澳门新濠新天地3559_www.3559.com新濠新天地网址

您的位置:新天地3559 > 机械设备 > 要在通用人工智能系统里实现对于意图的工程学

要在通用人工智能系统里实现对于意图的工程学

2019-11-23 07:43

how to design an artificial general intelligence system bearing intentions: an interdisciplinary inquiry based on anscombe's philosophy of intention

AI     artificial intelligence    人工智能

Ruslan Salakhutdinov, who leads Apple’s AI efforts, says emerging techniques could make the most popular approach in the field far more powerful.

作者简介:徐英瑾,哲学博士,复旦大学哲学学院教授,博士生导师,教育部青年长江学者。上海 200433

GA     genetic algorithms     遗传算法

by Will Knight  March 29, 2017

原发信息:《武汉大学学报:哲学社会科学版》第20186期

ANN    artificial neural network   人工神经网络

Apple’s director of artificial intelligence, Ruslan Salakhutdinov, believes that the deep neural networks that have produced spectacular results in recent years could be supercharged in coming years by the addition of memory, attention, and general knowledge.

内容提要:具有自主意图、只依赖小数据运作的通用人工智能系统的出现,并不会像有些人所预估的那样导致“机器奴役人类”的局面出现,因为此类技术对于小数据的容忍可以大大增加此类技术的潜在用户的数量,并使得体现不同用户价值观的通用人工智能系统能够大量出现。这样一来,具有不同意图的通用人工智能系统彼此之间的对冲效应,最终会使得任何一种具有特定意图的通用人工智能系统都无法占据主宰地位。相反,由于作为专用人工智能技术代表的深度学习技术的运用在原则上就需要大量数据的喂入,其对于民众隐私权的侵犯就成为一种难以被全面遏制的常态,因此,此类技术的发展在原则上就会加强一部分技术权贵对于大多数民众的统治地位。不过,要在通用人工智能系统里实现对于意图的工程学建模,就需要我们在哲学层面上首先厘清关于意图的种种哲学迷思。在这个问题上,美国女哲学家安斯康的意图理论是一个比较好的讨论起点。具体而言,安斯康关于“意图是在欲望驱使下做某事的理由”的观点,是可以在通用人工智能的语境中被实现的,但是她关于信念与意图之二元对立的观点,却在不少地方有失偏颇。而“非公理化推理系统”,则将为吸纳安斯康意图论的合理部分提供相应的工程学手段。

ML     machine learning     机器学习

Speaking at MIT Technology Review’s EmTech Digital conference in San Francisco on Tuesday, Salakhutdinov said these attributes could help solve some of the outstanding problems in artificial intelligence.

whether artificial intelligence will enslave human beings is very relevant to the meaning of "ai" itself.if "ai" means "artificial general intelligence" and the agi system in question can work well by tolerating a small size of inputs,then the potential number of the users of this technology will be significantly increased.and if these agi systems can have their habits of producing intentions in accordance with different users' values,then varieties of agi systems imbued with different values will make it tough for any single type of machine to dominate the society.hence,machines cannot enslave human-beings if machines are agi systems in this sense.in contrast,if "ai" only means deep learning systems which cannot function well without exploiting large quantity of data,then the issue on how to protect the human privacy will always be salient,and in this sense,agi systems with the capacities of having their own intentions is ethically superior to their deep learning counterparts.as to how to produce intentions in an agi system,g.e.m.anscombe's philosophy of intention may offer many inspirations,although some of her claims on the nature of intentions may be controversial.pei wang's non-axiomatic reasoning system will offer a technical realization of the plausible part of anscombe's theory of intention.

DL      deep learning   深度学习

Salakhutdinov, who retains a post as an associate professor at Carnegie Mellon University in Pittsburgh, pointed in his talk to limitations with deep-learning-driven machine vision and natural-language understanding.

关键词:通用人工智能/非公理化推理系统/纳思系统/深度学习/公众隐私/安斯康/artificial general intelligence/non-axiomatic reasoning system/nars/deep learning/public privacy/anscombe

DT      decision tree   决策树

Deep learning—a technique that involves using vast numbers of roughly simulated neurons arranged in many interconnected layers—has produced dramatic progress in machine perception over recent years, but there are many ways in which these networks are limited.

标题注释:国家社会科学基金重大项目,国家社会科学基金一般项目。

NBN     native bayesian network    朴素贝叶斯网络

Salakhutdinov showed, for example, how image captioning systems based on the technology can label images incorrectly because they tend to focus on everything in the image. He then pointed to a solution in the form of so-called “attention mechanisms,” a tweak to deep learning that has been developed in the last few years. The approach can remedy these errors by having a system focus on specific parts of an image when applying different words in a caption. The same approach can help improve natural-language understanding, too, by enabling a machine to focus on the relevant part of a sentence in order to infer its meaning.

