Artificial Intelligence
That’s Funny – but AI Models Don’t Get the Joke
Newswise — ITHACA, N.Y. — Large neural networks, a form of artificial intelligence, can generate thousands of jokes along the lines of “Why did the chicken cross the road?” But do they understand why they’re funny?
Using hundreds of entries from the New Yorker magazine’s Cartoon Caption Contest as a testbed, researchers challenged AI models and humans with three tasks: matching a joke to a cartoon; identifying a winning caption; and explaining why a winning caption is funny.
In all tasks, humans performed demonstrably better than machines, even as AI advances such as ChatGPT have closed the performance gap. So are machines beginning to “understand” humor? In short, they’re making some progress, but aren’t quite there yet.
“The way people challenge AI models for understanding is to build tests for them – multiple choice tests or other evaluations with an accuracy score,” said Jack Hessel, Ph.D. ’20, research scientist at the Allen Institute for AI (AI2). “And if a model eventually surpasses whatever humans get at this test, you think, ‘OK, does this mean it truly understands?’ It’s a defensible position to say that no machine can truly `understand’ because understanding is a human thing. But, whether the machine understands or not, it’s still impressive how well they do on these tasks.”
Hessel is lead author of “Do Androids Laugh at Electric Sheep? Humor ‘Understanding’ Benchmarks from The New Yorker Caption Contest,” which won a best-paper award at the 61st annual meeting of the Association for Computational Linguistics, held July 9-14 in Toronto.
Lillian Lee ’93, the Charles Roy Davis Professor in the Cornell Ann S. Bowers College of Computing and Information Science, and Yejin Choi, Ph.D. ’10, professor in the Paul G. Allen School of Computer Science and Engineering at the University of Washington, and the senior director of common-sense intelligence research at AI2, are also co-authors on the paper.
For their study, the researchers compiled 14 years’ worth of New Yorker caption contests – more than 700 in all. Each contest included: a captionless cartoon; that week’s entries; the three finalists selected by New Yorker editors; and, for some contests, crowd quality estimates for each submission.
For each contest, the researchers tested two kinds of AI – “from pixels” (computer vision) and “from description” (analysis of human summaries of cartoons) – for the three tasks.
“There are datasets of photos from Flickr with captions like, ‘This is my dog,’” Hessel said. “The interesting thing about the New Yorker case is that the relationships between the images and the captions are indirect, playful, and reference lots of real-world entities and norms. And so the task of ‘understanding’ the relationship between these things requires a bit more sophistication.”
In the experiment, matching required AI models to select the finalist caption for the given cartoon from among “distractors” that were finalists but for other contests; quality ranking required models to differentiate a finalist caption from a nonfinalist; and explanation required models to generate free text saying how a high-quality caption relates to the cartoon.
Hessel penned the majority of human-generated explanations himself, after crowdsourcing the task proved unsatisfactory. He generated 60-word explanations for more than 650 cartoons.
“A number like 650 doesn’t seem very big in a machine-learning context, where you often have thousands or millions of data points,” Hessel said, “until you start writing them out.”
This study revealed a significant gap between AI- and human-level “understanding” of why a cartoon is funny. The best AI performance in a multiple choice test of matching cartoon to caption was only 62% accuracy, far behind humans’ 94% in the same setting. And when it came to comparing human- vs. AI-generated explanations, humans’ were preferred roughly 2-to-1.
While AI might not be able to “understand” humor yet, the authors wrote, it could be a collaborative tool humorists could use to brainstorm ideas.
Other contributors include Ana Marasovic, assistant professor at the University of Utah School of Computing; Jena D. Hwang, research scientist at AI2; Jeff Da, research assistant at the University of Washington Rowan Zellers, researcher at OpenAI; and humorist Robert Mankoff, president of Cartoon Collections and long-time cartoon editor at the New Yorker.
The authors wrote this paper in the spirit of the subject matter, with playful comments and footnotes throughout.
“This three or four years of research wasn’t always super fun,” Lee said, “but something we try to do in our work, or at least in our writing, is to encourage more of a spirit of fun.”
This work was funded in part by the Defense Advanced Research Projects Agency; AI2; and a Google Focused Research Award.
Source: Cornell University
Author
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
Tech
How close are quantum computers to being really useful? Podcast
Quantum computers could revolutionize science by solving complex problems. However, scaling and error correction remain significant challenges before achieving practical applications.
Quantum computers have the potential to solve big scientific problems that are beyond the reach of today’s most powerful supercomputers, such as discovering new antibiotics or developing new materials.
