbirds, parrots and hummingbirds. No non-human primate is on that list. The apes of the great language studies were taught with their hands, or a board of symbols, because their throats would not do the work; Viki, raised in a household and drilled for nearly seven years, produced four words. A parrot is in the club and a chimpanzee is not.
Which makes one silence in the file peculiar. Three hundred and forty pages record fourteen birds speaking in continuous prose, with no note anywhere of pitch, of duration, of overlap, or of two voices from one delegate. Asked about it, the Recording Secretary replied that the stenographer had been instructed to render substance rather than sound.
What a Parrot Has Actually Done
The Bureau comes now to the one part of the file where the record is generous, and it is a good deal more generous than the correspondent appears to know.
Alex was a grey parrot, hatched on 18 May 1976 and bought from a pet shop. Irene Pepperberg worked with him for thirty years by the model–rival method: two trainers, one instructing while the other models right answers and wrong ones and competes with the pupil for the instructor’s attention, the roles traded back and forth so the bird sees the exchange from both ends. Over those thirty years he acquired a vocabulary of more than a hundred words, named fifty objects, distinguished seven colours and five shapes, handled bigger, smaller, same and different, and counted quantities up to six. In July 2005 Pepperberg reported that he had the concept of zero: shown two identical objects and asked what was different about them, he answered, “None.” He coined “banerry” for an apple, which a linguist read as a banana crossed with a cherry, the two fruits he already had. Looking into a mirror he asked what colour he was, was told grey six times, and so became the first non-human animal on record to have asked a question.
He was found dead in his cage on 6 September 2007, aged 31, in a species that keeps forty to sixty years in captivity; the necropsy gave no discernible cause, and the death was afterwards put to an arrhythmia, a heart attack or a stroke, associated with arteriosclerosis. The last thing he said to Pepperberg was the thing he said every evening as she left the laboratory: “You be good. I love you. See you tomorrow.”
The Bureau prints the objection beside all of that, the objection being part of the record too. Herbert Terrace held that what Alex produced was a complex discrimination under stimulus control — operant conditioning performed extremely well, and not the use of language. Thirty years of daily work, and the argument is still open. The Geneva transcript disposes of it on page four.
The Dates the Transcript Sits On Top Of
A hundred and fifteen years of controlled work on one question, and the line the correspondent has added at the bottom of it.
Oskar Pfungst takes Clever Hans out of his questioner’s sight. Eighty-nine per cent correct becomes six per cent.
The Gardners begin with the chimpanzee Washoe, who uses 85 signs reliably by her third year and 132 later.
Nim Chimpsky is born, and Herbert Terrace begins the project named after the linguist it was meant to refute.
Alex hatches. Irene Pepperberg buys him from a pet shop the next year, and the thirty years begin.
Terrace publishes: no evidence of an ape’s ability to use a grammar. Nim’s longest utterance runs to sixteen signs.
Beckers, Nelson and Suthers put a loudspeaker where a parakeet’s syrinx was, and catch the tongue doing the filtering.
Asked what is different about two identical objects, Alex answers “None”, and is credited with the concept of zero.
Alex dies at 31, in a species that keeps forty to sixty years in captivity.
A chatbot in the persona of a thirteen-year-old convinces a third of thirty judges in five minutes, and is announced as having passed the Turing test.
Olkowicz and colleagues count the neurons: bird forebrains hold primate-like numbers of them, in brains packed two to four times as densely as a mammal’s.
New York’s highest court declines, five to two, to extend habeas corpus to an elephant.
Fourteen parrots debate eleven philosophers for four hours on the nature of meaningful communication. No victor is declared.
The Control the Transcript Does Not Contain
The Bureau put one question to Geneva, and it is the only question in the file that would decide anything: during the debate, did the man asking the questions know the answers he was after?
Clever Hans was a horse in Berlin who did arithmetic by tapping a hoof, and his owner, Wilhelm von Osten, was no fraud. He believed in the horse, which is exactly what makes the case the permanent standard it became. In 1907 Oskar Pfungst pulled apart the two things that had always travelled together. With a questioner in view who knew the answer, Hans was right 89 per cent of the time — fifty of fifty-six. With the questioner out of sight, or ignorant of the answer himself, he was right 6 per cent of the time — two of thirty-five. The horse had been reading involuntary shifts of posture in a man who did not know he was making them. Everyone in the room was honest, and the result was worth nothing.
