Perhaps I am being unkind but over the last several years I have seen “artificial intelligence” waved in front of a myriad of questions like a magic talisman. Given my interest in languages, naturally my interest was piqued when I ran across this article at Discovery Magazine on using artificial intelligence to decipher writing that is presently undecipherable. The punch line is here:
Taking on ancient languages with AI does pose some unique problems, though. Machine learning algorithms are usually trained on massive datasets that they mine in order to learn through associations. Most ancient scripts have only a limited number of samples, making it difficult to feed an algorithm enough data for it to learn.
The process of training an algorithm also involves comparing its answers to known values. When a language is entirely undeciphered, however, this is impossible. You can’t tell an algorithm “Yes, that is a bike,†or “No, that word does not mean ‘stop’†if you don’t know what any of it means.
or, said another way, it isn’t particularly helpful. The problem with proto-Elamite, for example, is that it may be what’s called a “language isolate”, a language like Basque with no known related languages. The orthography is likely to be imperfect which makes it quite difficult to identify cognates. The number of samples is rather limited. Much the same is true of the writings of the Indus civilization. We’re not even 100% certain that they’re writings.
By comparison deciphering the Mayan glyphs, something I consider one of the great achievements in translation of the 20th century, was relatively easy. The hardest part was recognizing that they were writings.
However, there’s always hope.
Don’t you go losing your faith now, Dave. They’re all acoming – the flying car, fusion power, AI,… – and they’re all going to arrive on your birthday!
I have been skeptical about AI, but this week I saw some hope for it. We had the two lead engineers from a major international come talk with a group of my docs, which was kind of shocking. I have always assumed the real focus of the engineers of medical products is to make mediocre stuff that costs a lot, so why do you ever need to talk with the physicians who are going to use the stuff? They really seemed to be interested in our input, and their product they are trialing with us looks like it might have some very useful stuff making significant use of AI, though the engineers were at least honest about the limits of what it could do. Who knows? Maybe we arent so many decades away from AI being as useful as people think it can be.
Steve
I think that medical software or equipment incorporating expert systems have potential. I suspect that medical equipment or software making use of neural nets will have difficulty getting approved by the FDA.
In other words, the instant language translators used in Star Trek and other science fiction shows and media will forever remain fantasy.
However, there is the Babel Fish of Hitchhiker’s Guide to the Galaxy.
AI is not AI, as in it never has been nor ever will be actual intelligence, just hideously complicated algorithms that mimic intelligence. Some predestination types however argue that human intelligence is simply algorithms too, just a number of orders of magnitude greater than currently exists. I dunno. Too deep and complicated for me. I’d rather pretend that I have some free will in life.
It’s actually more complicated than that. “Artificial intelligence” is a grab bag of not particularly related techniques for accomplishing a variety of tasks vaguely related to natural intelligence. Some, as you say, are hideously complicated algorithms but some are actually pretty simple and vaguely mimic extremely primitive natural intelligence, e.g. neural nets.