Necessity is the father of invention
Civilization is the mother, and jugaad is miscarriage of invention. Plus, why Turing test is not a good idea.
I wrote a piece on why the Turing test is not a good way to identify intelligence without reading the Wikipedia page on the subject. When I eventually got to Wikipedia, I was mortified to learn that all the shortcomings that I wrote about were already known to those who had paid serious attention to it.
I was sheepishly preparing to abandon my essay when I learnt of a biologist named Mary Tai. Her story prompted me to write an entirely different essay. It also convinced me that my thoughts on the Turing test were worth publishing after all. This is therefore a double edition of the newsletter. To read about the Turing test, skip to the second part of this piece.
But first, about Mary Tai. So innocent was she of any knowledge of calculus that she wrote a paper reinventing the trapezoidal rule, a rule that forms the basis of integral calculus, and proudly named it after herself. The paper explaining the Tai’s formula went through peer review and was published in 1994, in an apparently reputable journal.
The discerning reader may ask if the author did not consult any previous literature on the subject, such as a Maths textbook. It turns out that she did. She read a Geometry textbook by one Irving Alder (sic) and while we will never know which of the formulas in the book she attempted to use, the paper informs us that she found it unsuitable for the purpose as the formula applied to only regular curves. There were apparently two other attempts to calculate the area under a curve, made by biologists, both of which are cited in the paper. One of them requires uniform time intervals, and it is not clear what the shortcomings of the other method were as compared to Tai’s method.
It is not known what became of Ms (Dr?) Tai’s career after 1994, but knowledge of this curious incident has caused me to reflect on the extent to which the ability to preserve and disseminate scientific knowledge is crucial to scientific progress. There is this fascinating Wired article published in May 2021 describing a few intrepid scientists’ investigation of the origins of one scientific theory - the idea that only viruses smaller than 5 microns could be airborne. This was the basis for WHO’s initial guidance that COVID was not airborne, and that masks would be ineffective against them. The investigations involved tracking down an out-of-print book and comparing the way experiments were described in that book with later interpretations, proving that the 5 micron rule was in fact a misinterpretation.
I have recently finished reading “The One Device” by Brian Merchant, a book that recounts the making of the iPhone. The book is not a must read, but the chapter on the development of Gorilla Glass is interesting.
The first shatter-proof glass was developed in 1909, and by the 1960s, a corporation named Corning had developed a variety that they attempted to sell to car manufacturers as raw material for windshields. The product was unsuccessful. The glass was too shatterproof. In an accident, glass would not crack on impact but human skulls would. In 1971, Corning shelved the product.
Decades later, when Steve Jobs threw a tantrum about the proto-iPhone getting scratched by the keys in his pocket, the engineers at Apple cast around for unbreakable and scratchproof glass. They chanced upon Corning, which, remarkably, had preserved the formula for the glass and had saved the blueprint for its manufacture.
Humanity has enjoyed extraordinary progress in the past 400 years. There is occasionally some debate about whether this progress is unprecedented. Fables have been constructed about advanced civilizations in the past that got destroyed without trace. Some people consider the fables serious hypotheses. They are obviously mistaken. Our own advanced civilization may get destroyed one day, but it is difficult to imagine all traces of plastic waste getting destroyed.
But I do believe that a weaker version of this theory is true, that stories such as that of Tai’s formula were more a norm than an exception through human history. Scientific truths were discovered, well-known among the educated and the curious, and then lost in the next war. A civilization may have advanced significantly in a particular area - the Indus Valley Civilization in sanitation and drainage, the Romans in construction methods, Ancient India in mathematics and astronomy, but the advances were inevitably lost when the next cataclysm occurred. I claim that if you go back to 1600, you will find that more scientific truths had been forgotten and technological advancements lost than were known at that time.