一、从“人工智能是否会奴役人类”谈起

DFS     depth-first search  深度优先搜索

A technique called memory networks, developed by researchers at Facebook, can improve how machines talk with people. As the name suggests, the approach adds a component of long-term memory to neural networks so that they remember the history of a chat.

随着近几年以来人工智能技术在工程学层面上的不断进步,关于“人工智能是否会在未来统治人类”的担忧,也日渐被人提起。但在笔者看来,这个问题本身已经包含了诸多语言混乱。如果不预先对这些混乱加以厘清,我们将很难对这一问题作出严肃的应答。具体而言,该问题所涉及的第一重语言歧义即:这里所说的“人工智能”究竟是指专用人工智能(即只能用于特定工作目的的人工智能系统),还是通用人工智能(即能够像人类那样灵活从事各种工作的人工智能系统)?有人或许会说,抓住这一点歧义不放乃是小题大做,因为所谓通用人工智能技术,无非就是既有的专用人工智能技术的集成。但持此论者却没有意识到如下三个问题:

BFS     breadth-first search   宽度优先搜索

Memory networks have been shown to improve another kind of AI as well, known as reinforcement learning. For example, two researchers at CMU recently showed how this could create a smarter game-playing algorithm. Researchers at DeepMind, an AI-focused subsidiary of Alphabet, have also demonstrated ways for deep-learning systems to build and access a form of memory.

就既有专业人工智能技术中发展最快的深度学习系统而言,此类系统的运作其实是需要大量的数据输入为其前提的。因此,深度学习系统并不具备根据少量数据进行有效推理的能力——换言之,它们缺乏“举一反三”的智能,尽管这种智能乃是任何一种理想的通用人工智能系统所不可或缺的。不得不提到的是,在“迁移学习”这一名目下,目前不少深度学习研究者都在研究如何将在一个深度学习网络中已经获得的网络权重分布“迁移”到一个新的网络中去。这姑且可以被视为某种最初步的“举一反三”。然而,这种意义上的迁移学习必须预设深度学习网络所从事的新任务与旧任务之间有足够的相似性,而无法模拟人类在非常不同的领域(如“孙子兵法”与商业活动)之间建立起类比推理关系的能力。

Reinforcement learning is rapidly emerging as a valuable way to solve hard-to-program problems in robotics and automated driving. It was one of MIT Technology Review’s 10 Breakthrough Technologies of 2017.

现有的深度学习架构都是以特定任务为导向的,而这些任务导向所导致的系统功能区分,既不与人类大脑的自然分区相符合(譬如,我们人类的大脑显然没有一个分区是专门用于下围棋的,而专门用于下围棋的“阿尔法狗”系统的内部结构则是为下围棋量身定做的),也缺乏彼此转换与沟通的一般机制。因此,深度学习系统自身架构若非经历革命性的改造,其自身是缺乏进阶为通用人工智能系统的潜力的。

Another exciting area of future research, Salakhutdinov said, would be finding ways to combine hand-built sources of knowledge with deep learning. He pointed to general-knowledge databases like Freebase and word-meaning repositories like WordNet.

目前真正从事通用人工智能研究的学术队伍,在全世界不过几百人,这与专业人工智能研究的庞大队伍相比,可谓九牛一毛①。

Just as humans rely heavily on general knowledge when parsing language or interpreting a visual scene, this could help make AI systems smarter, Salakhutdinov said. “How can we incorporate all that prior knowledge into deep learning?” he said during his talk. “That’s a big challenge.”

有鉴于特定技术流派的发展速度往往与从事该技术流派研究的人数成正比关系,所以,除非有证据证明通用人工智能的研究队伍会立即得到迅速扩充,否则我们就很难相信:通用人工智能研究在不久的将来就会取代专用人工智能研究,迅速成为人工智能研究的主流。而这一点又从另一个侧面印证了专用人工智能与通用人工智能之间的差异性。

Salakhutdinov spoke during a session that brought together researchers from several different schools of AI. A common theme among the speakers was the need for different approaches in order to take AI to the next level.

During the session Pedro Domingos, a professor at the University of Washington who studies different machine-learning approaches, said there is also a need to keep searching for completely new approaches to AI. “There’s a school of thought in machine learning that we don’t need fancy new algorithms, we just need more data,” he said. “I think there are really deep, fundamental ideas that need to be discovered before we can really solve AI.”

本文由新天地3559发布于机械设备,转载请注明出处:要在通用人工智能系统里实现对于意图的工程学

关键词: