Daily Archives: February 19, 2022

You might be confident in your ability to tell a real face from one created using artificial intelligence. But a new study has found that your chance of choosing accurately would be slightly better if you just flipped a coinand you are more likely to trust the fake face over the real one.

Published in the Proceedings of the National Academy of Sciences, the study was conducted by Hany Farid, a professor at the University of California, Berkeley, and Sophie J. Nightingale, a lecturer at Englands University of Lancaster.

Farid has been exploring synthetic imagesand how well people can tell them apart from the real onesfor years. He initially focused on the rise of computer-generated imagery. But the mediums path has accelerated in recent years as deep-learning-based neural networks known as GANs (generative adversarial networks) have become more sophisticated at generating truly realistic synthetic images. If you look at the rate of improvement of deep fakes and [GANs], its an order of magnitude faster than CGI, he says. We would argue that we are through the uncanny valley for still faces.

The problems with such realistic fakes are manifold. Fraudulent online profiles are a good example. Fraudulent passport photos. Still photos have some nefarious usage, Farid says. But where things are going to get really gnarly is with videos and audio.

Given the speed of these improvements, Farid and Nightingale wanted to explore if faces created via artificial intelligence were able to convince viewers of their authenticity. Their study included three experiments aimed at understanding whether people can discern a real face from a synthetic one created by Nvidias StyleGAN2. After identifying 800 images of real and fake faces, Farid and Nightingale asked participants to look at a selection of them and sort them into real and fake. Participants were correct less than half the time, with an average accuracy of 48.2%.

A second experiment showed that even giving participants some tips on spotting AI-generated faces and providing feedback as they made their decisions didnt drastically improve their deciphering ability. Participants identified which face was real and which was fake with 59% accuracy, but saw no improvement over time. Even with feedback, even with training trying to make them better, they did slightly better than chance, but theyre still struggling, Farid says. Its not like they got better and betterbasically it helps a little bit, and then it plateaus.

The difficulty people had spotting faces created by artificial intelligence didnt particularly surprise Farid and Nightingale. They didnt anticipate, however, that when participants were asked to rate a set of real and fake faces based on their perceived trustworthiness, people would find synthetically generated faces 7.7% more trustworthy than real onesa small but statistically significant difference.

We were really surprised by this result because our motivation was to find an indirect route to improve performance, and we thought trust would be thatwith real faces eliciting that more trustworthy feeling, Nightingale says.

Farid noted that in order to create more controlled experiments, he and Nightingale had worked to make provenance the only substantial difference between the real and fake faces. For every synthetic image, they used a mathematical model to find a similar one, in terms of expression and ethnicity, from databases of real faces. For every synthetic photo of a young Black woman, for example, there was a real counterpart.

Though the type of images GANs can convincingly create at the moment are still limited to passport-style photos, Nightingale says the deceptions pose a threat for everything from dating scams to social media.

In terms of online romance scams, these images would be perfect, she says. [For] things like Twitter disinformation attacks, rather than having a default egg image, you just take one of these images. People trust it, and if you trust something, youre probably more likely to share it. So you see how these types of images can already cause chaos.

So how do we protect against people using synthetic images for nefarious means? Farid is a champion of an approach called controlled capture, which is being built out by companies like TruePic and the Coalition for Content Provenance and Authentication. The technology captures metadata related to time and location for any photo taken within an app that has built-in camera function.

I think the only solution is to authenticate at the point of recording, using a controlled capture-type of technology, he says. And then, anything that has that, good; anything that doesnt, buyer beware. I think this [solution] is really going to start to get some traction, and my hope is in the coming years, we start taking trust and security more seriously online.

Beyond synthetic still images, the study comes as the world of synthetic media is growing. Synthesia, an Australian company, closed a $50 million series B round in December for AI avatars used in corporate communications; and London-based Metaphysic, the company behind viral deep fakes of Tom Cruise, raised $7.5 million earlier this year. As these technologies continue to improve and change whats possible to do with AI, Nightingale says researchers and companies will have to think seriously about the ethics involved.

If the risks are greater than the benefits of some new technology, should we really be doing it at all? she asks. First of all, should we be developing it? And second, should we be uploading it to something like Github, where anyone can just get their hands on it? . . . As we see, once its out there, we cant just take it back again because people have downloaded it, and its too late.

Answers1. Fake2. Real3. Real4. Fake

Original post:

AI-generated faces are now more trustworthy than real ones - Fast Company

DeepMind wants to use artificial intelligence to help scientists experiment with nuclear fusion, which it believes is a contender for "a source of clean, limitless energy" here on Earth.

The company says it collaborated with the Swiss Plasma Center at the EPFL technical university in Switzerland "to develop the first deep reinforcement learning (RL) system" devoted to the tools researchers are using to assess the nuclear fusion's viability as an energy source.

That reinforcement learning system was designed to "autonomously discover how to control" a tokamak, which DeepMind says is "a doughnut-shaped vacuum surrounded by magnetic coils" that is "used to contain a plasma of hydrogen that is hotter than the core of the Sun."

