Connect with us

The Conversation

AI misunderstands some people’s words more than others

Published

on

theconversation.com – Roberto Rey Agudo, Research Assistant Professor of Spanish and Portuguese, Dartmouth College – 2025-01-27 07:50:00

‘Sorry, I didn’t get that’: AI misunderstands some people’s words more than others

Speech recognition systems are less accurate for women and Black people, among other demographics.
Jacob Wackerhausen/iStock via Getty Images

Roberto Rey Agudo, Dartmouth College

The idea of a humanlike artificial intelligence assistant that you can speak with has been alive in many people’s imaginations since the release of “Her,” Spike Jonze’s 2013 film about a man who falls in love with a Siri-like AI named Samantha. Over the course of the film, the protagonist grapples with the ways in which Samantha, real as she may seem, is not and never will be human.

Twelve years on, this is no longer the stuff of science fiction. Generative AI tools like ChatGPT and digital assistants like Apple’s Siri and Amazon’s Alexa help people get driving directions, make grocery lists, and plenty else. But just like Samantha, automatic speech recognition systems still cannot do everything that a human listener can.

You have probably had the frustrating experience of calling your bank or utility company and needing to repeat yourself so that the digital customer service bot on the other line can understand you. Maybe you’ve dictated a note on your phone, only to spend time editing garbled words.

Linguistics and computer science researchers have shown that these systems work worse for some people than for others. They tend to make more errors if you have a non-native or a regional accent, are Black, speak in African American Vernacular English, code-switch, if you are a woman, are old, are too young or have a speech impediment.

Tin ear

Unlike you or me, automatic speech recognition systems are not what researchers call “sympathetic listeners.” Instead of trying to understand you by taking in other useful clues like intonation or facial gestures, they simply give up. Or they take a probabilistic guess, a move that can sometimes result in an error.

As companies and public agencies increasingly adopt automatic speech recognition tools in order to cut costs, people have little choice but to interact with them. But the more that these systems come into use in critical fields, ranging from emergency first responders and health care to education and law enforcement, the more likely there will be grave consequences when they fail to recognize what people say.

Imagine sometime in the near future you’ve been hurt in a car crash. You dial 911 to call for help, but instead of being connected to a human dispatcher, you get a bot that’s designed to weed out nonemergency calls. It takes you several rounds to be understood, wasting time and raising your anxiety level at the worst moment.

What causes this kind of error to occur? Some of the inequalities that result from these systems are baked into the reams of linguistic data that developers use to build large language models. Developers train artificial intelligence systems to understand and mimic human language by feeding them vast quantities of text and audio files containing real human speech. But whose speech are they feeding them?

If a system scores high accuracy rates when speaking with affluent white Americans in their mid-30s, it is reasonable to guess that it was trained using plenty of audio recordings of people who fit this profile.

With rigorous data collection from a diverse range of sources, AI developers could reduce these errors. But to build AI systems that can understand the infinite variations in human speech arising from things like gender, age, race, first vs. second language, socioeconomic status, ability and plenty else, requires significant resources and time.

‘Proper’ English

For people who do not speak English – which is to say, most people around the world – the challenges are even greater. Most of the world’s largest generative AI systems were built in English, and they work far better in English than in any other language. On paper, AI has lots of civic potential for translation and increasing people’s access to information in different languages, but for now, most languages have a smaller digital footprint, making it difficult for them to power large language models.

Even within languages well-served by large language models, like English and Spanish, your experience varies depending on which dialect of the language you speak.

Right now, most speech recognition systems and generative AI chatbots reflect the linguistic biases of the datasets they are trained on. They echo prescriptive, sometimes prejudiced notions of “correctness” in speech.

In fact, AI has been proved to “flatten” linguistic diversity. There are now AI startup companies that offer to erase the accents of their users, drawing on the assumption that their primary clientele would be customer service providers with call centers in foreign countries like India or the Philippines. The offering perpetuates the notion that some accents are less valid than others.

Human connection

AI will presumably get better at processing language, accounting for variables like accents, code-switching and the like. In the U.S., public services are obligated under federal law to guarantee equitable access to services regardless of what language a person speaks. But it is not clear whether that alone will be enough incentive for the tech industry to move toward eliminating linguistic inequities.

Many people might prefer to talk to a real person when asking questions about a bill or medical issue, or at least to have the ability to opt out of interacting with automated systems when seeking key services. That is not to say that miscommunication never happens in interpersonal communication, but when you speak to a real person, they are primed to be a sympathetic listener.

