【AI+教育】一些记录@2024.11.05

AI4T开放参考书第一部分

背景：
AI4T项目是一个为期三年的实验性倡议，旨在探索和促进在法国、爱尔兰、意大利、卢森堡和斯洛文尼亚的中学教育中使用人工智能。该项目由法国教育国际组织协调，联合了五个参与国家的教育部门、公共机构、研究中心和大学。AI4T合作实施了创新的教师培训方法，成功地为来自302所欧洲学校的1,005名教师提供了专业培训。项目通过定量和定性评估策略，分析了培训路径的影响，并根据2022年的初步结果对资源进行了改进。AI4T的MOOC和开放教科书正在被项目外的国家翻译，以帮助更多教师自信地使用基于AI的资源。
（总结自https://www.france-education-international.fr/en/expertises/cooperation-education/projets/ai4t-artificial-intelligence-and-teachers）

在开放参考书第一部分，最吸引我的是它们对于教育科技、教育科技公司可能给教育带来的风险的清醒认知、对于教师需要“监管”AI的重要性的强调，并以此论述教师需要学习AI的缘由。以下摘抄部分内容：

前言：

For some, the future is bright and the coming technologies will help make education available to all; it might even help when there are not enough teachers. It will permit the teacher to spend more time on the ‘noble’ tasks while the machine will take over the ‘boring’ ones, such as grading, organising the classroom.
For others, these AI algorithms represent a huge danger, and the billions of dollars the industry is prepared to invest prove that education is now viewed as a market. But it is not a market.
Somewhere in the middle, between these rather different positions, are researchers, educators and policy makers who are aware of a number of things:

Training teachers的必要性之一：. Making teachers aware of why such training is good. It can’t be an imposed decision; it has to be shared.

Teachers need to have an understanding of how AI works. The goal is not to make each person a biologist or a physicist – the goal is to make us understand the principles and ideas.

A teacher needs reassurance that the working environment is safe for all.

对于行业可能裹挟教育的洞察：For obvious economic reasons, the industry will push products on the teachers, ostensibly to help them solve a sometimes-unimportant task. But if it is considered ‘cool’, and is pushed by the seller, it could end up being seen as important.……Unfortunately, this is not the case: a lot of software is still immature, there are a lot of ethical concerns and in most cases the different ministries and governments have not approved lists of software.

Part Ⅰ Why Learn About AI？

How can artificial intelligence impact learning and teaching in my classroom?
Can it help me do what I want to do with my students?
How can it change the dynamics and interactions I have with my students?
How do I even know when it is being used correctly or incorrectly?
And, what should I be aware of if I want to put it to good use?

Teacher should always be in the loop, overseeing what AI does. Effective AI solutions in the classroom are those that empower the teacher.

How to separate the truth from oversell?
What are some potential problems that could come with using this?

AI cannot differentiate between what is ‘true’ and what is a result of stereotyping and other prejudices.
Problems creep up in AI whenever an individual case differs from the majority case (whether this represents faithfully the majority in the real world or just the majority as represented by the internet). In classrooms, the teacher has to compensate for the system’s lapses and, where necessary, direct student’s attention to the alternative text.……It is for the teacher to judge what is acceptable and to act where a decision taken by AI is not appropriate.

数据/隐私的重要性：It is imperative that teachers know what kind of AI data is useful.
How do we secure our classrooms while enjoying the fruits of a greedy industry?

对教学法的强调：Any tool should be AI FOR TEACHERS based on sound pedagogical theories. Furthermore, to be most effective, it should be co-created in teams involving teachers, pedagogical experts and computer scientists.

《人工智能时代已经开始 | 盖茨笔记》

对盖茨而言，两次印象深刻的技术演示：图形用户界面和OpenAI的ChatGPT。

《ChatGPT 中，G、P、T 分别是什么意思？》

Generative 生成式
「Generative 这个大的种类里面有几个小分支，DALLE 的画图的用的是对抗网络方式 GAN （这个晚些可以分析），现在最火的Stable Diffusion， MidJourney 走向了另外一个分支，叫做 Difusion，而 ChatGPT 又是一个分支，就是转换器 Transformer。
Transformer Generative 的语言模型的核心，通俗的说就是“顺口溜”。」

Pre-Training预训练
「Pre-Training 模型不是为了特定的目标训练，而是预先训练一个通用的模型。如果我有特定的需求，我可以在这个基础上进行第二次训练，基于原来已经预训练的模型，进行微调（Fine- Tuning）。」

Transformer转换器
「变电器就是一种 transformer：220伏电进，12伏出。
语言的转换器就是把语言的序列作为输入，然后用一个叫做编码器 encoder 的东西变成数字的表现（比如 GPT 就用 1536 个浮点数（也叫 1536 维向量）表示任何的单词，或者句子，段落，篇章等），然后经过转化，变成一串新的序列，最后再用 decoder 把它输出。这个转换器，是这个自然语言处理的核心。」

GPT生成式预训练转化器
「一个预先训练好的，用生成的方式，把输入文字转化成输出文字的翻译。」