IIT_CS480_Assignments_Week6_Participation_cs480课程作业-CSDN博客

本文链接：https://blog.csdn.net/dearsq/article/details/109425061

IIT CS 480 - Introduction to Artificial Intelligence
Assignments_Week6_Participation
github: https://github.com/IIT-MAI-2020Fall

文章目录

作业描述

This is a two-week discussion.

The next topic is Logic: chapters 7, 8, and 9. A seminal paper on AI and logic was the paper “Programs with Common Sense” written by John McCarthy in 1959. John McCarthy was one of the founders of the field AI.
下一个主题是逻辑:第7、8和9章。关于人工智能和逻辑的一篇开创性论文是1959年由约翰·麦卡锡撰写的论文《有常识的程序》。约翰·麦卡锡是人工智能领域的创始人之一。

Required reading
必须阅读的内容
Wiki page: https://en.wikipedia.org/wiki/John_McCarthy_(computer_scientist)
John McCarthy’s legacy: https://www.sciencedirect.com/science/article/pii/S0004370210001827
(Please read the full pdf; not just the abstract. Let me know if you cannot access the pdf)

维基百科
John McCarthy’s legacy ：/John McCarthy’s legacy.pdf
请阅读全文pdf;不仅仅是抽象的。如果无法访问pdf，请告诉我。

Optional reading
可选的阅读内容
Original article: http://jmc.stanford.edu/articles/mcc59/mcc59.pdf
McCarthy’s contributions and obituaries http://jmc.stanford.edu/contributions/index.html
McCarthy’s opinions on what AI is. Written in 2007. Highly recommended. http://jmc.stanford.edu/articles/whatisai/whatisai.pdf

Discussion
讨论

分享对麦卡锡人工智能逻辑的看法

Please do not start the discussion until we have a chance to talk about in class. I’ve posted the discussion so that you can start reading the relevant materials.
请不要开始讨论，直到我们有机会在课堂上讨论。我已经把讨论张贴出来了，这样你就可以开始阅读相关的资料了。
Share your thoughts about McCarthy’s logical approach to AI (discussed in the required reading #2: https://www.sciencedirect.com/science/article/pii/S0004370210001827).
请分享你对麦卡锡的人工智能逻辑方法的看法

你觉得人工智能的正确方向是什么？

One discussion among AI researchers have been what the right approach to AI is. Some people argued logic was the right approach while some people argued probability theory was the right approach, and yet others argued that circuits (e.g. neural networks) were sufficient for AI and that we didn’t need the logical approaches. Some people combined logic and probability theory. Recent advances in deep learning rekindled this discussion of whether neural networks were the solution to AI or whether we still needed logic for common sense reasoning. What do you think? Do you think logic is just plain old-fashioned AI and we don’t need it anymore? Do you think neural networks will be sufficient for machines to achieve common sense reasoning and planning or do you think neural networks are only tools for pattern recognition and the machines still need logic and knowledge bases for common sense reasoning and planning? Research this topic. List your references. Share your ideas.
人工智能研究人员之间的一个讨论是，人工智能的正确方法是什么。有些人认为逻辑是正确的方法，而有些人认为概率论是正确的方法，而另一些人则认为回路(例如神经网络)对人工智能来说已经足够了，我们不需要逻辑方法。有些人把逻辑和概率论结合起来。深度学习方面的最新进展重新引发了这样一场讨论:神经网络是否是人工智能的解决方案，或者我们是否仍需要逻辑来进行常识推理。你觉得怎么样?你认为逻辑只是简单的老式人工智能，我们不再需要它了吗?你认为神经网络足以让机器实现常识推理和规划吗?还是你认为神经网络只是模式识别的工具，机器还需要逻辑和知识基础来进行常识推理和规划?研究这个主题。列出你的引用。分享你的想法。

阅读材料重点梳理

wiki百科-John_McCarthy

约翰·麦卡锡（1927年9月4日至2011年10月24日）是美国计算机科学家和认知科学家。麦卡锡是人工智能学科的奠基人之一。
开发了Lisp的编程语言的家庭，显著影响了设计ALGOL编程语言。