With AI, at least for now, it either works or it doesn’t. If the system can process what you say, you are good to go. If it cannot, the onus is on you to make yourself understood.The Conversation

Roberto Rey Agudo, Research Assistant Professor of Spanish and Portuguese, Dartmouth College

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Read More

The post AI misunderstands some people’s words more than others appeared first on theconversation.com

The Conversation

Study shows surge of imagery and fakes can precede international and political violence

Published

on

theconversation.com – Tim Weninger, Collegiate Proessor of Engineering, University of Notre Dame – 2025-04-24 07:59:00

AI tools reveal how images have been manipulated.
William Theisen et al.

Tim Weninger, University of Notre Dame and Ernesto Verdeja, University of Notre Dame

Imagine a country with deep political divisions, where different groups don’t trust each other and violence seems likely. Now, imagine a flood of political images, hateful memes and mocking videos from domestic and foreign sources taking over social media. What is likely to happen next?

The widespread use of social media during times of political trouble and violence has made it harder to prevent conflict and build peace. Social media is changing, with new technologies and strategies available to influence what people think during political crises. These include new ways to promote beliefs and goals, gain support, dehumanize opponents, justify violence and create doubt or dismiss inconvenient facts.

At the same time, the technologies themselves are becoming more sophisticated. More and more, social media campaigns use images such as memes, videos and photos – whether edited or not – that have a bigger impact on people than just text.

It’s harder for AI systems to understand images compared with text. For example, it’s easier to track posts that say “Ukrainians are Nazis” than it is to find and understand fake images showing Ukrainian soldiers with Nazi symbols. But these kinds of images are becoming more common. Just as a picture is worth a thousand words, a meme is worth a thousand tweets.

Our team of computer and social scientists has tackled the challenge of interpreting image content by combining artificial intelligence methods with human subject matter experts to study how visual social media posts change in high-risk situations. Our research shows that these changes in social media posts, especially those with images, serve as strong indicators of coming mass violence.

Surge of memes

Our recent analysis found that in the two weeks leading up to Russia’s 2022 invasion of Ukraine there was a nearly 9,000% increase in the number of posts and a more than 5,000% increase in manipulated images from Russian milbloggers. Milbloggers are bloggers who focus on current military conflicts.

These huge increases show how intense Russia’s online propaganda campaign was and how it used social media to influence people’s opinions and justify the invasion.

This also shows the need to better monitor and analyze visual content on social media. To conduct our analysis, we collected the entire history of posts and images from the accounts of 989 Russian milbloggers on the messaging app Telegram. This includes nearly 6 million posts and over 3 million images. Each post and image was time-stamped and categorized to facilitate detailed analysis.

Media forensics

We had previously developed a suite of AI tools capable of detecting image alterations and manipulations. For instance, one detected image shows a pro-Russian meme mocking anti-Putin journalist and former Russian soldier Arkady Babchenko, whose death was faked by Ukrainian security services to expose an assassination plot against him.

The meme features the language “gamers don’t die, they respawn,” alluding to video game characters who return to life after dying. This makes light of Babchenko’s predicament and illustrates the use of manipulated images to convey political messages and influence public opinion.

This is just one example out of millions of images that were strategically manipulated to promote various narratives. Our statistical analysis revealed a massive increase in both the number of images and the extent of their manipulations prior to the invasion.

Political context is critical

Although these AI systems are very good at finding fakes, they are incapable of understanding the images’ political contexts. It is therefore critical that AI scientists work closely with social scientists in order to properly interpret these findings.

Our AI systems also categorized images by similarity, which then allowed subject experts to further analyze image clusters based on their narrative content and culturally and politically specific meanings. This is impossible to do at a large scale without AI support.

For example, a fake image of French president Emmanuel Macron with Ukrainian governor Vitalii Kim may be meaningless to an AI scientist. But to political scientists the image appears to laud Ukrainians’ outsize courage in contrast to foreign leaders who have appeared to be afraid of Russian nuclear threats. The goal was to reinforce Ukrainian doubts about their European allies.

image of of two men, one seated
This manipulated image combines French president Emmanuel Macron with Ukranian governor Vitalii Kim. It requires the expertise of political scientists to interpret the creator’s pro-Russian meaning.
William Theisen et al.

Meme warfare

The shift to visual media in recent years brings a new type of data that researchers haven’t yet studied much in detail.

Looking at images can help researchers understand how adversaries frame each other and how this can lead to political conflict. By studying visual content, researchers can see how stories and ideas are spread, which helps us understand the psychological and social factors involved.