But to achieve these breakthroughs, quantum computers will need to perform better than today’s best classical computers at solving real-world problems. And they’re not quite there yet. So what is still holding quantum computing back from becoming useful?
In this episode of The Conversation Weekly podcast, we speak to quantum computing expert Daniel Lidar at the University of Southern California in the US about what problems scientists are still wrestling with when it comes to scaling up quantum computing, and how close they are to overcoming them.
Quantum computers harness the power of quantum mechanics, the laws that govern subatomic particles. Instead of the classical bits of information used by microchips inside traditional computers, which are either a 0 or a 1, the chips in quantum computers use qubits, which can be both 0 and 1 at the same time or anywhere in between. Daniel Lidar explains:
“Put a lot of these qubits together and all of a sudden you have a computer that can simultaneously represent many, many different possibilities … and that is the starting point for the speed up that we can get from quantum computing.”
Faulty qubits
One of the biggest problems scientist face is how to scale up quantum computing power. Qubits are notoriously prone to errors – which means that they can quickly revert to being either a 0 or a 1, and so lose their advantage over classical computers.
Scientists have focused on trying to solve these errors through the concept of redundancy – linking strings of physical qubits together into what’s called a “logical qubit” to try and maximise the number of steps in a computation. And, little by little, they’re getting there.
In December 2024, Google announced that its new quantum chip, Willow, had demonstrated what’s called “beyond breakeven”, when its logical qubits worked better than the constituent parts and even kept on improving as it scaled up.
Lidar says right now the development of this technology is happening very fast:
“For quantum computing to scale and to take off is going to still take some real science breakthroughs, some real engineering breakthroughs, and probably overcoming some yet unforeseen surprises before we get to the point of true quantum utility. With that caution in mind, I think it’s still very fair to say that we are going to see truly functional, practical quantum computers kicking into gear, helping us solve real-life problems, within the next decade or so.”
Listen to Lidar explain more about how quantum computers and quantum error correction works on The Conversation Weekly podcast.
This episode of The Conversation Weekly was written and produced by Gemma Ware with assistance from Katie Flood and Mend Mariwany. Sound design was by Michelle Macklem, and theme music by Neeta Sarl.
Clips in this episode from Google Quantum AI and 10 Hours Channel.
You can find us on Instagram at theconversationdotcom or via e-mail. You can also subscribe to The Conversation’s free daily e-mail here.
Listen to The Conversation Weekly via any of the apps listed above, download it directly via our RSS feed or find out how else to listen here.
Gemma Ware, Host, The Conversation Weekly Podcast, The Conversation
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Author
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
Leo Porter, University of California, San Diego and Daniel Zingaro, University of Toronto
What do you think there are more of: professional computer programmers or computer users who do a little programming?
It’s the second group. There are millions of so-called end-user programmers. They’re not going into a career as a professional programmer or computer scientist. They’re going into business, teaching, law, or any number of professions – and they just need a little programming to be more efficient. The days of programmers being confined to software development companies are long gone.
If you’ve written formulas in Excel, filtered your email based on rules, modded a game, written a script in Photoshop, used R to analyze some data, or automated a repetitive work process, you’re an end-user programmer.
As educators who teach programming, we want to help students in fields other than computer science achieve their goals. But learning how to program well enough to write finished programs can be hard to accomplish in a single course because there is so much to learn about the programming language itself. Artificial intelligence can help.
Lost in the weeds
Learning the syntax of a programming language – for example, where to place colons and where indentation is required – takes a lot of time for many students. Spending time at the level of syntax is a waste for students who simply want to use coding to help solve problems rather than learn the skill of programming.
As a result, we feel our existing classes haven’t served these students well. Indeed, many students end up barely able to write small functions – short, discrete pieces of code – let alone write a full program that can help make their lives better.
Tools built on large language models such as GitHub Copilot may allow us to change these outcomes. These tools have already changed how professionals program, and we believe we can use them to help future end-user programmers write software that is meaningful to them.
These AIs almost always write syntactically correct code and can often write small functions based on prompts in plain English. Because students can use these tools to handle some of the lower-level details of programming, it frees them to focus on bigger-picture questions that are at the heart of writing software programs. Numerous universities now offer programming courses that use Copilot.
At the University of California, San Diego, we’ve created an introductory programming course primarily for those who are not computer science students that incorporates Copilot. In this course, students learn how to program with Copilot as their AI assistant, following the curriculum from our book. In our course, students learn high-level skills such as decomposing large tasks into smaller tasks, testing code to ensure its correctness, and reading and fixing buggy code.
