Google’s machine learning ‘course’ for journalists

I couldn’t resist dipping into this free course from the Google News Initiative, and what I found surprised me: eight short lessons that are available as PDFs.

The good news: The lessons are journalism-focused, and they provide a painless introduction to the subject. The bad news: This is not really a course or a class at all — although there is one quiz at the end. And you can get a certificate, for what it’s worth.

There’s a lot here that many journalists might not be aware of, and that’s a plus. You get a brief, clear description of Reuters’ News Tracer and Lynx Insight tools, both used in-house to help journalists discover new stories using social media or other data (Lesson 1). A report I recall hearing about — how automated real-estate stories brought significant new subscription revenue to a Swedish news publisher — is included in a quick summary of “robot reporting” (also Lesson 1).

Lesson 2 helpfully explains what machine learning is without getting into technical operations of the systems that do the “learning.” They don’t get into what training a model entails, but they make clear that once the model exists, it is used to make predictions. The predictions are not like what some tarot-card reader tells you but rather probability-based results that the model is able to produce, based on its prior training.

Noting that machine learning is a subset of the wider field called artificial intelligence is, of course, accurate. What is inaccurate is the definition “specific applications that use data to train a model to perform a given task independently and learn from experience.” They left out Q-learning, a type of reinforcement learning (a subset of machine learning), which does not use a model. It’s okay that they left it out, but they shouldn’t imply that all machine learning requires a trained model.

The explosion of machine learning and AI in the past 10 years is explained nicely and concisely in Lesson 2. The lesson also touches on misconceptions and confusion surrounding AI:

“The lack of an officially agreed definition, the legacy of science-fiction, and a general low level of literacy on AI-related topics are all contributing factors.”

—Lesson 2, Is Machine Learning the same thing as AI?

I’ll be looking at Lessons 3 and 4 tomorrow.

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

Free courses in machine learning

Two days ago, I came upon this newly published course from FastAI: Practical Deep Learning for Coders. I actually stumbled across it via a video on YouTube, which I’ve watched now, and it made me feel optimistic about the course. I’m in the middle of the CS50 AI course from Harvard, though, so I need to hold off on the FastAI course for now.

Above: Screenshot from FastAi course

The first video got me thinking.

First, they said (as many others have said) that Python is the main programming language used for machine learning today. (This makes me happy, as I know Python.) But I wonder whether there’s more to this claim than I’m aware of.

Second, they said PyTorch has superseded TensorFlow as the framework of choice for machine learning. They said PyTorch is “much easier to use and much more useful for researchers.”

“Within the last 12 months, the percentage of papers at major conferences that use PyTorch has gone from 20 percent to 80 percent and vice versa — those that use TensorFlow have gone from 80 percent to 20 percent.”

—Jeremy Howard, in the FastAI video “Lesson 1 – Deep Learning for Coders (2020)”

Note, I don’t know if this is true. But it caught my attention.

FastAI is a library “that sits on top of PyTorch,” they explain. They say it is “the most popular higher-level API for PyTorch,” and it removes a lot of the struggle necessary to get started with PyTorch.

This leads me back to the CS50 AI course. The CS50 phenomenon was documented in The New Yorker in July 2020. One insanely popular course, Introduction to Computer Science, has spawned multiple follow-on courses, including the seven-module course about the principles of artificial intelligence in which I am currently enrolled (not for credit).

In the various online forums and Facebook groups devoted to CS50, you can see a lot of people asking whether they need to take the intro course prior to starting the AI course. Some of them admit they have never programmed before. They know nothing about coding. But they think they might take an AI programming course as their very first computer science course.

This is what I was thinking about as the speakers in the FastAI video both praised how easy FastAI makes it to train a model and cautioned that machine learning is not a task for code newbies.

Training and testing a machine learning system — a system that will make predictions to be used in some industry, some social context, where human lives might be affected — should not be dependent on someone who learned how to do it in one online course.

Some other free online courses:

.

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

How weird is AI?

For Friday AI Fun, I’m sharing one of the first videos I ever watched about artificial intelligence. It’s a 10-minute TED Talk by Janelle Shane, and it’s actually pretty funny. I do mean “funny ha-ha.”

