Connect with us

Published

4 months ago

on

This story originally appeared in The Algorithm, our weekly newsletter on AI. To get stories like this in your inbox first, sign up here.

Think of a teacher. Close your eyes. What does that person look like? If you ask Stable Diffusion or DALL-E 2, two of the most popular AI image generators, it’s a white man with glasses. 

Last week, I published a story about new tools developed by researchers at AI startup Hugging Face and the University of Leipzig that let people see for themselves what kinds of inherent biases AI models have about different genders and ethnicities.

Although I’ve written a lot about how our biases are reflected in AI models, it still felt jarring to see exactly how pale, male, and stale the humans of AI are. That was particularly true for DALL-E 2, which generates white men 97% of the time when given prompts like “CEO” or “director.”

And the bias problem runs even deeper than you might think into the broader world created by AI. These models are built by American companies and trained on North American data, and thus when they’re asked to generate even mundane everyday items, from doors to houses, they create objects that look American, Federico Bianchi, a researcher at Stanford University, tells me.

As the world becomes increasingly filled with AI-generated imagery, we are going to mostly see images that reflect America’s biases, culture, and values. Who knew AI could end up being a major instrument of American soft power?
So how do we address these problems? A lot of work has gone into fixing biases in the data sets AI models are trained on. But two recent research papers propose interesting new approaches.

What if, instead of making the training data less biased, you could simply ask the model to give you less biased answers?

A team of researchers at the Technical University of Darmstadt, Germany, and AI startup Hugging Face developed a tool called Fair Diffusion that makes it easier to tweak AI models to generate the types of images you want. For example, you can generate stock photos of CEOs in different settings and then use Fair Diffusion to swap out the white men in the images for women or people of different ethnicities.

As the Hugging Face tools show, AI models that generate images on the basis of image-text pairs in their training data default to very strong biases about professions, gender, and ethnicity. The German researchers’ Fair Diffusion tool is based on a technique they developed called semantic guidance, which allows users to guide how the AI system generates images of people and edit the results.

The AI system stays very close to the original image, says Kristian Kersting, a computer science professor at TU Darmstadt who participated in the work. 

image 3 1

This method lets people create the images they want without having to undertake the cumbersome and time-consuming task of trying to improve the biased data set that was used to train the AI model, says Felix Friedrich, a PhD student at TU Darmstadt who worked on the tool.

However, the tool is not perfect. Changing the images for some occupations, such as “dishwasher,” didn’t work as well because the word means both a machine and a job. The tool also only works with two genders. And ultimately, the diversity of the people the model can generate is still limited by the images in the AI system’s training set. Still, while more research is needed, this tool could be an important step in mitigating biases.

A similar technique also seems to work for language models. Research from the AI lab Anthropic shows how simple instructions can steer large language models to produce less toxic content, as my colleague Niall Firth reported recently. The Anthropic team tested different language models of varying sizes and found that if the models are large enough, they self-correct for some biases after simply being asked to.

Researchers don’t know why text- and image-generating AI models do this. The Anthropic team thinks it might be because larger models have larger training data sets, which include lots of examples of biased or stereotypical behavior—but also examples of people pushing back against this biased behavior.

AI tools are becoming increasingly popular for generating stock images. Tools like Fair Diffusion could be useful for companies that want their promotional pictures to reflect society’s diversity,

Read More

————

By: Melissa Heikkilä
Title: What if we could just ask AI to be less biased?
Sourced From: www.technologyreview.com/2023/03/28/1070390/what-if-we-could-just-ask-ai-to-be-less-biased/
Published Date: Tue, 28 Mar 2023 08:22:40 +0000

Total
0
Shares
Share 0
Tweet 0
Pin it 0
Share 0
Related Topics:
Continue Reading

Tech

These scientists live like astronauts without leaving Earth

Published

10 hours ago

on

September 22, 2023

By

201009 B2 exterior 0005 scaled

This article was originally published on Undark. Read the original article.

In January 2023, Tara Sweeney’s plane landed on Thwaites Glacier, a 74,000-square-mile mass of frozen water in West Antarctica. She arrived with an international research team to study the glacier’s geology and ice fabric, and how its ice melt might contribute to sea level rise. But while near Earth’s southernmost point, Sweeney kept thinking about the moon.

“It felt every bit of what I think it will feel like being a space explorer,” said Sweeney, a former Air Force officer who’s now working on a doctorate in lunar geology at the University of Texas at El Paso. “You have all of these resources, and you get to be the one to go out and do the exploring and do the science. And that was really spectacular.”

That similarity is why space scientists study the physiology and psychology of people living in Antarctic and other remote outposts: For around 25 years, people have played out what existence might be like on, or en route to, another world. Polar explorers are, in a way, analogous to astronauts who land on alien planets. And while Sweeney wasn’t technically on an “analog astronaut” mission — her primary objective being the geological exploration of Earth — her days played out much the same as a space explorer’s might.