This is especially important for finding more advanced and subtle ways people are influenced. Projects like this also can contribute to improving early warning efforts and reduce the risks of violence and instability.The Conversation

Tim Weninger, Collegiate Proessor of Engineering, University of Notre Dame and Ernesto Verdeja, Associate Professor of Peace Studies and Global Politics, University of Notre Dame

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Read More

The post Study shows surge of imagery and fakes can precede international and political violence appeared first on theconversation.com

Continue Reading

The Conversation

Colors are objective, according to two philosophers − even though the blue you see doesn’t match what I see

Published

on

theconversation.com – Elay Shech, Professor of Philosophy, Auburn University – 2025-04-25 07:55:00

What appear to be blue and green spirals are actually the same color.
Akiyoshi Kitaoka

Elay Shech, Auburn University and Michael Watkins, Auburn University

Is your green my green? Probably not. What appears as pure green to me will likely look a bit yellowish or blueish to you. This is because visual systems vary from person to person. Moreover, an object’s color may appear differently against different backgrounds or under different lighting.

These facts might naturally lead you to think that colors are subjective. That, unlike features such as length and temperature, colors are not objective features. Either nothing has a true color, or colors are relative to observers and their viewing conditions.

But perceptual variation has misled you. We are philosophers who study colors, objectivity and science, and we argue in our book “The Metaphysics of Colors” that colors are as objective as length and temperature.

Perceptual variation

There is a surprising amount of variation in how people perceive the world. If you offer a group of people a spectrum of color chips ranging from chartreuse to purple and asked them to pick the unique green chip – the chip with no yellow or blue in it – their choices would vary considerably. Indeed, there wouldn’t be a single chip that most observers would agree is unique green.

Generally, an object’s background can result in dramatic changes in how you perceive its colors. If you place a gray object against a lighter background, it will appear darker than if you place it against a darker background. This variation in perception is perhaps most striking when viewing an object under different lighting, where a red apple could look green or blue.

Of course, that you experience something differently does not prove that what is experienced is not objective. Water that feels cold to one person may not feel cold to another. And although we do not know who is feeling the water “correctly,” or whether that question even makes sense, we can know the temperature of the water and presume that this temperature is independent of your experience.

Similarly, that you can change the appearance of something’s color is not the same as changing its color. You can make an apple look green or blue, but that is not evidence that the apple is not red.

Apple under a gradient of red to blue light
Under different lighting conditions, objects take on different colors.
Gyozo Vaczi/iStock via Getty Images Plus

For comparison, the Moon appears larger when it’s on the horizon than when it appears near its zenith. But the size of the Moon has not changed, only its appearance. Hence, that the appearance of an object’s color or size varies is, by itself, no reason to think that its color and size are not objective features of the object. In other words, the properties of an object are independent of how they appear to you.

That said, given that there is so much variation in how objects appear, how do you determine what color something actually is? Is there a way to determine the color of something despite the many different experiences you might have of it?

Matching colors

Perhaps determining the color of something is to determine whether it is red or blue. But we suggest a different approach. Notice that squares that appear to be the same shade of pink against different backgrounds look different against the same background.

Green, purple and orange squares with smaller squares in shades of pink placed at their centers and at the bottom of the image
The smaller squares may appear to be the same color, but if you compare them with the strip of squares at the bottom, they’re actually different shades.
Shobdohin/Wikimedia Commons, CC BY-SA

It’s easy to assume that to prove colors are objective would require knowing which observers, lighting conditions and backgrounds are the best, or “normal.” But determining the right observers and viewing conditions is not required for determining the very specific color of an object, regardless of its name. And it is not required to determine whether two objects have the same color.

To determine whether two objects have the same color, an observer would need to view the objects side by side against the same background and under various lighting conditions. If you painted part of a room and find that you don’t have enough paint, for instance, finding a match might be very tricky. A color match requires that no observer under any lighting condition will see a difference between the new paint and the old.

YouTube video
Is the dress yellow and white or black and blue?

That two people can determine whether two objects have the same color even if they don’t agree on exactly what that color is – just as a pool of water can have a particular temperature without feeling the same to me and you – seems like compelling evidence to us that colors are objective features of our world.

Colors, science and indispensability

Everyday interactions with colors – such as matching paint samples, determining whether your shirt and pants clash, and even your ability to interpret works of art – are hard to explain if colors are not objective features of objects. But if you turn to science and look at the many ways that researchers think about colors, it becomes harder still.

For example, in the field of color science, scientific laws are used to explain how objects and light affect perception and the colors of other objects. Such laws, for instance, predict what happens when you mix colored pigments, when you view contrasting colors simultaneously or successively, and when you look at colored objects in various lighting conditions.

The philosophers Hilary Putnam and Willard van Orman Quine made famous what is known as the indispensability argument. The basic idea is that if something is indispensable to science, then it must be real and objective – otherwise, science wouldn’t work as well as it does.

