Reflecting on what worked yesterday....
Yesterday I gave a shuir at Siach on the following:
https://ashlag-cause-and-kook-affect.blogspot.com/2022/08/blog-post_15.html?m=1
The topic was AI - unsupervised learning in Halacha
People were interested - the room was full, not only the students but two teachers/Rabbis.
The Mishna and Gemara were perfect, they were learning an adjacent Sugya, but the Mishna had a simple twist.
So I started with an analysis of the Mishna in supervised learning mode, what is the case, what is the Halacha (action required), case law, simple. There are two cases in the Mishna, first half, disagreement between Tanna Kama and Rabbi Elazar Ben Azariah, second half, second case Rabbi Yehuda and Rabbi Meir. No connection. I asked them what titles would they give to each half, great.
Then I challenged that view, unsupervised, no case, really the Mishna is holistic. Look it says Rabbi Yehuda says, a continuation of the first half....
Tried to listen to the group, asked what is the value of a Ketuba, I could have done a better job listening, reframing their answer - oh it is a set of handcuffs? oh a marriage is maintained through threat of force?
Moved right along to the Gemara, try to read it supervised, I failed, but I did not have the time, nor inclination to let them read it and watch them fail. I think this worked anyway. The gemara ostensibly is teaching a Halacha, but it really isn't, much too complicated and no clear answer to what behavior should be accepted...
Back to the Mishna, what can we learn from the parallel holistic Mishna, oh Rabbi Meir provides a discreet formulated relationship, Rabbi Yehuda enables conversation, I tried to think of an example, cooking on Shabbat, is that a relationship with Halacha, I decide how to behave based on how I feel, in conversation with the Platta? Yet a Ketuba provides an opportunity for a discussion according to Rabbi Yehuda, ok now leverage that perspective into the first disagreement, the Tosefet Ketuba according to Tanna Kama is a conversation.
I explained the difference between the technique and tools of learning - sequence of text, Rabbi says vs. said Rabbi, the metrics of learning which are supervised but minimalistic, vs. the open/closed system and meaning imbued to the learning. I did *not* talk about levels and hierarchies. I also did *not* spend much time on energy/Information theory. -- I don't think they fully understood this, a question at the end seemed to indicate that this was missed.
Then I shifted gears back to AI, ChatGPT, LLM. I made the observation that GPT is unsupervised and learns to generate text based on next word, a strong assumption in sequence. A question. And then I added my understanding of the value of 'attention' in LLMs.
These two ideas were very powerful, they really liked them. The Ramban's opening remarks that the entire Torah is a sequence of letters of God's name, became Human History is a sequence of God's name. The sequence provides enough structure that learning is possible - without supervision. In addition, 'attention' provides a unique individual perspective -- my contribution to Human History.
Meaning can only be found in an open system. Computer Chess, Turing test - failure to see the open system, Go, all indicate that meaning is found in an open system.
They enjoyed it, but need to let it get absorbed...
notes:
1. Define mashiach as historical process, sequence, then semi supervised, unsupervised learning is a freedom providing method
