When someone mentions ‘Artificial Intelligence’ I usually ignore them completely. But the subject comes up so frequently that it is impossible to ignore. Andy Haldane fretted a few weeks ago that AI might cause ‘widespread unemployment, despite an Organisation for Economic Cooperation and Development report earlier this year saying that fears of mass unemployment for automation were ‘overblown’ as most jobs were harder to automate than previous studies had suggested.
One of the better articles came out yesterday (9 October) quoting Roger Schank who predicts a new AI winter, ‘a reference to the period in the early 1980s when disappointment over the progress of the technology led to a retreat from the field’.
Schank is spot on. I studied AI in the 1980s when it had become clear that the 1970s optimism over artificial intelligence was misplaced. But the winter was long before that. As I noted in a presentation to the FSA in 2001, proposals for mechanical translation pre-date the invention of the digital computer. The first recognisable application was a dictionary look-up system developed at Birkbeck College, London in 1948. Researchers had been involved in code-breaking during the Second World War, and thought that documents written in one language could be viewed as having been written in code, i.e. Russian is simply ‘English in code’. This made it seemed like a problem in cryptography. ‘When I look at an article in Russian, I say: “This is really written in English, but it has been coded in some strange symbols. I will now proceed to decode.”’
US funding of Machine Translation research cost the U.S. public $20 million by the mid 1960s. The Automatic Language Processing Advisory Committee (ALPAC) produced a report on the results of the funding and concluded that ‘there had been no machine translation of general scientific text, and none is in immediate prospect’.
I doubt this has changed. In the presentation I noted an amusing Google translation of ‘Seinen Lebensabend verbrachte in bad kleinen, in der Nähe seiner Geburtsstadt Wismar’ as ‘His life was spent in small bathroom, near his hometown of Wismar’. Google should check with Wikipedia first.
I also noted the following well-known example:
The teacher sent the boy to the headmaster because
- he wanted to see him
- he had been throwing stones
- he was fed up with his bad behaviour
Humans instantly understand the pronoun back reference, which is nothing to do with syntax, but computers get puzzled. For a separate project I contacted a user of Stanford CoreNLP tagger, ‘a state of the art neural system’, giving two similar examples to solve. Deep Thought got both of them wrong, although the Hobb algorithm, developed in the 1970s, got one of them right. Yet pronoun resolution (as it is called) is key to understanding any commercial document.
At the end of the FSA presentation, in a slide titled ‘Solving the Problem’, I commented ‘this page deliberately left blank’. No solution yet, but the good news is that humans will remain in charge for now.