lec 1
What is AI? (1) Machine Learning
（1）Detection of systematic patterns between input and output
（2）General task: Predict output given specific features of the input
(minimizing the prediction errors)
（3）Very similar to “regular” statistical modeling
- Input features: independent variables
- Output class: dependent variable
- (in fact, ‘neural networks’ can be seen as a form of logistic regression
models)
- Key difference to statistical modeling:
We care about predicting something, not about understanding a (causal)
process
Models are highly complex (and multicollinear) and generally seen as
‘black box’
Deep learning
- Fancy term for machine learning with very large models
- Based on:
Very large neural networks (with a specific structure)
Trained on enormous amounts of data, e.g. “all of the internet”
,Using massive computing power, especially of GPUs
Key innovations:
Feature layers find patterns in raw input
Networks can be (pre-)trained based on unannotated data
Patterns from (pre-)training are transferred to actual task, and fine-tuned
on annotated data
Conclusions
AI is a group of techniques clustered around machine learning
AI is transforming society at an increasing pace
AI associated with various problems connected to power
- Bias
- Inequality
Regulating/fixing AI requires a deep understanding of …
- What AI is?
- How it affects (various aspects of) society?
- How solutions/regulation interacts with technology, users, owners,
society
Interdisciplinary research and solutions are key
lec2
Why is “Deep learning” revolutionary
,- Key insights: transfer learning based on unannotated training
material
没有标记的材料
- ML was always limited by lack of (expensive) annotated training data;
steps to find ‘features’ can work on raw data
-- Finding faces, detecting similar words, connotations of words
- Patterns / “Knowledge” extracted from these data sets can be
transferred to new tasks, so not every task has to start from scratch 从旧
知识到引用
- Next week: more details on embeddings 词向量 and move from CNNs
to transformers (GPT)
What makes news special?
What makes theguardian.com different from amazon.com?
Normatively, news supply has direct societal consequences
- Informing citizens (Democratic citizenship)
- Exposing problems (Watchdog role, press as fourth estate) 揭露问题
- Enabling debate (Public sphere) 促进公共讨论
→ It matters what content is recommended
Technically, news has very short shelf life “保质期”非常短
Economically, news is hard to monetize 变现
News gathering vs distribution
, (Good) news gathering is difficult & expensive
-“Professional Journalism is Expensive” (Nielsen 2020)
- Finding sources, balanced reporting, preparation
News distribution is (now) practically free
-“Everyone has a megaphone” (Wolfsfeld 2014)
Tension between costs and income
- News as a good: High fixed, low marginal cost 每多做一个东西，几乎不花
什么额外成本; non-exclusive 一个东西可以被很多人同时使用，不会互相影响
→ commoditization of news 新闻的商品化
- Incomes (e.g. ads) especially at point of distribution
- More distributors (FB, Google news, born-online sites such as nu.nl)
- Fewer gatherers (AP, CNN, BBC, newspapers, ..)
AI in the The journalistic value chain
News production
- Finding & selecting stories
- Information gathering
- Writing stories
News distribution
- Personalizing news
- Monetizing consumption
Recommender algorithms
What is AI? (1) Machine Learning
（1）Detection of systematic patterns between input and output
（2）General task: Predict output given specific features of the input
(minimizing the prediction errors)
（3）Very similar to “regular” statistical modeling
- Input features: independent variables
- Output class: dependent variable
- (in fact, ‘neural networks’ can be seen as a form of logistic regression
models)
- Key difference to statistical modeling:
We care about predicting something, not about understanding a (causal)
process
Models are highly complex (and multicollinear) and generally seen as
‘black box’
Deep learning
- Fancy term for machine learning with very large models
- Based on:
Very large neural networks (with a specific structure)
Trained on enormous amounts of data, e.g. “all of the internet”
,Using massive computing power, especially of GPUs
Key innovations:
Feature layers find patterns in raw input
Networks can be (pre-)trained based on unannotated data
Patterns from (pre-)training are transferred to actual task, and fine-tuned
on annotated data
Conclusions
AI is a group of techniques clustered around machine learning
AI is transforming society at an increasing pace
AI associated with various problems connected to power
- Bias
- Inequality
Regulating/fixing AI requires a deep understanding of …
- What AI is?
- How it affects (various aspects of) society?
- How solutions/regulation interacts with technology, users, owners,
society
Interdisciplinary research and solutions are key
lec2
Why is “Deep learning” revolutionary
,- Key insights: transfer learning based on unannotated training
material
没有标记的材料
- ML was always limited by lack of (expensive) annotated training data;
steps to find ‘features’ can work on raw data
-- Finding faces, detecting similar words, connotations of words
- Patterns / “Knowledge” extracted from these data sets can be
transferred to new tasks, so not every task has to start from scratch 从旧
知识到引用
- Next week: more details on embeddings 词向量 and move from CNNs
to transformers (GPT)
What makes news special?
What makes theguardian.com different from amazon.com?
Normatively, news supply has direct societal consequences
- Informing citizens (Democratic citizenship)
- Exposing problems (Watchdog role, press as fourth estate) 揭露问题
- Enabling debate (Public sphere) 促进公共讨论
→ It matters what content is recommended
Technically, news has very short shelf life “保质期”非常短
Economically, news is hard to monetize 变现
News gathering vs distribution
, (Good) news gathering is difficult & expensive
-“Professional Journalism is Expensive” (Nielsen 2020)
- Finding sources, balanced reporting, preparation
News distribution is (now) practically free
-“Everyone has a megaphone” (Wolfsfeld 2014)
Tension between costs and income
- News as a good: High fixed, low marginal cost 每多做一个东西，几乎不花
什么额外成本; non-exclusive 一个东西可以被很多人同时使用，不会互相影响
→ commoditization of news 新闻的商品化
- Incomes (e.g. ads) especially at point of distribution
- More distributors (FB, Google news, born-online sites such as nu.nl)
- Fewer gatherers (AP, CNN, BBC, newspapers, ..)
AI in the The journalistic value chain
News production
- Finding & selecting stories
- Information gathering
- Writing stories
News distribution
- Personalizing news
- Monetizing consumption
Recommender algorithms