For 16 days, Sweeney and her colleagues lived in tents on the ice, spending half their time trapped inside as storms blew snow against their tents. When the weather permitted, Sweeney snowmobiled to and from seismometer sites, once getting caught in a whiteout that, she said, felt like zooming inside a ping-pong ball.

On the glacier, Sweeney was always cold, sometimes bored, often frustrated. But she was also alive, elated. And she felt a form of focus that eluded her on her home continent. “I had three objectives: to be a good crewmate, to do good science, and to stay alive,” she said. “That’s all I had to do.”

None of that was easy, of course. But it may have been easier than landing back on the earth of El Paso. “My mission ended, and it’s over,” she said. “And how do I process through all these things that I’m feeling?”

Then, in May, she attended the 2023 Analog Astronaut Conference, a gathering of people who simulate long-term space travel from the relative safety and comfort of Earth. Sweeney had learned about the event when she visited an analog facility in the country of Jordan. There, she’d met one of the conference’s founders, Jas Purewal, who invited her to the gathering.

The meeting was held, appropriately, at Biosphere 2, a glass-paneled, self-contained habitat in the Arizona desert that resembles a 1980s sci-fi vision of a space settlement — one of the first facilities built, in part, to understand whether humans could create a habitable environment on a hostile planet.

wide view of the Biosphere 2 facilities
The 40-acre Biosphere 2 campus in Oracle, Arizona. It was one of the first facilities built for analog astronaut missions. UNIVERSITY OF ARIZONA

A speaker at the conference had spent eight months locked inside a simulated space habitat in Moscow, Russia, and she talked about how the post-mission period had been hard for her. The psychological toll of reintegration became a chattering theme throughout the whole meeting. Sweeney, it turned out, wasn’t alone.

Across the world, around 20 analog space facilities host people who volunteer to be study subjects, isolating themselves for weeks or months in polar stations, desert outposts, or even sealed habitats inside NASA centers. These places are intended to mimic how people might fare on Mars or the moon, or on long-term orbital stations. Such research, scientists say, can help test out medical and software tools, enhance indoor agriculture, and address the difficulties analog astronauts face, including, like Sweeney’s, those that come when their “missions” are over.

Lately, a community of researchers has started to make the field more formalized: laying out standards so that results are comparable; gathering research papers into a single database so investigators can

Read More

————

By: Sarah Scoles
Title: These scientists live like astronauts without leaving Earth
Sourced From: www.technologyreview.com/2023/09/22/1080035/scientists-live-like-astronauts-without-leaving-earth/
Published Date: Fri, 22 Sep 2023 15:38:48 +0000

Did you miss our previous article…
https://mansbrand.com/making-sense-of-sensor-data/

Continue Reading

Tech

Making sense of sensor data

Published

12 hours ago

on

September 22, 2023

By

Kinetica web ready 16 9 cover

Consider a supply chain where delivery vehicles, shipping containers, and individual products are sensor-equipped. Real-time insights enable workers to optimize routes, reduce delays, and efficiently manage inventory. This smart orchestration boosts efficiency, minimizes waste, and lowers costs.

Many industries are rapidly integrating sensors, creating vast data streams that can be leveraged to open profound business possibilities. In energy management, growing use of sensors and drone footage promises to enable efficient energy distribution, lower costs, and reduced environmental impact. In smart cities, sensor networks can enhance urban life by monitoring traffic flow, energy consumption, safety concerns, and waste management.

Kinetica web ready 16 9 cover 1

These aren’t glimpses of a distant future, but realities made possible today by the increasingly digitally instrumented world. Internet of Things (IoT) sensors have been rapidly integrated across industries, and now constantly track and measure properties like temperature, pressure, humidity, motion, light levels, signal strength, speed, weather events, inventory, heart rate and traffic.

The information these devices collect—sensor and machine data—provides insight into the real-time status and trends of these physical parameters. This data can then be used to make informed decisions and take action—capabilities that unlock transformative business opportunities, from streamlined supply chains to futuristic smart cities.

John Rydning, research vice president at IDC, projects that sensor and machine data volumes will soar over the next five years, achieving a greater than 40% compound annual growth rate through 2027. He attributes that not primarily to an increasing number of devices, as IoT devices are already quite prevalent, but rather due to more data being generated by each one as businesses learn to make use of their ability to produce real-time streaming data.

V2 Kinetica web ready 16 9 quote 1

Meanwhile, sensors are growing more interconnected and sophisticated, while the data they generate increasingly includes a location in addition to a timestamp. These spatial and temporal features not only capture data changes over time, but also create intricate maps of how these shifts unfold across locations—facilitating more comprehensive insights and predictions.

But as sensor data grows more complex and voluminous, legacy data infrastructure struggles to keep pace. Continuous readings over time and space captured by sensor devices now require a new set of design patterns to unlock maximum value. While businesses have capitalized on spatial and time-series data independently for over a decade, its true potential is only realized when considered in tandem, in context, and with the capacity for real-time insights.

Download the report.

This content was produced by Insights, the custom content arm of MIT Technology Review. It was not written by MIT Technology Review’s editorial staff.

