当博尔赫斯遇见人工智能的先驱之一
When Jorge Luis Borges met one of the founders of AI

原始链接: https://resobscura.substack.com/p/when-jorge-luis-borges-met-one-of

这位历史学家悲叹当前对学术界的攻击,并以此与1970年Jorge Luis Borges和Herbert A. Simon之间一次引人入胜的对话作类比。这次会面由Simon对Borges的钦佩发起,探讨了文学、哲学、人工智能和机器学习的交叉点。 Borges和Simon讨论了自由意志、决定论以及用计算机模拟人类行为的可能性。Simon主张相容论,认为人类行为如同计算机程序一样,但个体性依然存在。Borges追问了全知存在对行为的可预测性,导致Simon同意基于个体过去,行为是不可避免的。 作者受到这场跨学科对话的启发,设想未来人文科学和STEM领域合作,利用原始资料和先进的语言模型来模拟历史人物和事件。作者个人反思浮现,提及其母亲的日记,暗示未来一个项目:创造其母亲思想和梦境的幽灵AI痕迹,这与Borges-Simon的对话产生了共鸣。作者希望在当今支离破碎的知识景观中,促进更多这样的对话。

一篇关于博尔赫斯与人工智能创始人会面的文章引发了Hacker News的讨论。评论者们辩论着通过汇编博尔赫斯的大量阅读和感官体验数据来创造一个类似博尔赫斯风格的语言模型的可能性。一位评论者认为这个想法类似于斯坦尼斯瓦夫·莱姆的《ไซเบอร์》中一个故事的情节,故事中一个机器人试图模拟世界的进化来创造一个人工智能诗人。其他人则提到了博尔赫斯自己的故事《皮埃尔·梅纳尔,堂吉诃德的作者》作为相关的概念。 一位评论者强烈推荐博尔赫斯的短篇故事,特别是《Tlön, Uqbar, Orbis Tertius》,并认为大型语言模型缺乏真正再现像博尔赫斯这样的作者所需的先验综合知识。他们对比了赫伯特·西蒙的“感觉束”观点和康德的知识论方法。另一位评论者发现博尔赫斯-西蒙的对话很有趣,但鉴于大型语言模型目前的性能,缺乏更深入的见解。
相关文章

原文

One reason I became a historian is the joy of encountering moments in the past that are foreign, yet also oddly familiar. These moments seem to ripple outward, lapping up against the present in unexpected ways.

Lately, I have been deeply dispirited by the current attack on academia here in the US — the deportation of graduate students and researchers, the dismantling of a huge range of important projects via funding cuts to NIH and elsewhere. And I’ve been reflecting on one of these moments of contact with the past. It happened in Buenos Aires in 1970. What I love about this exchange is that it takes place across the supposed “two cultures” of science and the humanities, jumping in a remarkably freewheeling way between literature, philosophy, and the fields of AI and machine learning, which were at the time barely even born.

The participants were Jorge Luis Borges, the great Argentine writer, and Herbert A. Simon, an economist, psychologist, and pioneer of artificial intelligence. Their conversation, part of which I’ve reproduced below, models something we badly need today: not just the simple joy of exploring ideas for their own sake, but also the ability to avoid seeing your work and life as something defined by what it excludes.

I rarely see historians and computer scientists (or people who write for literary magazines and people who work in AI labs) having this sort of conversation today.

I wish it happened more. That’s partly why I’m writing this newsletter.

Share

Their meeting took place in Borges’s office at the Argentine National Library. Simon had requested the meeting because of his admiration for Borges’s writing—his first and only “audience with a celebrity,” as he put it later . He was visiting Buenos Aires to lecture on systems theory, but what he really wanted was to speak with the author of “The Library of Babel.”

“I have been seized by an unaccountable wish to meet you,” Simon wrote in a letter to Borges. “I am a social scientist who tries to understand human behavior by building mathematical models (or, more recently, computer simulations).”

Once the conversation began, Borges quickly guided Simon away from casual pleasantries about translation and into deeper philosophical waters. Borges, in his gently ironic manner, asked:

BORGES: “Now I would like you to tell me a bit about the so-called behaviorism. What is its underlying principle? Is it free will or predestination?”

It is, and was, a hugely open-ended question. But Simon was at this time immersed in the study of early computer simulations, and he took this to be a question about whether the human mind is fundamentally comparable to a computer.

This was his response:

SIMON: Well, I always end up talking about computers. I’m in love with computers. We could put it this way. Faced with a certain problem, a computer will also behave in a certain way. And we may wonder whether it acted out of free will. We can say it employed free will in the sense that had it been programmed differently, it would have chosen to do otherwise, it would have behaved in a different way.

BORGES: What do you mean by behaved? Because we are talking about a purely mechanical process here. I mean, in the case of computers, of course.

SIMON: Yes, it’s a mechanical process. But I believe, like many others in my professional field, that human beings also display a mechanical type of behavior, much like computers.

BORGES: Do you mean to say that we act by force of habit?

SIMON: Rather accordingly to programs that we have stored in our brain. And we have free will in the sense that our resulting behavior will depend on who we are and the situation we are in. People respond differently when confronting the same situation.

BORGES: So, in your opinion, when faced with a dilemma, say, a situation in which there is a choice to be made between two possible behaviors, we can choose one of them?

SIMON: Your mental programming does the choosing. Yes, you choose.

It seems to me that Simon is here arguing for what philosophers call “compatibilism” — the idea that determinism can coexist with meaningful human choice and responsibility.