For example, you may wonder whether unobservable entities such as electrons and electromagnetic fields really exist. But, so the argument goes, the best scientific explanations assume the existence of such entities and so they must exist. Similarly, because mathematics is indispensable to contemporary science, some philosophers argue that this means mathematical objects are objective and exist independently of a person’s mind.

Blue damselfish, seeming iridescent against a black background
The color of an animal can exert evolutionary pressure.
Paul Starosta/Stone via Getty Images

Likewise, we suggest that color plays an indispensable role in evolutionary biology. For example, researchers have argued that aposematism – the use of colors to signal a warning for predators – also benefits an animal’s ability to gather resources. Here, an animal’s coloration works directly to expand its food-gathering niche insofar as it informs potential predators that the animal is poisonous or venomous.

In fact, animals can exploit the fact that the same color pattern can be perceived differently by different perceivers. For instance, some damselfish have ultraviolet face patterns that help them be recognized by other members of their species and communicate with potential mates while remaining largely hidden to predators unable to perceive ultraviolet colors.

In sum, our ability to determine whether objects are colored the same or differently and the indispensable roles they play in science suggest that colors are as real and objective as length and temperature.The Conversation

Elay Shech, Professor of Philosophy, Auburn University and Michael Watkins, Professor of Philosophy, Auburn University

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Read More

The post Colors are objective, according to two philosophers − even though the blue you see doesn’t match what I see appeared first on theconversation.com

Continue Reading

The Conversation

‘Extraordinary claims require extraordinary evidence’ − an astronomer explains how much evidence scientists need to claim discoveries like extraterrestrial life

Published

on

theconversation.com – Chris Impey, University Distinguished Professor of Astronomy, University of Arizona – 2025-04-25 07:54:00

The universe is filled with countless galaxies, stars and planets. Astronomers may find life one day, but they will need extraordinary proof.
ESA/Euclid/Euclid Consortium/NASA, image processing by J.-C. Cuillandre (CEA Paris-Saclay), G. Anselmi

Chris Impey, University of Arizona

The detection of life beyond Earth would be one of the most profound discoveries in the history of science. The Milky Way galaxy alone hosts hundreds of millions of potentially habitable planets. Astronomers are using powerful space telescopes to look for molecular indicators of biology in the atmospheres of the most Earth-like of these planets.

But so far, no solid evidence of life has ever been found beyond the Earth. A paper published in April 2025 claimed to detect a signature of life in the atmosphere of the planet K2-18b. And while this discovery is intriguing, most astronomers – including the paper’s authors – aren’t ready to claim that it means extraterrestrial life exists. A detection of life would be a remarkable development.

The astronomer Carl Sagan used the phrase, “Extraordinary claims require extraordinary evidence,” in regard to searching for alien life. It conveys the idea that there should be a high bar for evidence to support a remarkable claim.

I’m an astronomer who has written a book about astrobiology. Over my career, I’ve seen some compelling scientific discoveries. But to reach this threshold of finding life beyond Earth, a result needs to fit several important criteria.

When is a result important and reliable?

There are three criteria for a scientific result to represent a true discovery and not be subject to uncertainty and doubt. How does the claim of life on K2-18b measure up?

First, the experiment needs to measure a meaningful and important quantity. Researchers observed K2-18b’s atmosphere with the James Webb Space Telescope and saw a spectral feature that they identified as dimethyl sulfide.

On Earth, dimethyl sulfide is associated with biology, in particular bacteria and plankton in the oceans. However, it can also arise by other means, so this single molecule is not conclusive proof of life.

Second, the detection needs to be strong. Every detector has some noise from the random motion of electrons. The signal should be strong enough to have a low probability of arising by chance from this noise.

The K2-18b detection has a significance of 3-sigma, which means it has a 0.3% probability of arising by chance.

That sounds low, but most scientists would consider that a weak detection. There are many molecules that could create a feature in the same spectral range.

The “gold standard” for scientific detection is 5-sigma, which means the probability of the finding happening by chance is less than 0.00006%. For example, physicists at CERN gathered data patiently for two years until they had a 5-sigma detection of the Higgs boson particle, leading to a Nobel Prize one year later in 2013.

YouTube video
The announcement of the discovery of the Higgs boson took decades from the time Peter Higgs first predicted the existence of the particle. Scientists, such as Joe Incandela shown here, waited until they’d reached that 5-sigma level to say, ‘I think we have it.’