Read More

————

By: MIT Technology Review Insights
Title: Making sense of sensor data
Sourced From: www.technologyreview.com/2023/09/21/1079924/making-sense-of-sensor-data/
Published Date: Thu, 21 Sep 2023 17:00:00 +0000

Continue Reading

Tech

The Download: inverse vaccines, and Microsoft’s big deal

Published

14 hours ago

on

September 22, 2023

By

This is today’s edition of The Download our weekday newsletter that provides a daily dose of what’s going on in the world of technology.

How inverse vaccines might tackle diseases like multiple sclerosis

On the whole, typical vaccines prime the immune system to respond. But scientists are also working on “inverse vaccines” that teach the immune system to stand down.

Last week Jeffrey Hubbell and his colleagues at the University of Chicago reported that an inverse vaccine they developed had successfully reversed a disease similar to multiple sclerosis in mice. Hubbell has tested this approach before, but only as a way of preventing the disease—not curing it.

These immune-dampening shots could lead to a whole host of therapies to treat autoimmune diseases. And in fact, Anokion, a company Hubbell cofounded, has already launched clinical trials to test whether this type of inverse vaccine might help people with multiple sclerosis and celiac disease. Read the full story.

—Cassandra Willyard

This story is from The Checkup, our weekly biotech and health newsletter. Sign up to receive it in your inbox every Thursday.

To learn more about vaccines, why not check out:

+ What to know about this autumn’s covid vaccines. New variants will pose a challenge, but early signs suggest the shots will still boost antibody responses. Read the full story.

+ Who benefits most from the new covid vaccines? Data show that older adults and people with underlying illnesses need the vaccine most. Read the full story.

+ What’s next for mRNA vaccines. mRNA vaccines helped us through the covid-19 pandemic—but they could also help defend against many other infectious diseases, offer universal protection against flu, and even treat cancer. Read the full story.

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 The UK is revisiting Microsoft’s offer to acquire Activision
But it’s not a done deal yet. (WSJ $)
UK regulators have accepted Microsoft’s tweaks to the deal, in principle. (FT $)

2 The US has granted Ukraine a lot more military hardware
But long-range missiles, which Ukraine covets, won’t be included in the package. (ABC News)
The US equipment will plug the gap now Poland has ceased sending weapons. (Economist $)+ Decoy weapons are successfully fooling Russian troops. (FT $)
Inside the messy ethics of making war with machines. (MIT Technology Review)

3 The blockchain’s future looks surprisingly crypto-free
Startups have more modest goals now, ideally without legal complications. (Rest of World)+ Crypto parties are still raging if you know where to look, though. (Bloomberg $)

4 These law geeks are taking us inside Google’s antitrust trial
Google may have stopped the trial from being live streamed, but it can’t prevent members of the public from sitting in. (Wired $)

5 People are still queuing to buy the new iPhone
Lines of eager customers in China, the UK, Dubai and Australia suggest Apple’s appeal is as strong as ever. (Bloomberg $)

6 A man-made organism defies the rules of biology
In theory, using unnatural amino acids could make organisms less prone to viral infections. (Proto.Life)

7 Your intimate health information is just data to Big Tech
Data that can generally be sold onto willing advertisers. (The Atlantic $)
How your brain data could be used against you. (MIT Technology Review)

8 Gen Z isn’t immune to online scams
In fact, in many cases, they fall for them more than boomers. (Vox)

9 Absolutely nobody loves PowerPoint
Nonetheless, it persists. (FT $)
Next slide, please: A brief history of the corporate presentation. (MIT Technology Review)

10 Don’t throw out your used coffee grounds
They can become the foundation for all sorts of 3D-printing projects instead. (Ars Technica)
Watch this team of drones 3D-print a tower. (MIT Technology Review)

Quote of the day

“It feels bad, you feel hurt. Then you give yourself time to grieve, you find someone else and you get excited again.”

—Rowan Rosenthal, a former principal product designer for Grindr, likens the decision to leave the company because of its strict return-to-office mandate to the end of a relationship, they tell the Washington Post.

The big story

California’s coming offshore wind boom faces big engineering hurdles

7pOBQSa 1 g6Wmm0vaaGfCKh agnOh5X8fvo0szO5tQj76IoJnZj t wPj1lCwX1A2cpHFhzI3Nguwd51VMtbMqRxFWKc4NZx8lRc1SmxplS W8cyjmrKecjVXWAmQA7NDWoz6 Y TRdCG3zXQ5vCZo

December 2022

Last December, dozens of companies fought for the right to lease the first commercial wind power sites off the coast of California in an auction

Read More

————

By: Rhiannon Williams
Title: The Download: inverse vaccines, and Microsoft’s big deal
Sourced From: www.technologyreview.com/2023/09/22/1080043/the-download-inverse-vaccines-and-microsofts-big-deal/
Published Date: Fri, 22 Sep 2023 12:10:00 +0000

Did you miss our previous article…
https://mansbrand.com/bolstering-enterprise-llms-with-machine-learning-operations-foundations/

Continue Reading

Trending