The remedy is not a fuller transcript. It is less contact. When the bonobo Kanzi was tested, the interrogator was out of sight and the questions arrived through headphones; over 180 trials on 35 different items he was right 93 per cent of the time, and the 93 means something only because of the headphones. Nim Chimpsky was tested the other way, in a room full of prompting teachers, and his longest recorded utterance ran to sixteen signs: “Give orange me give eat orange me eat orange give me eat orange give me you.” Terrace published the finding in 1979 — no evidence of an ape’s ability to use a grammar.
The seating plan from Geneva puts the moderator four metres from the perches, in full view, with his question sheet on the table in front of him. Asked whether the moderator knew the answers to his own questions, the Recording Secretary replied that a moderator who did not would be poorly prepared.
The Imitation Game, Held in the Open
There is a test for this, and it is older than the file. In 1950, in “Computing Machinery and Intelligence”, Alan Turing proposed replacing the question whether machines can think with a party game: an interrogator, a text channel, and no sight or sound of the party at the other end. He guessed that in about fifty years a machine would play the imitation game well enough that an average interrogator would have no better than a 70 per cent chance of identifying it after five minutes.
That bar has since been cleared by things nobody takes for minds. On 7 June 2014, at the Royal Society, a chatbot called Eugene Goostman convinced 33 per cent of thirty judges in five-minute conversations and was declared to have passed. It was written as a thirteen-year-old boy from Odesa with a pet guinea pig, a persona chosen — its author said — as not too old to know everything and not too young to know nothing. Bad grammar became charm; gaps in knowledge became youth. The machine did not rise to the judges. The judges were lowered to the machine.
The Geneva debate is an imitation game with the screen taken away, and it never puts the question that would settle anything. Half a century of argument about what is peculiar to human language narrowed, in a 2002 paper by Hauser, Chomsky and Fitch, to one candidate: not vocabulary, not naming, not counting, but recursion — embedding a structure inside another of the same kind, without a natural stop. Nobody in three hundred and forty pages asks a bird to embed a clause.
Searle settled in 1980 what a transcript is worth on its own. A man alone in a room with a rulebook returns flawless Chinese and understands none of it, because syntax is not sufficient for semantics. The Bureau takes no view on the argument. Its lesson for a filing clerk is exact: a perfect transcript is consistent with a mind behind it and equally with nothing behind it, which is precisely why it cannot tell you which.
The Lion, and the Register He Speaks In
Wittgenstein wrote that if a lion could talk, we could not understand him. The remark is not about a lion’s throat. It is that understanding runs on a shared way of living, and that a creature whose whole existence is arranged differently could use words we were perfectly able to parse and mean by them something we could not follow.
The most remarkable feature of three hundred and forty pages from Geneva is that everybody understood everybody perfectly. The birds argue in idiomatic English, in complete paragraphs, at the length and in the register of the men across the floor from them. They cite. They concede a point and come back to it later. They handle consciousness, autonomy and the good in the vocabulary the philosophers brought into the room with them, and they take their turns without once talking over anybody, which is more than the transcript records of the philosophers.
The exchange the correspondent is proudest of is the one that gives him away. A philosopher asks about the nature of freedom, and a parrot answers with a remark about flight and the confines of a cage. It is a good line. It is also the image a man reaches for when he is asked to imagine what a caged bird would say.
What the Bureau Is Able to Certify
The Bureau is able to certify that parrots are vocal learners and apes are not; that a grey parrot named Alex learned more than a hundred words, named fifty objects and answered “none”; that bird forebrains carry primate-like numbers of neurons, in brains packed two to four times as densely as a mammal’s; and that people have been running careful controls on this question since 1907. It is not able to certify that any of it has ever produced a debate.
The communiqué at the back of the transcript asks that the delegates be recognised as persons. There the record is clear and recent. The Nonhuman Rights Project has petitioned American courts for two chimpanzees, Tommy and Kiko, and for an elephant named Happy; on 14 June 2022 New York’s highest court refused the elephant, five to two, holding that nothing in the precedent of any state or federal court supported extending the writ to a non-human animal. That is where the matter stood twenty-three years before the communiqué was drafted.
We wrote once more and asked for one thing that could be weighed: a recording, a spectrogram, a moulted feather with a ring number on it, a hall booking. The Recording Secretary replied that the Forum keeps no recordings, a recording of a debate not being the debate; that the delegation had declined to be ringed; that the hall was booked in the caterer’s name; and that we would find all of it a great deal easier in twenty-one years, when he suggests we write again. He asks that we file the transcript now, so that it is ready.
The file remains open. It cannot close before 2045.