Freed to solve problems
In this course, we’ve been giving students large, open-ended projects and couldn’t be happier with what they have created.
For example, in a project where students had to find and analyze online datasets, we had a neuroscience major create a data visualization tool that illustrated how age and other factors affected stroke risk. Or, for example, in another project, students were able to integrate their personal art into a collage, after applying filters that they had created using the programming language Python. These projects were well beyond the scope of what we could ask students to do before the advent of large language model AIs.
Given the rhetoric about how AI is ruining education by writing papers for students and doing their homework, you might be surprised to hear educators like us talking about its benefits. AI, like any other tool people have created, can be helpful in some circumstances and unhelpful in others.
In our introductory programming course with a majority of students who are not computer science majors, we see firsthand how AI can empower students in specific ways – and promises to expand the ranks of end-user programmers.
Leo Porter, Teaching Professor of Computer Science and Engineering, University of California, San Diego and Daniel Zingaro, Associate Professor of Mathematical and Computational Sciences, University of Toronto
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
Tech
From shrimp Jesus to fake self-portraits, AI-generated images have become the latest form of social media spam
Renee DiResta, Stanford University; Abhiram Reddy, Georgetown University, and Josh A. Goldstein, Georgetown University
Suppose you’ve spent time on Facebook over the past six months. In that case, you may have noticed photorealistic images that are too good to be true: children holding paintings that look like the work of professional artists, or majestic log cabin interiors that are the stuff of Airbnb dreams.
Others, such as renderings of Jesus made out of crustaceans, are just bizarre.
Like the AI image of the pope in a puffer jacket that went viral in May 2023, these AI-generated images are increasingly prevalent – and popular – on social media platforms. Even as many of them border on the surreal, they’re often used to bait engagement from ordinary users.
Our team of researchers from the Stanford Internet Observatory and Georgetown University’s Center for Security and Emerging Technology investigated over 100 Facebook pages that posted high volumes of AI-generated content. We published the results in March 2024 as a preprint paper, meaning the findings have not yet gone through peer review.
We explored patterns of images, unearthed evidence of coordination between some of the pages, and tried to discern the likely goals of the posters.
Page operators seemed to be posting pictures of AI-generated babies, kitchens or birthday cakes for a range of reasons.
There were content creators innocuously looking to grow their followings with synthetic content; scammers using pages stolen from small businesses to advertise products that don’t seem to exist; and spammers sharing AI-generated images of animals while referring users to websites filled with advertisements, which allow the owners to collect ad revenue without creating high-quality content.
Our findings suggest that these AI-generated images draw in users – and Facebook’s recommendation algorithm may be organically promoting these posts.
Generative AI meets scams and spam
Internet spammers and scammers are nothing new.
For more than two decades, they’ve used unsolicited bulk email to promote pyramid schemes. They’ve targeted senior citizens while posing as Medicare representatives or computer technicians.
On social media, profiteers have used clickbait articles to drive users to ad-laden websites. Recall the 2016 U.S. presidential election, when Macedonian teenagers shared sensational political memes on Facebook and collected advertising revenue after users visited the URLs they posted. The teens didn’t care who won the election. They just wanted to make a buck.
In the early 2010s, spammers captured people’s attention with ads promising that anyone could lose belly fat or learn a new language with “one weird trick.”
AI-generated content has become another “weird trick.”
It’s visually appealing and cheap to produce, allowing scammers and spammers to generate high volumes of engaging posts. Some of the pages we observed uploaded dozens of unique images per day. In doing so, they followed Meta’s own advice for page creators. Frequent posting, the company suggests, helps creators get the kind of algorithmic pickup that leads their content to appear in the “Feed,” formerly known as the “News Feed.”
Much of the content is still, in a sense, clickbait: Shrimp Jesus makes people pause to gawk and inspires shares purely because it is so bizarre.
Many users react by liking the post or leaving a comment. This signals to the algorithmic curators that perhaps the content should be pushed into the feeds of even more people.
Some of the more established spammers we observed, likely recognizing this, improved their engagement by pivoting from posting URLs to posting AI-generated images. They would then comment on the post of the AI-generated images with the URLs of the ad-laden content farms they wanted users to click.
But more ordinary creators capitalized on the engagement of AI-generated images, too, without obviously violating platform policies.
Rate ‘my’ work!