It turns out that experimenting with something hotter than the Sun can be difficult. EPFL says that if a tokamak's settings aren't carefully managed the plasma within "could collide with the vessel walls and deteriorate." So researchers have to run their experiments in simulators first.

But those simulators can be hard to use, too, not least because of time constraints. DeepMind says that "plasma simulators are slow and require many hours of computer time to simulate one second of real time." That's hardly ideal for scientists racing to investigate nuclear fusion.

It's also where AI comes in. DeepMind and the Swiss Plasma Center published a study in Nature describing a system that's said to have allowed them to create "controllers that can both keep the plasma steady and be used to accurately sculpt it into different shapes" for further research.

"Similar to progress weve seen when applying AI to other scientific domains," DeepMind says, "our successful demonstration of tokamak control shows the power of AI to accelerate and assist fusion science, and we expect increasing sophistication in the use of AI going forward."

Sign up for What's New Now to get our top stories delivered to your inbox every morning.

This newsletter may contain advertising, deals, or affiliate links. Subscribing to a newsletter indicates your consent to our Terms of Use and Privacy Policy. You may unsubscribe from the newsletters at any time.

Read more from the original source:

DeepMind Teaches AI to Assist With Nuclear Fusion Experiments - PCMag

Foundation Fellow Nathan Safir is happiest when working on a complicated problem

As Nathan Safir was finishing up his masters in artificial intelligence at the University of Georgia, he had a big decision to make: Did he want to accept the Marshall Scholarship to pursue a Ph.D. in computer science in England or accept an offer to start work at Google this summer?

Safir, a Foundation Fellow, applied his problem-solving brain to the decision.

Mostly I tried to simplify it into a decision of Am I more excited to work now or am I more excited to go to grad school now? said Safir, who has a B.S. in computer science with a minor in geography.

Safir in the Institute for Artificial Intelligence in the Boyd Research and Education Center. (Photo by Chamberlain Smith/UGA)

Google was the winner. For now. Safir hopes to get his Ph.D. in artificial intelligence sometime in the future.

Whatever way he gets there, the end goal is the same he wants to work in artificial intelligence. Whether its working on a new cool application, existing AI methods, or actually doing the research, Id really love to work somewhere in that space, said Safir.

This flexibility will serve Safir well at Google, because he doesnt yet know what hell be doing at the company when he arrives in California this August. Theyll start team matching closer to my start date, said Safir. All I know is that Ill be in the Bay Area and possibly working on the Google Ads team.

During a Google internship in summer 2021, he worked with the Cloud AI team.

UGA has helped him find his passion for AI in several ways. One was a difficult math class his first year in college. I was a super eager freshman, so I took Math 3500. Its notorious for being very tough, and it was. I was glad I took it, but I realized I was less interested in the proof-based rigorous mathematics.

Nathan Safir outside of the Institute for Artificial Intelligence in the Boyd Research and Education Center. (Photo by Chamberlain Smith/UGA)

He decided to pick up machine learning instead. It presented an interesting way to use mathematical thinking and be able to think creatively.

Also helpful in guiding his path was UGAs Foundation Fellowship, a merit-based scholarship based in the Jere W. Morehead Honors College that is awarded to between 20-25 high school seniors out of 1,200 applicants. Its the reason Safir, who grew up in Kansas City, Missouri, ultimately chose UGA out of his ranked list of 13 potential schools.

He said he most appreciates the travel opportunities, although theyve been a bit truncated during the COVID years, and the people hes met along the way. The network of people you can talk to about careers or cool places you could work that type of thing was super helpful for me trying to imagine what I wanted to do in the future. Safir said he interacts with staff, peers and friends from the Fellowship on a daily basis.

Safirs gateway into programming was a documentary he saw in middle school on Mark Zuckerberg. It said he learned C++ from a book when he was about my age so I checked out the same C++ book from the library. I dont think I picked it up as quickly as he did, but that was the beginning of me trying to learn programming.

Ive been into math and quantitative problem solving since I was really young. Programming was just an extension of that logical problem-solving, he said.

Safir is a down-to-earth, well-rounded person who can move between playing in local Monday pick-up hockey games, coaching Athens Youth Hockey and working on his tennis game, to high-level problem-solving. But hes happiest when he has a tough question to work on.

My thesis is working on a theoretical problem of how to extend an architecture called variational autoencoders to work with both labeled and unlabeled data, he said. My lab, led by Dr. Quinn of the CS department, is at the intersection of biomedical imaging and artificial intelligence. Its a theoretical problem inspired by a real-world problem. Its an especially open fieldI feel like there are a million different things you can work on. Im just happy I can be a part of it.

Continue reading here:

An interest in AI led this student to a job at Google - University of Georgia

From music streaming platforms to social media feeds and search engines, algorithms are used behind the scenes to tailor services to the unique preferences of individuals. Though the use of algorithms has been explored in health care since the origins of artificial intelligence, new strides in deep learning methods over the last decade are allowing clinicians to go after mass amounts of data that were previously inaccessible, transforming how doctors and clinical researchers detect, diagnose and treat disease.

