Artificial Stupidity

Artificial intelligence is a great buzzword and it's can be a marvellous plot device in fiction, but despite the hype, it doesn't have a thing to do with real intelligence. The term AI is routinely used for technologies such as pattern-recognition, expert systems and computer-based problem solving. They're all potentially useful and they're all making an increasingly large contribution to our day to day lives, but they all have limitations. There has unquestionably been huge progress since I started using computing in the late seventies, but the fact remains, we're still decades away from anything resembling a truly intelligent computer.

The problem with AI is that it has no genuine understanding of the real world. It doesn't matter if the technology is using standard rules-based computer code or fancy neural networks, it's still capable of the most extreme stupidity, because it's inherently limited by the information that's used to train it.

The Alexa syndrome

Ask Alexa to tell you the time and they'll give it to you accurate to the nearest second. Ask Alexa to play your favourite music track, and it'll either stream it instantly or require you to make half a dozen attempts before it manages to recognise the name. Cough in its general direction, and it could easily interpret it as a request to buy something expensive from Amazon.

It's not the programmers' fault, because the reason that systems like Alex fail is because they don't have the slightest understanding of what your words mean. Instead, they use advanced pattern recognition to match your commands against a dictionary stored on the internet. That's fine if you're asking something that thousands of people have asked before, but if you ask for something obscure, the system is guaranteed to fail with unexpected and often surprising results.

It doesn't matter if you're using Siri, or Google, or Cortana, you're still talking to something that has a worse genuine understanding of what you're saying than the average dog.

Machine translation

You might think that machine translation is a triumph of modern AI, but the reality depends on what you're attempting to translate. Ask Google to translate between a pair of related languages like French and English, and you'll probably get a good enough translation to make sense of the result. Ask it to translate a book cover from Japanese into English, and you'll be lucky if it provides you with an accurate title. If the cover has pictures, there's an excellent chance that will attempt to translate those pictures as if they were Japanese words. It's annoying for people like me, because as an origami enthusiast, I've bought a number of books written in Japanese, many of which I treasure. I can understand the instructions with no problems, as they're provided in a graphical Yoshizawa-Randlett notation that shows the folding operations in a visual form. However, since I can't read a word of Japanese, I've no idea what most of the models I've folded are actually called, and Google Translate can't help me with that because it only has a limited experience with translating Japanese origami books.

Machine translation is a handy thing to have, and when it works, it can be genuinely useful, but it's no substitute for a human being who can understand the context as well as just the individual words.

Self driving cars?

Self driving cars are still in their infancy, but despite the description, they have no sense of "self". Armed with a battery of high-tech electronics, including 360 degree cameras and advanced radar systems, they're still struggling with coping with unexpected situations or imperfect conditions such as rain or fog. If the roads were not full of vehicles driven by unpredictable humans, if there were no pedestrians, and if the driving conditions were guaranteed to be perfect, I'm sure that self driving cars would be a pretty neat idea. But after decades of automation, we rightly insist that aeroplanes have pilots, because although they could theoretically fly the plans by clicking a couple of switches, they're vital security when the real world pops up with something surprising.

Real intelligence

A few years ago, in those happy days when I was working in Sleaford, I saw a majestic creature swoop out of nowhere and drop into the water, making barely a ripple. It was a duck. I've also seen pigeons, birds which are not noted for their intelligence, take a casual glance at a street lamp and dive straight into the sky, landing neatly on the top. The birds were using no radar and no high definition video feeds. They were doing it because after millions of years of evolution, their brains had developed the ability to navigate their bodies around the real world. Try building that kind of smarts into a drone and see how far you get.

I'm not one of those people who believe that true artificial intelligence is impossible or that research into the subject is a bad idea, because I'm convinced that we can learn a lot about the human condition if we can understand how our minds work. I also think that we'll get better at making machines interpret human speech, that machine translation will become more reliable and that self driving cars will eventually become a reality. The issue for me, is that none of that is going to happen quickly, because the problem with trying to make a machine intelligent is outrageously hard. It took billions of years for it to evolve in organic brains like our own, and it's sheer hubris to believe that we're going to be able to replicate it using electronics, at least in the foreseeable future.

The issue for the moment isn't artificial intelligence: it's artificial stupidity. We're becoming increasingly reliant on algorithms that have no conception of the problems they're trying to solve. Human beings are either being taken out of the loop or forced to read their responses from a pre-generated selection provided by a computer. And as anyone who's tried to use an automated helpline will tell you, the answers that you receive can be both infuriating and stupid.

Conclusion

AI is a useful tool, but it's not something we can entirely trust, because it's not actually intelligent. A pattern recognition system is only as good as the patterns used to train it. An expert system is only useful if its knowledge-base covers all the possibilities rather than just the obvious ones. A problem solving algorithm can be massively effective if the problem it's trying to solve is governed by clearly defined rules, which is true for games such as Chess or Go, but useless for situations where the parameters are more complicated, such as the ones that occur in the real world. Smarter machines aren't going to take over the world any time soon, but stupid machines might, and they could do a lot more damage than smart ones. We've all experienced situations where we've been using a computer and it's popped a suggestion that is so bizarre, it's positively dangerous. My personal favourite was back in the day, when I used a Mac and I tried to insert a disc that was created on a PC. A message popped up with "Do you want me to format this disc?". If I'd clicked "Yes", I'd have wiped the disc and deleted everything it contained, potentially losing hours of my work. It was stupid because the person who programmed it had never imagined that anyone could be using both a Mac and a PC at the same time. Modern AI systems are capable of the same type of stupidity, and that's why they're scary.