I’m not wild about the ice-cream-flavors example she starts out with, because what does an AI know about ice cream, anyway? It’s got no tongue. It’s got no taste buds.

But starting at 2:07, she shows illustrations and animations of what an AI does in a simulation when it is instructed to go from point A to point B. For a robot with legs, you can imagine it walking, yes? Well, watch the video to see what really happens.

This brings up something I’ve only recently begun to appreciate: The results of an AI doing something may be entirely satisfactory — but the manner in which it produces those results is very unlike the way a human would do it. With both machine vision and game playing, I’ve seen how utterly un-human the hidden processes are. This doesn’t scare me, but it does make me wonder about how our human future will change as we rely more on these un-human ways of solving problems.

“When you’re working with AI, it’s less like working with another human and a lot more like working with some kind of weird force of nature.”

—Janelle Shane

At 6:23 in the video, Shane shows another example that I really love. It shows the attributes (in a photo) that an image recognition system decided to use when identifying a particular species of fish. You or I would look at the tail, the fins, the head — yes? Check out what the AI looks for.

Shane has a new book, You Look Like a Thing and I Love You: How Artificial Intelligence Works and Why It’s Making the World a Weirder Place. I haven’t read it yet. Have you?

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

What is called ‘AI’ but really isn’t

Because “artificial intelligence” and “AI” have become such potent buzzwords in business — and so many firms are trying to sell some kind of “AI” system or software or strategy to every business possible — we should all take a step back and evaluate whether there is actual AI operating in some of these systems.

That won’t always be easy to discern. If a company claims there is “AI” in its product, they are not going to divulge exactly how it works. If they want to convince you, their literature or their engineers will likely throw out a tangled net of terms that, while accurate, might not help anyone but another engineer understand what’s inside the black box.

I was thinking about this recently as I worked on assignments for an online computer science course in AI. One of the early projects was to program a tic-tac-toe game in which a human can play against “an AI.” Just like most humans, the AI can force a tie in every tic-tac-toe game unless the human makes a mistake, and then the human will lose. I wrote the code that enables the AI to play — that was the assignment. But I didn’t invent the code from nothing. I was taught in the course to use an algorithm called minimax. Further, I was encouraged to make my program faster by using another algorithm called alpha-beta pruning.

Illustration of alpha-beta pruning (Wikipedia, by Jez9999, GNU license)

There is no machine learning involved in those two algorithms. They are simply a time-tested way for a computer language to direct a certain kind of look-ahead in a two-player game (not only tic-tac-toe).

Don’t despair or tune out — look at the diagram and understand that the computer, through instructions in my code, is able to rapidly advance through every possible outcome in tic-tac-toe and see how to: (a) prevent a win for the opponent, and (b) win if a win is possible.

There is no magic here.

Tic-tac-toe with “AI” playing X, human playing O.

Another assignment in the same course has the students programming “an AI” that plays Minesweeper. This game is quite different from tic-tac-toe in that there is only one player, and there is hidden knowledge: The player doesn’t know where the mines are. One move at a time, the player builds knowledge about the game board.

Completed Minesweeper game, with AI playing all moves.

A human player doesn’t click on a mine, because she chooses squares that are next to a 0 (indicating no mines touch that square) and marks a mine square when it becomes obvious that a mine is hidden there.

The “AI” builds knowledge in a way that it is programmed to do (that is the assignment). In this case, there is no pre-existing algorithm, but there are principles of logic. I programmed “knowledge” that was stored in the program each time the AI clicked a square and a number was revealed. The knowledge is: (a) that number, and (b) the coordinates of all the surrounding squares. Thus the AI “knows” that, for example, among eight specified squares there are two mines.

If among eight specified squares there are zero mines, my code tells the AI to mark all eight of those squares as safe. My code also tells the AI that if there are any safe moves left to be made, then make a safe move. If not, make a random move. That is the only time when the AI can possibly set off a mine.

Once again, there is no magic here.