In addition to higher data-computing capacities and advanced algorithms, clinicians can now input data through written and spoken words rather than only quantitative lab and imaging results. As they talk with patients about subjective feelings and pain levels, detailed interpretations can be coded to augment poking and prodding data collected through sensors, giving machine-learning algorithms a fuller picture. With enough input, algorithms will be able to output a series of patterns which physicians can then use in their clinical practice for better diagnoses and understandings of disease.

What has happened in the last eight or so years is that new results coming out of these deep learning methods are allowing us to go after hard data problems that werent accessible before, said Don Brown, the University of Virginias senior associate dean for research, Quantitative Foundation Distinguished Professor in Data Science and a professor in the Department of Engineering Systems and Environment.

Five years ago, Brown was approached by pediatric gastroenterologist Dr. Sana Syed, who was then working as a gastroenterology fellow in her first year on the UVA faculty. Syed was part of a team researching environmental enteropathy, a disease caused by long-term exposure to poor sanitation and hygiene that results in intestinal inflammation and is a major contributor to growth stunting in children living in developing countries around the world.

During her time as a fellow, Syeds role was essentially that of an algorithm. By sifting through thousands of images taken by a video capsule endoscopy a tiny wireless camera that works its way through a patients gut in six to eight hours Syed was tasked with tagging any abnormal images, such as polyps, bleeding and ulcers, to understand tissue structure and function in the context of inflammation. The identification of disease patterns would ideally minimize the need for doctors to conduct endoscopies (the collection of tissue from a patients intestine, which is not a sustainable practice in many low- and middle-income countries) to confirm diagnoses. The tedious task pushed Syed and her colleagues to explore the use of AI as a way of picking up patterns of disease in tissue samples, saving countless hours of analysis.

That is when Syed approached Don Brown in a pursuit to intersect big data and medicine. With such an accomplishment, physicians could begin to more quickly and more accurately predict important inflammatory bowel disease outcomes not only assisting physicians, but also giving patients more peace of mind.

What is going to happen is that a patient shows up, and their data is plugged into these larger algorithms that are trained to learn off patterns that may be representative of you. Then you will be able to predict a specific patients outcome, Syed said. The idea is that you will be able to get more tailored data and have the ability to give specific risk percentages.

In addition to identifying patterns in environmental enteropathy, Syed is researching the effectiveness of AI in other inflammatory bowel diseases, such as Crohns and celiac disease. According to Syed, algorithms that predict scar tissue development in Crohns patients, for example, or individualized risk percentages for thyroid disease and diabetes in celiac patients, would be game-changers in determining more accurate diagnosis and likelihood of disease, allowing clinicians to develop targeted medications and treatments.

That is the thinking we are moving toward: Lets not just try to stop the disease; we want to prevent it and cure it. That is the goal.

To ensure that AI in medicine is effective, Brown said that data scientists and clinical researchers must be aware of inherent biases prevalent in smaller data sets that are many times missed when that data is generalized.

Algorithms work in a way that gets it the highest reward, and if it can get that reward through a biased set of data, it will use it, Brown said. You have to make sure that you are giving [the algorithm] a fair cross section of data so that it can give you results that are truly answering the question that you are after.

For Syed, one of the biggest issues lies in the lack of publicly available data sets. When determining patterns in celiac patients, Syed and her team parsed through 2,000 to 3,000 electronic health records to collect data from the desired population. Instead of recreating the wheel by labeling hundreds of thousands of biopsies, Syed hopes to gather medical information through large industry partners such as Takeda, a R&D-driven global biopharmaceutical company.

A lot of this cross-thinking happens when there is open access to data, but that has to be cleaned and sorted and thoughtfully put out there, Syed said.

Once data is collected and confirmed to be bias-free, the final step is ensuring interpretability. When many clinicians do not have a background in data science, the only way to cross the disciplinary boundary is to ensure that the data is communicated effectively, which is a responsibility that falls back on the data science community.

At the heart of this is making it interpretable to the clinicians so that they get what it is that the system is telling them and how it can be used, and what worth it is, Brown said. And then, they can look at individual patients and decide what makes sense.

Machine learning algorithms are being used across medical disciplines throughout the world of health. Within UVA Grounds, algorithms are being used to predict the effects of cardiac disease treatments, understand health conditions through smart watches in UVAs Link Lab, and analyze real-time diabetes-monitoring data, to name a few. As Syed, Brown and other researchers continue to evolve their AI capabilities, clinicians will begin to transform the ways in which they can accurately predict, determine and treat disease.

The more I have learned about [big data], the more I have understood its potential impact in medicine across all specialties, Syed said.

Read more:

Medicine Meets Big Data: Clinicians Look to AI for Disease Prediction, Prevention - University of Virginia