回答

Share your thoughts about McCarthy’s logical approach to AI (discussed in the required reading #2) From the article, I learned that McCarthy made outstanding achievements in three aspects of Reasoning about Action, Reasoning about knowledge and Nonmonotonic Reasoning：

McCarthy 在很多方面具有前瞻性，并且规划了人工智能的发展方向。虽然有的具有缺陷，但是瑕不掩瑜。
One discussion among AI researchers have been what the right approach to AI is. Some people argued logic was the right approach while some people argued probability theory was the right approach, and yet others argued that circuits (e.g. neural networks) were sufficient for AI and that we didn’t need the logical approaches. Some people combined logic and probability theory. Recent advances in deep learning rekindled this discussion of whether neural networks were the solution to AI or whether we still needed logic for common sense reasoning. What do you think? Do you think logic is just plain old-fashioned AI and we don’t need it anymore? Do you think neural networks will be sufficient for machines to achieve common sense reasoning and planning or do you think neural networks are only tools for pattern recognition and the machines still need logic and knowledge bases for common sense reasoning and planning? Research this topic. List your references. Share your ideas.

人工智能研究人员之间的一个讨论是，人工智能的正确方法是什么。有些人认为逻辑是正确的方法，而有些人认为概率论是正确的方法，而另一些人则认为回路(例如神经网络)对人工智能来说已经足够了，我们不需要逻辑方法。有些人把逻辑和概率论结合起来。深度学习方面的最新进展重新引发了这样一场讨论:
神经网络是否是人工智能的解决方案，或者我们是否仍需要逻辑来进行常识推理。你觉得怎么样?
你认为逻辑只是简单的老式人工智能，我们不再需要它了吗?
你认为神经网络足以让机器实现常识推理和规划吗?
还是你认为神经网络只是模式识别的工具，机器还需要逻辑和知识基础来进行常识推理和规划?
研究这个主题。列出你的引用。分享你的想法。
我认为，我们需要使用逻辑进行常识推理，并且结合神经网络。才能更好地解决问题。
第一代知识驱动的 AI，利用知识、算法、算力 3个要素构造AI。
第二代数据驱动的 AI，利用数据、算法、算力 3个要素构造AI。
接下来 AI 的发展目标是安全、可信、可靠、可扩展。所以其思路是，将第一代的知识驱动和第二代的数据驱动相结合，通过同时利用知识、数据、算法、算力四个要素来构造 AI。目前的解决方案是基于双空间模型和单一空间模型构造的三空间融合模型。
I think we need to use logic to do common sense reasoning, combined with neural networks
The first generation of knowledge-driven AI uses knowledge, algorithm and computing power to construct AI.
The second generation of data-driven AI uses data, algorithm and computing power to construct AI.
The next development goal of AI is to be safe, reliable, reliable and extensible. Therefore, the idea is to combine the knowledge driven in the first generation with the data driven in the second generation, and construct AI by using the four elements of knowledge, data, algorithm and computing power simultaneously.The present solution is a Triple-Space integration model based on Dual-space model and a Single-space model.
双空间模型：一种类脑模型，符号空间模拟大脑的认知行为，亚符号（向量）空间模拟大脑的感知行为。这个目标的实现依赖于1.知识和推理 2. 感知 3.强化学习
Dual space model: a kind of brain model, the symbol space simulates the cognitive behavior of the brain, and the subsymbol (vector) space simulates the perceptual behavior of the brain.This goal depends on 1. Knowledge and reasoning 2. Perception 3. Reinforcement learning
单一空间模型：以深度学习为基础，将所有的处理都放在了亚符号（向量）空间。
Single-space model: Based on deep learning, all the processing is placed in the subsymbol (vector) space.
三空间融合模型：
Triple-Space integration model.

参考文献：

Zhang B, Zhu J, Su H. Toward the third generation of artificial intelligence (in Chinese). Sci Sin Inform, 2020
Zhang Y, Dai W Y, Pan S J. Transfer Learning. Cambridge: Cambridge Press, 2020