In contrast to these two simple examples of a computer successfully playing a game, AlphaGo (which I wrote about previously) uses real AI and could not have beaten a human Go master otherwise. Some games can’t be programmed with only simple algorithms or logic — if they are to win, they need something akin to intuition.

Programming a computer to develop and use an approximation of human intuition is what we have in today’s machine learning with deep neural networks. It’s still not magic, but it’s a lot more complicated than the kind of strictly mapped-out processes I wrote for playing tic-tac-toe or Minesweeper.

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

Who labels the data for AI?

In yesterday’s post, I referred to the labels that are required for supervised machine learning. To train a model — which enables an AI system to correctly identify or sort images or documents or iris flowers (and so much more) — each data record must include one or more labels. For an image of a dog, for example, the labels might be dog and Great Dane. For an iris flower, the label is the name of the exact species of that individual flower.

Nowadays there are people all around the world sitting at computers and labeling data.

In the 6-minute video above, BBC journalist Dave Lee travels to Kenya, where about 2,000 people work in a Nairobi office for Samasource, which produces training data for use in machine learning.

You’ll see exactly how every single item in one video frame is marked and tagged — this is what a vision system for a self-driving car needs if it is to avoid crashing into mailboxes or people.

In the Nairobi office, 52 percent of the workers are women. The pay is terribly low by Silicon Valley standards, but high for Kenya. Lee doesn’t gloss over this aspect of the story — in fact, it’s central to the telling.

Financial Times journalist Madhumita Murgia wrote about Samasource in July 2019. Her story also covers iMerit, a similar company with offices in Kolkata, India, as well as California and Louisiana.

“An hour of video takes eight hours to annotate. In fact, a McKinsey report from 2018 listed data labeling as the biggest obstacle to AI adoption in industry.”

—Financial Times

Some very large and widely used datasets such as ImageNet were labeled by self-employed workers for extremely low rates of pay — often through the Amazon-owned Mechanical Turk crowdsourcing website (which also offers up far worse tasks for similarly low compensation). In contrast, Samasource’s CEO Leila Janah told Murgia that the company’s pay rate is “almost quadruple” the previous income of their workers in developing countries.

Janah also pointed out that these workers are not just labeling cats and dogs. They have been trained, for example, to label diseased cells in photos of cross-sections of plants for one particular project. They are providing real human intelligence that is specialized to very particular problem sets.

Fortune journalist Jeremy Kahn wrote about other companies that also provide data-labeling services for top multinational firms. Labelbox and Scale AI have received heaps of funding from venture capitalists, but I couldn’t find any information about their workers who label the data. Is this something we should be concerned about? Probably so.

Both Samasource and iMerit are upfront about who their workers are and where they do the work (this might have changed since the spread of COVID-19 in early 2020). Are the dozens of other companies supplying labeled data to corporations and universities in the wealthy countries paying their workers a living wage?

“Often companies have a need for both general and more expert labeling and employ a combination of outsourcing firms, freelancers, and in-house experts to affix these annotations.”

—Fortune

Labelbox, in fact, doesn’t employ people who do the labeling work, according to Fortune. It provides “a tool for managing labeling projects and data across different contract labelers, who often work for large outsourcing firms.”

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

ImageNet and labels for data

Supervised learning is a type of machine learning in which a model is trained using labeled data. You begin with a very large collection of labeled data. (In the case of ImageNet, the data were all digital images. For the Iris Data Set, the data all refer to individual iris flowers, which can be divided into three related species. For the MNIST dataset, the data are images of about 70,000 handwritten numbers, 0 through 9.)

You divide the dataset into two parts, the training data and the test data. The split might be 70/30, or 80/20. You don’t choose which data goes into which group. Then you run the training data many, many, many times, adjusting certain parameters in the code along the way, until the code consistently returns good results — that is, the thing the code identifies (an object in an image, an iris species, a number) matches the label (which is hidden from the code).

At that point, you have a trained model. You feed the test data set to it and see whether the accuracy rate is also high. (It’s important that none of the test data were used to train the model.) Again, the proof is in the labels.