Third, a result needs to be repeatable. Results are considered reliable when they’ve been repeated – ideally corroborated by other investigators or confirmed using a different instrument. For K2-18b, this might mean detecting other molecules that indicate biology, such as oxygen in the planet’s atmosphere. Without more and better data, most researchers are viewing the claim of life on K2-18b with skepticism.

Claims of life on Mars

In the past, some scientists have claimed to have found life much closer to home, on the planet Mars.

Over a century ago, retired Boston merchant turned astronomer Percival Lowell claimed that linear features he saw on the surface of Mars were canals, constructed by a dying civilization to transport water from the poles to the equator. Artificial waterways on Mars would certainly have been a major discovery, but this example failed the other two criteria: strong evidence and repeatability.

Lowell was misled by his visual observations, and he was engaging in wishful thinking. No other astronomers could confirm his findings.

An image of Mars in space
Mars, as taken by the OSIRIS instrument on the ESA Rosetta spacecraft during its February 2007 flyby of the planet and adjusted to show color.
ESA & MPS for OSIRIS Team MPS/UPD/LAM/IAA/RSSD/INTA/UPM/DASP/IDA, CC BY-SA

In 1996, NASA held a press conference where a team of scientists presented evidence for biology in the Martian meteorite ALH 84001. Their evidence included an evocative image that seemed to show microfossils in the meteorite.

However, scientists have come up with explanations for the meteorite’s unusual features that do not involve biology. That extraordinary claim has dissipated.

More recently, astronomers detected low levels of methane in the atmosphere of Mars. Like dimethyl sulfide and oxygen, methane on Earth is made primarily – but not exclusively – by life. Different spacecraft and rovers on the Martian surface have returned conflicting results, where a detection with one spacecraft was not confirmed by another.

The low level and variability of methane on Mars is still a mystery. And in the absence of definitive evidence that this very low level of methane has a biological origin, nobody is claiming definitive evidence of life on Mars.

Claims of advanced civilizations

Detecting microbial life on Mars or an exoplanet would be dramatic, but the discovery of extraterrestrial civilizations would be truly spectacular.

The search for extraterrestrial intelligence, or SETI, has been underway for 75 years. No messages have ever been received, but in 1977 a radio telescope in Ohio detected a strong signal that lasted only for a minute.

This signal was so unusual that an astronomer working at the telescope wrote “Wow!” on the printout, giving the signal its name. Unfortunately, nothing like it has since been detected from that region of the sky, so the Wow! Signal fails the test of repeatability.

An illustration of a long, thin rock flying through space.
‘Oumuamua is the first object passing through the solar system that astronomers have identified as having interstellar origins.
European Southern Observatory/M. Kornmesser

In 2017, a rocky, cigar-shaped object called ‘Oumuamua was the first known interstellar object to visit the solar system. ‘Oumuamua’s strange shape and trajectory led Harvard astronomer Avi Loeb to argue that it was an alien artifact. However, the object has already left the solar system, so there’s no chance for astronomers to observe it again. And some researchers have gathered evidence suggesting that it’s just a comet.

While many scientists think we aren’t alone, given the enormous amount of habitable real estate beyond Earth, no detection has cleared the threshold enunciated by Carl Sagan.

Claims about the universe

These same criteria apply to research about the entire universe. One particular concern in cosmology is the fact that, unlike the case of planets, there is only one universe to study.

A cautionary tale comes from attempts to show that the universe went through a period of extremely rapid expansion a fraction of a second after the Big Bang. Cosmologists call this event inflation, and it is invoked to explain why the universe is now smooth and flat.

In 2014, astronomers claimed to have found evidence for inflation in a subtle signal from microwaves left over after the Big Bang. Within a year, however, the team retracted the result because the signal had a mundane explanation: They had confused dust in our galaxy with a signature of inflation.

On the other hand, the discovery of the universe’s acceleration shows the success of the scientific method. In 1929, astronomer Edwin Hubble found that the universe was expanding. Then, in 1998, evidence emerged that this cosmic expansion is accelerating. Physicists were startled by this result.

Two research groups used supernovae to separately trace the expansion. In a friendly rivalry, they used different sets of supernovae but got the same result. Independent corroboration increased their confidence that the universe was accelerating. They called the force behind this accelerating expansion dark energy and received a Nobel Prize in 2011 for its discovery.

On scales large and small, astronomers try to set a high bar of evidence before claiming a discovery.The Conversation

Chris Impey, University Distinguished Professor of Astronomy, University of Arizona

This article is republished from The Conversation under a Creative Commons license. Read the original article.

Read More

The post ‘Extraordinary claims require extraordinary evidence’ − an astronomer explains how much evidence scientists need to claim discoveries like extraterrestrial life appeared first on theconversation.com

Continue Reading

Trending