Image: Laitr Keiows Image: Laitr Keiows

The story of Solomon V Shereshevskii, the man blessed with an amazing memory, but was lousy with faces, illustrates the extraordinary breakthrough AI has made in facial recognition.

His story begins with a talk given by the neuropsychologist Alexander Luria. The frantic scribbling of journalists, the smell of ink on paper, dominated the room as the great Luria enunciated his theories. But at the back of the room, one hack sat, listening, no pen in hand, no notes. After the talk, Luria cornered the journalist, and upon questioning, to his astonishment found that the fellow, Mr Shereshevskii, could recite Luria’s talk, word for word; he didn’t need to take notes because of his extraordinary memory. Remarkably, Shereshevskii didn’t even realise his gift for memory was unique, or indeed unusual.

It was too good an opportunity for Luria, and Shereshevskii became the subject of much of the neuropsychologist’s work.

It turned out that Shereshevskii’s memory was born out of his imagination. Objects, ideas, and names took on a distinct shape and texture in Shereshevskii’s mind. He described a musical tone sung by Luria as ‘looking like pinkish fireworks with a rough texture and an ugly taste of briny pickles”. He remembered because he had unwittingly developed a sophisticated system of mnemonics.

“One time,” said Shereshevskii, “I went to buy some ice cream… I walked over to the vendor and asked her what kind of ice cream she had. ‘Fruit ice cream,’ she said. But she answered in such a tone that a whole pile of coals, of black cinders, came bursting out of her mouth, and I couldn’t bring myself to buy any ice cream after she had answered in that way.”

Of course, computers can have even better memories than Shereshevskii. But there is one thing they have in common with the late Russian journalist. They are, or in the case of Shereshevskii were, both lousy with faces.

It is said that if someone coughed while Shereshevskii was looking at their face, then his memory of that face would always be linked with the sight of them coughing. Alas he was unable to recognise this person, unless they were coughing. Or so it is said. This became his problem. We never look exactly the same twice. It appears Shereshevskii’s memory was too good; he remembered fine detail, but when the details on a face changed, he struggled with recognising that face.

We do have a problem here. Shereshevskii died in 1958. The studies of his memory and his shortcomings were not rigorous, and much of the information we have is anecdotal.

But for many years, it seemed as if the story of Shereshevskii illustrated the problem faced by computers is being able to recognise faces.

Recognising faces is devilishly difficult, but a task that the human brain is well suit to – with exceptions such as Shereshevskii.

But it is a problem that AI is cracking. They say that computers can do the things, that we find difficult, easily, and struggle with the things we find easy, but now that AI can recognise faces, it seems that even that is changing