In a later post I will discuss how data come to be labeled. (Hint: It’s not elves.) In this post, I will discuss bad labels. Specifically, I want to highlight the work that AI researcher Kate Crawford and artist-researcher Trevor Paglen did around the famous ImageNet dataset.

In the video above, Crawford and Paglen present this work and show a lot of great examples. They also published a long article about the work, if you’d rather read than watch.

ImageNet is a huge collection of labeled images. More than 14 million images. They were labeled according to a set of categories and synonym groupings from WordNet, an English-language lexical database. The images were labeled by humans.

And that, it seems, is at the root of the problem.

Crawford and Paglen were interested in the ImageNet photos of people. Person is a category in WordNet. Within the category, there are many descriptive terms for people, such as “cheerleaders, scuba divers, welders, Boy Scouts, fire walkers, and flower girls.” So the photos of people in ImageNet are labeled with these terms. However, not all terms are neutral.

“A young man drinking beer is categorized as an ‘alcoholic, alky, dipsomaniac, boozer, lush, soaker, souse.’ A child wearing sunglasses is classified as a ‘failure, loser, non-starter, unsuccessful person.’”

—Crawford and Paglen

You might say, well, where’s the harm? They are only labels in a database, after all.

The ImageNet database has been used to train many convolutional neural networks used in image-recognition software.

When you feed a photo of yourself into an image-recognition application, you might be surprised at the labels that are applied to you. For example, an image of Paglen (a white man with a shaved head) was labeled as “Klansman, Ku Kluxer.”

Paglen built a web app called ImageNet Roulette so that anyone could upload a photo of themselves or a friend and see what labels were applied. (The app is no longer online.) It became clear that perfectly innocuous people in photos were being labeled as criminals or dangerous, or with racist or sexist terms.

About 952,000 of ImageNet’s 14 million images were in the person category as of 2010 (source). Many of those images — with their labels — were removed after the opening of Crawford and Paglen’s art exhibition, Training Humans, in Milan in September 2019.

ImageNet has been used to train countless image-recognition systems since 2010.

Additional information:

Leading online database to remove 600,000 images after art project reveals its racist bias (September 2019), The Art Newspaper.

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

Ask a computer to draw what it sees

If a computer can correctly identify an object (an apple, a tricycle) or an animal such as a zebra, can it produce a drawing of that object or animal? This is something most people can do, even if their drawing skills are minimal. After all, almost anyone can play Pictionary.

This 8-minute video shows us what happened when a programmer-artist reversed the process of an AI that recognizes objects and animals in digital images. I really admire the deft storytelling here.

Object recognition has improved amazingly in the past 10 years, but that does not mean these AI systems see the same way as a human does. In some cases, that might not matter at all. In other cases, it can mean the difference between life and death.

In yesterday’s post I mentioned the way a convolutional neural network (part of a machine learning system) processes an image through many stacked layers of detection units (sometimes called neurons), identifying edges and shapes that eventually lead to a conclusion that the image is likely to contain such-and-such an object, animal, or person. Today’s video shows a bit more about the training process that an AI goes through before it can perform these identifications.

Training is necessary in the type of machine learning called supervised learning. The training data (in this case, digital images of objects and animals) must be labeled in advance. That is, the system receives thousands of images labeled “tiger” before it is able to recognize a tiger in a random photo or video. If a system can identify 20 different animals, that system was trained on thousands of images of each animal.

If the system was never trained on tigers, it cannot recognize a tiger.

So today’s video gives us a nice glimpse into how and why that training works, and what its limitations are. What’s really fascinating to me, though, are the images produced by programmer-artist Tom White‘s system.

“I have created a drawing system that allows neural networks to produce abstract ink prints that reveal their visual concepts. Surprisingly, these prints are recognized not only by the neural networks that created them, but also universally across most AI systems which have been trained to recognize the same objects.”

—Tom White

In the video, you’ll see that humans cannot recognize what the AI drew. The rendering is too abstract, too unlike what we see and what we would draw ourselves. Note what White says, though, about other AI systems: they can recognize the object in these AI-produced drawings.

This is, I think, related to what is called adversarial AI, which I’ll discuss in a future post.

