Feeds:
Posts
Comments

Posts Tagged ‘ai’

The other night I was deceived by my clock. I should perhaps explain that Chez Hamlett we still have a clock which hangs on the wall. This habit originated many years ago when our son was very small and insisted on a cuckoo clock.

The rather annoying mechanical bird expired quite quickly, thank goodness, and the empty hook was taken over by a silent station clock with a pendulum. The pendulum was purely ornamental and stopped working a long time ago without affecting the function of the clock.

The station clock finally grew tatty after decades of faithful service and was replaced by a simple, and cheap, version from Ikea; this is nothing fancy and just has the basic 12 numbers and two hands.

The clock is easily visible from my desk so I tend to use it despite the competition provided by my wristwatch, computer and telephone, all of which will happily tell the time if asked.

So it came about that I was watching a particularly riveting episode of Killing Eve on Netflix, and this was followed by another riveting episode, and another, during which time apparently stood still. Surely it could not still be 10.30? Indeed it was not. The clock had stopped and needed a new battery.

This was harmless enough. I no longer have to get up to go to work so a late night is nothing to lose sleep over, as it were. It was chastening to find how easily you could be deceived by a simple piece of machinery, even when it was easy enough to check.

Of course there was no malice here. The clock was not being mischievous. Phones, keyboards and mice all need attention to their electricity needs and will refuse to perform unless regularly fed. Battery-operated items of the traditional kind need changes, the car needs petrol, and so it goes.

Yet this is the age of deceptive machinery. This is not a completely new thing. Stalin used to have purge victims painted out of official photography. People wondered whether Robert Capa’s shot militiaman was really … well, shot. One news photographer of my acquaintance admitted that he always took a small stuffed toy on disaster assignments because pictures of ruins looked so much more poignant with an abandoned plaything in them.

The distinguished war photographer Tim Page mused in his memoirs whether it was acceptable to photograph a dead soldier under a poncho, and if so whether it was also acceptable to rearrange the poncho to improve the picture.

At a more homely level I once congratulated the chief photographer of the Morecambe Visitor, where I was a naive and newly recruited sports editor, on the frequency with which he and his colleagues managed to capture the goal-scoring moment, with the ball flying past the goalkeeper’s flailing fingers. He then explained that this was due to technical skill rather than timing; the diving goalkeeper was genuine but the ball was inserted in the darkroom while the picture was being printed.

On the whole though, with due allowance for the possibility that people posed when they knew they were being pictured, photographs were regarded as more or less reliable records of what people did and looked like.

I was always more suspicious of TV and film, because of early exposure to the amount of setting up and technical preparation involved in quite simple bits of moving picture. If the scene was supposed to be “real” you were to some extent dependent on the honesty of the enterprise, and needed to remember that out of the picture were “the crew”. Over the years the crew has shrunk. In my early RTHK days there was a producer/director, cameraman, sound man, lights man, man in charge of the props, just in case any were needed, and the driver of the van required by all these people and me. Nowadays you may be interviewed by a reporter with a phone.

Anyway we have changed all that. The moment when I realised that you could no longer rely at all on video came when I watched a Youtube video of cats doing Olympic dives off a springboard. This was not an attempt at deception – nobody was going to buy the idea of human-sized cats – it was just a display of what you could do with AI.

Which is, it seems, practically anything. Nothing can be relied on any more. Idiots who believe Hillary Clinton runs a pedophile ring out of a Washington pizzeria can now order up a video of Mrs C ushering dazed kids into the basement for unmentionable purposes.

Lying by machinery has now become so common that they seem to be genuinely frightened of each other. How else can we explain the constant insistence that the ordinary human netizen should prove he or she is “not a robot”.

This is infuriating. The usual test used to be a mosaic of nine tiny pictures, which you were supposed to tick if they had a bicycle or a traffic light in them. Sometimes they wanted you to look for a ”crosswalk”, something unknown to English English. But until you had passed the picture test you could not Google it.

We now seem to have moved on to a bit of cursive writing, which you are supposed to decipher and type in the space provided. I imagine this is difficult for any robot; it regularly defeats me. Alternatively you can listen to something and type that.

The resulting burst of sound is completely incomprehensible. I must in fairness acknowledge that this may be because my VPN places me in a carefully selected and very neutral Scandinavian city. I do not speak Swedish. What tangled webs we are all weaving these days.

Read Full Post »

Happily I retired from the university teaching scene before Chat GPT and his mates came along, because it has apparently made life rather difficult.

Generally the most subtle assessment method, at least in the humanities, is the essay. You set the topic, send the student away, and assess the resulting masterpiece a week later.

People in other necks of the woods have other choices, of course. Mathematicians can be asked to tackle a problem on the spot. Subjects which consist primarily of absorbing a lot of memorable material can be tested with multiple-choice questions, which have the additional advantage that they can be graded by a computer.

The most picturesque assessment method was applied to trainee aircraft engineers. The class assembled at 9 am and each member was presented with a small engine. At 1 pm the examiner returned, by which time the engine had to be dismantled entirely into its component parts. The class then had lunch, then reassembled to reassemble their engines.

At 5 pm the examiner returned and tried to start each engine. If it started first time you had passed, and if not…

But I digress. The problem with setting essays now is that you do not know if the result was partly, or indeed wholly, written by artificial intelligence. This has in turn produced a lot of interest in detection, which is sometimes possible. The person who told me about this problem recalled a case in which the essay included the word “albeit”.

This is not the sort of word which comes up often in sociable chat so the writer was asked what it meant… and could not answer.

Clearly though there is going to be an ongoing arms race here between AI users and programmers, trying to produce ever more convincing forgeries, and teachers and other detectives looking for ways to spot traces of computer composition.

The answer, I fear, is going to be a return to the primitive practices of the past.

When I took my first degree, half a century ago, there was no continuous assessment, multiple choice was regarded as simplistic, and computers were monstrous machines which had to be fed punched cards.

We wrote essays at least once a week and these had to be submitted and discussed, but they were not part of the assessment. There was also a small examination each term intended to reassure your personal academic adviser that you were still alive and working, but this was not part of the assessment either.

After three years of this regime you encountered an ordeal more like the old Chinese civil service exam (in which candidates were locked up for a week and, according to legend, invited to write down everything they knew) than modern gentle testing methods.

Starting on Wednesday we had three-hour exams each morning and afternoon until Sunday, when you had a day off, a relic of the days when universities were mainly intended to train clergymen. The ordeal resumed on Monday and ended at lunchtime on Tuesday, at which point there was an understandable tendency for the survivors to get resoundingly drunk.

The actual examination was quite Spartan. You were allowed (times had changed) to use a Biro. The paper was white and completely blank. People were agitating for lined paper, to help authors whose writing tended to diverge gradually from the horizontal as they got down the page. This came eventually, but too late to help me.

There was a list of about a dozen questions, of which in your three hours you were expected to attempt four. As the marks available were divided equally between your four answers it was an elementary point of exam technique to get to four somehow.

There were no “open books”, bringing in notes of any kind was cheating, and the examiners were not – with rare exceptions – the people who had taught you. We were not asked our opinions of the teachers, but because the examiners were applying consistent standards, the outcomes of individuals and groups could be compared.

This was a stressful system and I would not recommend copying it in all its details. It favoured glib bullshitters and people who could write quickly. It was a great preparation for journalism, which is perhaps not a recommendation.

But as AI tramples its way across the academic landscape it may be time to rediscover the merits of putting the students in a room with paper, pen, questions and a time limit. At least you know who has written what.

Read Full Post »