When Borges pressed him on whether an all-knowing being could predict our every action, though, Simon didn't shy away from the implications:

BORGES: Would this imply that if any all-powerful being, any god, knew everything about my past, my childhood, even about the time before I was born, my ancestors … would this imply that he would be able to predict my behavior in any one situation?

SIMON: According to my scientific beliefs, I would say so. With such knowledge, we can predict an individual’s behavior.

BORGES: So, what I’m saying right now is …

SIMON: … it’s a result of your past …

BORGES: … it’s inevitable.

SIMON: It is inevitable, yes. However, you still retain your identity, your individuality. You embody your own past.

BORGES: I understand. Well, I like to think I do. Now, does this account for all of our actions? That is, if my right hand is resting on my left hand, is it because it has to be this way? I believe people do quite a lot of things without any thinking.

SIMON: That’s the doing of our subconscious mind. You’re right, yes, otherwise, we would hardly be able to tie our shoelaces. Most things happen in this way. But that’s because we are heavily programmed.

BORGES: Would you say then things are to be also inevitable in this sense?

SIMON: They might be different, but always depending on programming. Any determinant could affect your programming and lead you to act differently. And if we introduce chance into the picture, scientists will always ultimately rule it out. At some point they may have to admit their inability to explain a particular phenomenon, but they will keep on working on the assumption that actions are determined by certain causes. And therefore, when we study a person who is in the process of solving a problem, we start from the assumption that every little thing has a cause. We are not always able to identify those causes.

BORGES: Well, of course. In order to study a person’s behavior, professionals have to go back to that individual’s history. Even to the historic past, to the origin of humanity, the cosmos.

SIMON: No, it’s not like that. Because the past influences a person’s present behavior to the extent that this past is already in the person. So, we can always find a starting point.

What I love about this is that here we have two significant thinkers, in two very different fields, who are both reflecting on not just on the meaning and importance of history (“it’s a result of your past”) but also about how and whether they have, themselves, been able to act within history.

If you compiled an enormous dataset of everything Borges read, and combined it with an exquisitely sensitive record of every sensory experience he ever had, could you create a Borges LLM? It’s a question that I think a lot of my peers in the humanities would not take much interest in, today (the two cultures, again). But I love that Borges himself was thinking about it.

Lately I have been spending a good amount of time thinking about whether, and in what ways, there can be value in seeking to simulate historical figures (or even entire historical moments — like the collective medical opinion of 1820s London, say) by combining a large amount of historical primary sources with an advanced language model.

I think this will absolutely be a powerful tool. I’m just not sure when and how. What I do know is that the people who figure that out will be a team that combines humanities and STEM. It is encouraging to see Simon and Borges embodying that kind of thinking way back in 1970.

I have been writing Res Obscura for 13 years now, and it has always been unpaywalled and free of any advertisements. The generosity of paid subscribers makes this possible.

Someday, when the feeling is a little less raw, I’ll write about the hundred or so journals that my mother left behind when she died. It was during Covid, and my wife was in her second trimester of pregnancy with our older daughter, Yara. My mom had been diagnosed with an aggressive form of cancer and was in a lot of pain. She was incredibly brave throughout, and, as a lifelong diarist and writer, she kept writing in her journal up to the end.

This is one of the last passages in her last journal. It’s about her hoping to hold on long enough to see Yara (who she ended up missing, alas):

You embody your own past.

I haven’t had the heart to actually read my mother’s journals yet. I only have this photo because she showed it to me and wanted me to share with Yara when she got older. Someday, probably far in the future, I will not only read them, but experiment with using them as data for an AI system which, to some extent, might embody some sort of spectral trace of my mother and her past.

She wanted her journals to be read and understood. She wanted to be able to talk to Yara. Someday, “she” will. Or, at least, a simulacrum composed of the collection of thoughts and dreams and impressions that she chose to record will.

That day is not even close today. But it will happen. And when it does, I’ll think about Borges and Simon in Buenos Aires in 1970.

The full dialogue is reproduced here, in an article in The Journal of the History of Economic Thought by Ricardo F. Crespo.

Share

• “They discovered that the Jahai, hunter-gatherers living at the border of Malaysia and Thailand, have a rich vocabulary of abstract smell words. One Jahai term, itpit, refers to the ‘intense smell of durian, perfume, soap, Aquillaria wood, and bearcat’… Jahai and English speakers were asked to identify and name twelve smells, including cinnamon, turpentine, gasoline, and onion. English speakers, despite their greater familiarity with the odors, faltered. They mostly gave rambling source-based answers and showed almost no agreement among themselves. One English speaker presented with cinnamon responded, ‘I don’t know how to say that, sweet, yeah; I have tasted that gum like Big Red or something tastes like, what do I want to say? I can’t get the word. Jesus it’s that gum smell like something like Big Red. Can I say that? Ok. Big Red. Big Red gum.’ But Jahai speakers named smells with relative ease.” (From this great New Yorker article by UC Davis anthropologist Manvir Singh on how language shapes thought)

• As a longtime fan and sometime contributor to Lapham’s Quarterly, I was happy to see that it has been revived: more here.

• “In comments to The Washington Post, the Brazilian president went further, delving into the minutiae of early 20th-century aerospace engineering and mourning what he described as the wrongful denial of Brazilian valor. ‘Everyone knows that Santos Dumont was the first to make something heavier than air fly, in an autonomous way, without any assistance,’ he vented. ‘But the Americans have the movie industry and were able to promote the Wright brothers.’” (The Washington Post)

Leave a comment

Share

联系我们 contact @ memedata.com