Q&A: Artificial Intelligence
Artificial Intelligence
Question
Hello,
If they develop artificial intelligence that gathers data from all the books of Jewish law, the responsa literature, and so on, and it can produce an answer to every question with 99 percent accuracy, and basically everyone turns to the AI—the AI will pull up an answer within a second, and there will be rabbis who only quickly review the answers to make sure they are correct—would that make all study for the purpose of halakhic ruling unnecessary (aside from the few individuals who would quickly review the answers for verification)?
After all, for example, that is how security organizations and so on work. And if it is good enough for some of the best security organizations in the world, where a mistake means life and death and even war, then apparently the technology is very, very sophisticated and can do such a thing.
So one could say there are unique cases and so on, but how many such unique cases are there already? And every time there is something unique, the AI will say it has no answer and the person asking will turn to a rabbi who will issue a halakhic ruling; but after the ruling, the answer will be absorbed by the AI, and so after such-and-such years there will barely be unique questions left either.
Answer
I do not know where you got the idea that this is how various security organizations operate. In any case, I do not think this makes study unnecessary, because there is no substitute for human judgment.
From your description, it sounds like you view such software as a collection of information. But that is not how this is done. We are talking about a neural network that is trained through feedback from the correct answers. But I do not know what the correct answers would be that you would give it, since there are disputes about a great many issues in Jewish law. So I do not see how one would train such a network to operate. You assume there is a given collection of halakhic rulings, but I do not think so. This is not necessarily connected to the question of halakhic monism (whether there is one correct answer to every question). It could be done on the basis of the rulings of one particular halakhic decisor, and then perhaps you would get a good reconstruction of that decisor, but that still would not make your own study unnecessary, because your rulings do not necessarily overlap with his.
Discussion on Answer
Producing targets means, for example, identifying—through visual analysis of satellite images (or photographs from Air Force sorties)—the characteristic appearance of missile hiding places and changes in terrain and construction. Or confirming from an attack drone that this really is the wanted person and this really is a missile battery, and so on. In my time in the army, intelligence researchers would independently analyze aerial photographs together with many other surrounding indications, and for that one can definitely make use of artificial intelligence.
Artificial intelligence also helps decide what information looks interesting within a hostile organization’s computer and is especially worth exfiltrating back home, and also directs the researchers’ attention to it. Given an enormous amount of information, it can also search for needles in haystacks whose identification seems like a stroke of genius (for example, a request to obtain material X, an attempt to smuggle restricted material Y, a warehouse in the port from which material Z was stolen in an area where operatives of a certain organization were seen or clues about it came up, and then the hypothesis that maybe the bad guys are trying to produce device A. These are things brilliant intelligence researchers with phenomenal memories used to do while getting through an astronomical amount of information).
But classical artificial intelligence will not be able to decide whether it is worth attacking while taking into account the total collateral and diplomatic damage and the fear of retaliation. That is a huge difference.
It seems that the questioner is dealing with textual analysis and understanding content, and indeed there is tremendous communal activity in that branch of artificial intelligence as well. For example, about two years ago IBM came out with a very, very impressive “Debater” tool that is capable of conducting an intelligent discussion on various topics by means of prior mining of huge amounts of information, understanding the subject being asked about, and formulating arguments (I read at the time that they also tried to fuse knowledge in order to create new arguments out of existing structures, and not merely identify existing arguments). Their monster could presumably also be applied to halakhic ruling within certain limits.
Today there is no artificial intelligence.
When there will be in the future, the study you mentioned will be unnecessary. Human beings will also be unnecessary.
Hello,
Thank you for the answer.
You can read about artificial intelligence, for example, here – https://www.israelhayom.co.il/news/defense/article/1515844/
Here is an excerpt from the article –
The campaign in Gaza was in fact the first artificial-intelligence war. The targets generated during the fighting were identified through a combination of human and machine. The Intelligence Directorate has superpower-level artificial intelligence capabilities, and many intelligence organizations, including the NSA, come to learn from Israel.
And if you search for various expressions such as, for example, artificial intelligence in armies, you can read about how this is being integrated in other places as well.
What percentage of halakhic questions are questions that require human judgment, and how many are completely automatic? For example, to answer the question whether one may light a fire on the Sabbath you do not need a rabbi, whereas to answer a question about permitting an agunah in a complicated situation you do need one. If most or the overwhelming majority are automatic answers, then presumably software can replace that, no?