小说与文学2021tedunderwood.com
科幻小说没能让我们准备好想象机器学习
Science fiction hasn’t prepared us to imagine machine learning
导读
这篇文章讨论为什么科幻文学没有充分帮助我们想象机器学习，以及技术题材为何难以进入叙事。评论补充了创作可行性、经典科幻资源和可证伪预测等角度，值得写作者思考如何把抽象、乏味的技术转化为有张力的故事。
可以直接用的写作要点
- 01
技术题材写作要找到可戏剧化的冲突和人物后果，否则读者很难对抽象概念投入情感。
- 02
把机器学习写成故事的底层条件而非唯一卖点，可能比直接写一部“关于机器学习的小说”更有效。
- 03
想象新技术时可以借用经典科幻的既有母题，但也要避免高估当前技术能力或把旧模型简单套用。
- 04
讨论技术趋势时尽量给出具体、可证伪的预测，这种训练能帮助写作者把空泛想象落成可感知的细节。
HN 评论精选
机器学习难以戏剧化，所以科幻常把它写成叙事的底层条件而非核心卖点。
书面科幻的一个问题是它至少要在商业上 vaguely 可行，这就逼着作者安排戏剧性情节和/或保证可读性。对大多数非技术读者来说，机器学习非常无聊。你怎么把它戏剧化？我出版过两本涉及这个领域的小说：2011年的《Rule 34》和即将出版的《Invisible Sun》。在这两本书里，机器学习都是叙事的从属要素：没有它就没有情节，但在我看来，很难像早年的“盒中大脑”式AI那样，把机器学习写成《2001太空漫游》《P-1的青春期》或《巨人》的核心构思——那些是1960、70年代的例子，当时这个母题还很新鲜，而不是已经被用到烂。
cstross查看英文原文
One problem in written science fiction is the requirement for it to be at least vaguely commercially viable, which necessitates dramatic plotting and/or readability. ML is, to most non-technical people, intensely boring. How do you dramatize it? I've published two novels that go there: 2011's "Rule 34", and -- coming out this September -- "Invisible Sun", and in both books ML is very much a subaltern aspect of the narrative. That is, neither book would have a plot without the presence of ML, but it's not really possible in my opinion to write a novel about ML in the same way that the earlier brain-in-a-box model of AI could form the key conceit of books like "2001: A Space Odyssey", "The Adolescence of P-1", or "Colossus" (examples from the 1960s and 1970s when it was still a fresh, new trope rather than being pounded into the dirt). So it probably shouldn't surprise anyone much that SF hasn't said much about the field so far, any more than it had much to say about personal computers before 1980 or about the internet before 1990 -- what examples there are were notable exceptions, recognized after the event.
对技术的怀疑应转化为可证伪的具体预测，而不是空泛的怀疑姿态。
我觉得这类话题下很多评论令人沮丧，它们更像怀疑姿态：“不过是统计！”或“当前AI没有理解，所以解决不了X”。我认为这种思路没有预测力。如果你因为深度学习“只是统计”而怀疑它，这种怀疑能让你预测2020年自动驾驶没解决、但蛋白质折叠解决了吗？最好的测试是：说出你认为未来两年肯定不会发生的最不起眼AI成就；或者列出你认为有50%概率在两年、五年内发生的成就。做出可证伪的预测。如果你不愿意，为什么？
emtel查看英文原文
I find a lot of the comments on this topic (and anytime AI comes up) very frustrating. Many of them strike me as little more than skeptical posturing: “it’s just statistics!” or “Current AI systems have no understanding and so can’t solve problem X” I don’t think this sort of thinking has any predictive power. If you are skeptical of deep learning because it is “just statistics”, did that skepticism allow you to predict that self driving cars would not be solved in 2020, but protein folding would? I think the best test for skeptics is the one proposed by Eliezer Yudkowsky: what is this least impressive AI achievement you are sure won’t happen in the next two years. If you really want to impress me, give me a list of N achievements that you think have a 50% chance of happening in 2 years. Or 5 years. Just make some falsifiable prediction. If you aren’t willing to do that, or something like it, why not?
经典科幻已经提供了思考资源，作者可能低估了科幻传统。
一些科幻经典对此有很多话可说，而且远超当前和近期现实可能达到的程度。作者要么大大高估了当前AI的能力，要么没读过阿西莫夫之类的作品。
dgb23查看英文原文
Some of the science fiction classics have a lot to say about this and go well beyond what is currently possible and realistic in the near future. The author either vastly overestimates the capabilities of current AI or didn’t read Asimov and the like.
同类还值得读
乔治·奥威尔1940年对《我的奋斗》的书评
George Orwell’s 1940 Review of Mein Kampf
这是奥威尔 1940 年对《我的奋斗》的书评，他没有简单谴责，而是认真分析希特勒的宣传手法为何奏效、又暴露了哪些思维特征。
小说与文学2022 · 499 分 · 545 评4 条写作要点编辑精选我们都是极客：尼尔·斯蒂芬森的文学作品
We are all nerds: The literary works of Neal Stephenson
这是一篇讨论 Neal Stephenson 科幻小说的文章，评论区读者围绕《Anathem》《钻石年代》《Snow Crash》等作品，聊世界构建、结尾处理和写作风格的变化。
小说与文学2022 · 307 分 · 220 评4 条写作要点编辑精选又一部我不会再去写的小说
Yet another novel I will no longer write
科幻作家 Charles Stross 讲述自己为何放弃一部已经构思好的小说：COVID-19 疫情让他的病毒灾难设定显得多余，现实给所有人上了一堂生动的疫情应对课。
小说与文学2020 · 304 分 · 169 评4 条写作要点编辑精选用图画写作：理查德·斯卡里与儿童文学的艺术
Writing in Pictures: Richard Scarry and the art of children's literature
这是《耶鲁评论》上漫画家 Chris Ware 写的文章，分析童书作家 Richard Scarry 如何用图画而不是文字来讲故事。
小说与文学2024 · 297 分 · 131 评4 条写作要点编辑精选如何写一部小说
How to Write a Novel
这篇来自 The Millions 的文章讨论写小说的两种路径——“计划派”（planners）与“即兴派”（pantsers，即“跟着感觉走、边写边编”）。
小说与文学2016 · 238 分 · 112 评4 条写作要点编辑精选