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

AI programs that play games

One of the very best media items I’ve found is this feature-length documentary about the program that beat an international master at the game of Go in 2016. It’s excellent as a documentary film — well-paced, sparking curiosity, exciting in some parts, and never pedantic.

You don’t need to understand anything about the game (which is immensely popular in China, Japan, and Korea, but not widely played elsewhere). It’s explained visually so that you can appreciate what’s going on. The film is free to watch on YouTube.

As a resource for learning about AI — or, more specifically, about machine learning — the film excels at helping us understand the work of the team of humans that created and trained the AlphaGo program. We don’t see a lot of people sitting at computer keyboards, typing. There are clustered people pointing at a screen, talking enthusiastically, or saying, “What happened there? Why did it do that?”

Probably my favorite moment in the film is after Lee Se-dol, the human Go master, has played a move that is so great, it was later referred to as “the God move.” The AlphaGo team begins analyzing the program’s responses in real-time, watching the graphs of its probability calculations on a large screen in their command center. For all the talk of AI as a black box that makes decisions humans can”t comprehend, this scene demonstrates that AI can be made transparent and accountable.

There’s much, much more to love about this documentary. The director, Greg Kohs, had extraordinary access to the DeepMind team during the months leading up to the five-game match with Lee. In the end, Google financed a general-audience-friendly film. (Google acquired DeepMind in 2014.)

In an interview with CNET, Kohs said the film “had very modest beginnings.”

“A couple members of Google’s creative lab that I’d worked with before gave me a ring and said we’d have access behind the curtain with [DeepMind founder and CEO] Demis Hassabis and his team. So I jumped on board with the expectation we would just film what happens for archival purposes and then put it on a shelf on a hard drive and that would be the end of it.”

Greg Kohs

Another wonderful aspect of the film is its humanity. I’ve seen a fair number of “scare essays” that predict the end of everything as AI gains dominance over its creators — but here we hear a more nuanced and thought-provoking set of views and reactions.

First, there is Lee, possibly the best (human) Go player who has ever lived, in closeup, in the very moment of his realization that the machine has bested him. Then there are the other Go experts, who understand more than you or I what the machine has actually done. Finally, there are the team members of DeepMind, who built the machine. Of course they are happy, ecstatically happy — but they are humbled, and even awed, as well.

At the end of 2019, Lee Se-dol retired as a professional Go player, at age 36. He is the only human who has ever defeated AlphaGo in tournament play.

More about AlphaGo:

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.

.

Getting thrown into machine learning

Early in 2018, I had several senior journalism students who wanted to learn about machine learning. I knew nothing about it, and they knew that, and we plowed forward together.

The three student teams chose these topics:

  • Sentiment analysis on subreddits for NBA teams
  • Analysis of county court documents naming our university
  • Analysis of tweets by one news organization for audience reactions, engagements

We quickly learned that knowing Python was a big plus. (Fortunately, we all knew Python.) Each of the teams found a different Python library to work with, and after a few weeks, projects were completed and demonstrated — although desired results were not achieved in all cases.

I crammed information mainly from two sources — a YouTube video series called Machine Learning Recipes with Josh Gordon, and something I’ve lost that explained in detail how a model was trained on the Iris Data Set. These provided a surprisingly solid foundation for beginning to understand how today’s machine learning projects are done.

Above: Histograms and features from the Iris Data Set

Since then, I’ve continued to read casually about AI and machine learning. As more and more articles have appeared in the general press and news reports about face recognition and self-driving cars (among other topics related to AI), it’s become clear to me that journalism students need to know more about these technologies — if for no other reason than to avoid being bamboozled by buzzword-spewing politicians or tech-company flacks.

Since May 2020, I’ve been collecting resources, reading and researching, with an intention to teach a course about AI for communications students in spring 2021. This new blog is going to help me organize and prioritize articles, posts, videos, and more.

If it helps other people get a handle on AI, so much the better!

Creative Commons License
AI in Media and Society by Mindy McAdams is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Include the author’s name (Mindy McAdams) and a link to the original post in any reuse of this content.