91av

Siri on the iPhone is smart, but not smart enough to pass for a human

Alan Turing was a visionary thinker on artificial intelligence (AI), devising the Turing test, which is still used as a key gauge of how close machines have come to human intelligence. He also published prescient ideas about simulating a brain with computers.

Toward the end of his life, he was also beginning tantalising work in biology, devising a mathematical theory of “morphogenesis” – in essence, how a leopard gets its spots

Artificial Brains

Turing was curious about the brain. He believed that the infant brain could be simulated on a computer. In 1948, he , and in doing so gave an early description of the artificial neural networks used to simulate neurons today.

His paper was prescient, but was not published until 1968 – years after his death – in part because his supervisor at the National Physical Laboratory, Charles Galton Darwin, described it as a “schoolboy essay”.

The paper describes a model of the brain based on simple processing units – neurons – that take two inputs and have a single output. They are connected together in a random fashion to make a vast network of interconnected units. The signals, passing along interconnections equivalent to the brain’s synapses, consisted of 1s or 0s. Today this is called a “boolean neural network”; Turing called it an unorganised A-type machine.

The A-type machine could not learn anything, so Turing used it as the basis for a teachable B-type machine. The B-type was identical to the A-type except that the interconnections between neurons had switches that could be “educated”. The education took the form of telling a switch to be on (allowing a signal to pass down the synapse) or off (blocking the signal). Turing theorised that such education could be used to teach the network of neurons.

After his death, Turing’s ideas were rediscovered and his simple binary-based neural networks were shown to be teachable. For example, they can learn to recognise simple patterns like the shapes of Os and Xs.

Later, independently, more complex neural networks became the focus of AI research and they are used to this day in robotics, pattern recognition and game playing.

Mathematical Biology

How does the leopard get its spots? Why is the black-and-white pattern on the skin of a Friesian cow haphazard and asymmetrical? These were questions that intrigued Turing. In the 1950s, he found an answer that had previously eluded biologists.

From childhood, Turing had been fascinated by mathematical patterns in nature, and, in particular, the recurrence of the Fibonacci sequence in plants. The Fibonacci sequence is 1, 1, 2, 3, 5, 8, 13 and so on. Each number in the sequence is the sum of the previous two. Turing had counted the number of petals on daisies, and found the total would always add up to a Fibonacci number. The same applies to other flowers. He believed that the spiral pattern on the head of a sunflower was determined by the Fibonacci sequence.

“Turing counted the number of petals on daisies and found they always added up to a Fibonacci number”

As an adult, Turing became curious about a biological concept called morphogenesis – essentially, how does a living body take shape as it grows? At the time, a fundamental practical question for life was: how did a single, simple cell with a simple shape take on the complex form of a living being just by dividing? Why didn’t division simply result in ever more copies of the same cell? How did odd shapes like hands or eyes develop?

In attempting to answer those questions, Turing turned to mathematics. He posited the existence of chemicals he termed “morphogens” that cause the asymmetry seen in living beings. His 1952 paper “The chemical basis of morphogenesis” is now seen as the starting point for ().

Turing believed that specific chemical reactions were responsible for the irregular spots and patches on the skin of animals like leopards or cows, and the ridges inside the roof of the mouth. So he used the first computers to simulate the process that he thought might be occurring.

Turing modelled a pair of interacting chemicals undergoing diffusion and reaction. The two chemicals diffuse, or spread out, at different rates resulting in varying concentrations of the two. They also react with one another to produce more of the same chemicals, which in turn diffuse and react.

Turing’s simple combination of reaction and diffusion results in striking, irregular patterns. He theorised that if one of the chemicals caused activity in cells, and another prevented it, the result would be the types of patterns seen on the coats of animals. In his 1952 paper, he revealed a computer simulation of reaction and diffusion that resulted in patterns similar to those seen on Friesian cows.

But his idea remained just a theory. Actual morphogens were not found during his lifetime. In January 2012, a group of researchers at King’s College London showed that Turing had been right and that . They showed that two chemicals, acting as Turing predicted, control the formation of ridge patterns inside a mouse’s mouth ().

The Turing Test

In 1950, Turing described what we now call the Turing test. To this day, his test is a standard by which “intelligent” machines are judged, and it is remarkable in its simplicity and ingenuity.

Turing referred to his method of determining whether a machine could be called intelligent as the Imitation Game. He proposed that a machine would be intelligent if it could not be distinguished from a human ().

In his test, a judge communicates with both a human and a machine in written language, via a computer screen or teleprinter. This means the judge can use only the conversation to determine which is which. If the judge cannot distinguish the machine and the human, the machine is deemed to be intelligent.

The concept will be familiar to anyone who has interacted with an artificial intelligence such as Apple’s digital personal assistant, Siri, or a chatbot. Siri does not pass the test, and although chatbots may have fooled some individuals in recent years, none have passed the Turing test unequivocally. In fact, the limitations of even the best modern AIs mean that they are quickly outed as machines. Still, Turing imagined a day when AI would prove indistinguishable from the human form.