When we looked up the posts’ captions on CrowdTangle – a social media monitoring platform owned by Meta and set to sunset in August – we found that they were “copypasta” captions, which means that they were repeated across posts.
Some of the copypasta captions baited interaction by directly asking users to, for instance, rate a “painting” by a first-time artist – even when the image was generated by AI – or to wish an elderly person a happy birthday. Facebook users often replied to AI-generated images with comments of encouragement and congratulations
Algorithms push AI-generated content
Our investigation noticeably altered our own Facebook feeds: Within days of visiting the pages – and without commenting on, liking or following any of the material – Facebook’s algorithm recommended reams of other AI-generated content.
Interestingly, the fact that we had viewed clusters of, for example, AI-generated miniature cow pages didn’t lead to a short-term increase in recommendations for pages focused on actual miniature cows, normal-sized cows or other farm animals. Rather, the algorithm recommended pages on a range of topics and themes, but with one thing in common: They contained AI-generated images.
In 2022, the technology website Verge detailed an internal Facebook memo about proposed changes to the company’s algorithm.
The algorithm, according to the memo, would become a “discovery-engine,” allowing users to come into contact with posts from individuals and pages they didn’t explicitly seek out, akin to TikTok’s “For You” page.
We analyzed Facebook’s own “Widely Viewed Content Reports,” which lists the most popular content, domains, links, pages and posts on the platform per quarter.
It showed that the proportion of content that users saw from pages and people they don’t follow steadily increased between 2021 and 2023. Changes to the algorithm have allowed more room for AI-generated content to be organically recommended without prior engagement – perhaps explaining our experiences and those of other users.
‘This post was brought to you by AI’
Since Meta currently does not flag AI-generated content by default, we sometimes observed users warning others about scams or spam AI content with infographics.
Meta, however, seems to be aware of potential issues if AI-generated content blends into the information environment without notice. The company has released several announcements about how it plans to deal with AI-generated content.
In May 2024, Facebook will begin applying a “Made with AI” label to content it can reliably detect as synthetic.
But the devil is in the details. How accurate will the detection models be? What AI-generated content will slip through? What content will be inappropriately flagged? And what will the public make of such labels?
While our work focused on Facebook spam and scams, there are broader implications.
Reporters have written about AI-generated videos targeting kids on YouTube and influencers on TikTok who use generative AI to turn a profit.
Social media platforms will have to reckon with how to treat AI-generated content; it’s certainly possible that user engagement will wane if online worlds become filled with artificially generated posts, images and videos.
Shrimp Jesus may be an obvious fake. But the challenge of assessing what’s real is only heating up.
Renee DiResta, Research Manager of the Stanford Internet Observatory, Stanford University; Abhiram Reddy, Research Assistant at the Center for Security and Emerging Technology, Georgetown University, and Josh A. Goldstein, Research Fellow at the Center for Security and Emerging Technology, Georgetown University
This article is republished from The Conversation under a Creative Commons license. Read the original article.
The science section of our news blog STM Daily News provides readers with captivating and up-to-date information on the latest scientific discoveries, breakthroughs, and innovations across various fields. We offer engaging and accessible content, ensuring that readers with different levels of scientific knowledge can stay informed. Whether it’s exploring advancements in medicine, astronomy, technology, or environmental sciences, our science section strives to shed light on the intriguing world of scientific exploration and its profound impact on our daily lives. From thought-provoking articles to informative interviews with experts in the field, STM Daily News Science offers a harmonious blend of factual reporting, analysis, and exploration, making it a go-to source for science enthusiasts and curious minds alike. https://stmdailynews.com/category/science/
Author
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
-
Urbanism1 year agoSignal Hill, California: A Historic Enclave Surrounded by Long Beach
-
News2 years agoDiana Gregory Talks to us about Diana Gregory’s Outreach Services
-
Senior Pickleball Report2 years agoThe Absolute Most Comfortable Pickleball Shoe I’ve Ever Worn!
-
STM Blog2 years agoWorld Naked Gardening Day: Celebrating Body Acceptance and Nature
-
Senior Pickleball Report2 years agoACE PICKLEBALL CLUB TO DEBUT THEIR HIGHLY ANTICIPATED INDOOR PICKLEBALL FRANCHISES IN THE US, IN EARLY 2023
-
Travel2 years agoUnique Experiences at the CitizenM
-
Automotive2 years ago2023 Nissan Sentra pricing starts at $19,950
-
Senior Pickleball Report2 years ago“THE PEOPLE’S CHOICE AWARDS OF PICKLEBALL” – VOTING OPEN