I believe that necessity is not the mother, but the father of invention. He provides the seed. It is civilization who is the mother. She provides the womb and nourishes it, firstly through prosperity that generates the economic surplus that allows smart people to spend time developing their insights. The second ingredient is the ability for smart people to meet and build on one another's insights rather than rediscover them every generation; as also the ability to preserve and transmit discoveries and inventions. The third is the capital to convert abstract insight into concrete invention1.
These conditions have come and gone in the past in specific regions of the world. The scale and extent of human progress of the past 400 years is unprecedented, but not the length of time. What is different about the past 400 years is that the womb that nurtures invention got a lot bigger, encompassing the entire world. Previously, a period of human progress would be followed by a longer period of regression where much of the knowledge gained was lost. In this period, countries have advanced and declined, but as we have developed better abilities to preserve and transmit knowledge and to collaborate across the world, human progress has persisted where earlier it would have reversed.
Civilizations have advanced for longer periods before. Given that the extraordinary amount of capital we have accumulated can fund weapons that can destroy the world as well, I am not confident that we will break the record this time.
The Turing test is not a good test
Colloquially, we use “Passing the Turing test” as short for “achieving artificial general intelligence”. As a concept, this is fine. We don’t have a good understanding of intelligence, but we know that it is a thing human beings possess and something that makes us uniquely human. So rather than devise a test for intelligence, we ask the question “Can we distinguish it from a human being?” as a proxy.
But when the Turing test moves from the realm of the conceptual to the concrete, it runs into Goodhart’s Law. This law states that when a measure becomes a target, it ceases to be a good measure. Challenging an intelligent machine to pass the Turing test is to set it a target. When a human sets out to devise a Turing test that can sniff out a machine hidden among humans, he is also seeking to achieve a target. Following Goodhart’s law, this will render the Turing test useless. Let me lay out the reasons.
Right now, if you ask Chat GPT if it is a human, it will answer, truthfully, that it is not. This is a rather trivial Turing test to devise and a trivial one to fail. A machine trying to pass a Turing test must be free to lie, or it wouldn’t be fair.
I am not so concerned with the ethics of programming a machine to lie. The problem though is that the machine is more or less forced to lie, but a human being does not have to. This makes the Turing test unfair from the machine’s point of view. The examiner can create a Turing test that asks probing questions about the examined entity’s life story. While the human can answer them truthfully, the machine will have to fabricate an elaborate story that is consistent and doesn't yield to the cleverest questioning. This task would be difficult for even an intelligent human being, so we must concede that it would prove difficult for an intelligent machine. This would cause it to fail the Turing test when it should pass.
There are other attacks that the Turing tester can devise. For example, machines can do calculations much faster than humans can. A Turing test could trick the machine into giving results too quickly, and the machine would have to build in a delay into its responses and even introduce the occasional error in order to better simulate a human.
Human beings get tired and their performance degrades as the day passes. A machine that displays consistently high level of intelligence will give away the fact that it is a machine, so when it pretends to be a human, it must simulate an android degradation in performance.
These are but a sample of the many ways in which a literal Turing test would fail. A good test of intelligence should successfully distinguish man from machine if and only if the machine is less intelligent than man. An actual Turing test would fail at the “only if” part.
To make it fair, we will need to limit the Turing test to only those questions that relate to intelligence. But the raison d'être of the Turing test is that it is a way to identify intelligence when we don't have a good way to define, let alone detect, intelligence.
In that, I think that the Turing test will fail, and that leaves us with no choice but to understand what intelligence is, and attempt a more rigorous way of determining if a machine has achieved general intelligence.
In the absence of this nourishment, we have jugaad, a miscarriage of invention. Figuring out that washing machines can be used to make lassi is only the first step of innovation. Using this inspiration to build a lassi maker and starting a company to manufacture and sell it is what counts as innovation. When we celebrate jugaad, we are celebrating the absence of an environment that enables this.
The middle part of the article is similar to your blog post on how nuclear weapon stockpiling and mutually assured destruction has led to preservation of knowledge as wanton wars used to periodically destroy knowledge and civilizations which had to be rediscovered anew.