1 Billion Meals Go Uneaten Every Day, Helping to Fuel Global Warming: Clean Your Plate to Reduce Food Loss and Waste

By United Nations

Food waste is helping to fuel global warming, one half-eaten sandwich at a time.

In fact, throwing out lunchtime leftovers or wilted lettuce only adds to the mountain of discarded food decomposing in landfills that produces the powerful greenhouse gas methane, accounting for some eight to 10 per cent of global emissions.

This includes spoiled crops – food that is lost after harvest and before even reaching stores.

Image: Zoshua Colah - Unsplash

A billion meals wasted

Every day, over a billion meals are left uneaten, whether due to excess food production, spoilage or poor meal planning, according to the UN Environment Programme (UNEP).

The issue is in the spotlight on the International Day of Awareness of Food Loss and Waste, observed on 29 September – convened by UNEP and the UN Food and Agriculture Organization (FAO).

This year, focus is on the link with protecting forests and ensuring sustainable agrifood systems.

“Using the food we already produce more efficiently can reduce pressure on land and forests and help feed more people,” said Michael Muratha, a Programme Manager at UNEP, speaking from its headquarters in Nairobi.

Moreover, having healthy forests also support food production in several ways, including by protecting soils and regulating rainfall.

Cold comfort

UNEP is highlighting how taking action can reduce losses by promoting measures such as investing in better food handling, storage and transport.

Mr. Muratha stressed that strengthening cold chain systems for perishable foods is also important.

“In Kenya, a cooperative of 179 farmers near Nairobi was losing up to 40 per cent of its crops after harvest before receiving cold-storage and refrigerated transport equipment,” he said.

Better storage can also help farmers to cope with climate impacts, such as droughts, floods and periods of food scarcity, while reducing losses of nutritious foods in a world where 2.6 billion people cannot afford to eat a healthy diet rich in fruits and vegetables.

“That’s why it’s really important to take these measures to preserve and conserve and reduce losses of food,” he said.

UNEP is also backing efforts to redirect safe surplus food to people who need it, such as children at daycare centres in underserved communities.

Innovation at all levels

The agency is further showcasing inventive ways – from simple to cutting edge - to keep food out of landfills.

“Innovation can help at different points where food is lost or wasted, and it does not always have to be high-tech,” Mr. Muratha insisted.

He pointed to the example of Favela Orgânica, a project in Brazil that teaches communities, street vendors and chefs to use ingredients that are often thrown away such as banana peels and broccoli stalks.

“AI is also showing promise,” he added. “In a UNEP-supported initiative involving 13 hotels, AI tools helped measure discarded food and adjust purchasing and preparation, reducing post-consumer breakfast waste by 62 per cent in four months.”

Last year, UNEP and partners launched the Food Waste Breakthrough initiative at the UN COP30 climate change conference in Belém Brazil with the goal of halving food waste by 2030, reducing methane emissions and supporting climate action.

Mr. Muratha said taking action could help reduce some of the more than $1 trillion lost globally each year through food loss and waste, while every dollar invested in prevention saves cities $8 and households $84.

“For communities, it means more food reaching people and greater resilience when harvests, prices or supply chains are disrupted,” he said.

“It will not end hunger on its own, but it is an important part of building more secure and sustainable food systems.”

Fact-Checked by Irfan Ahmad.

Editor’s Note: Published with updated links and minor clarity edits, including a correction to a COP reference that appeared as a typo in the original.

Read next: 

• Study Finds Health Claims Can Mask Nutritional Trade-Offs in Food Products

• Is it safe to charge your phone overnight?
by External Contributor via Digital Information World

How Press Restrictions, Direct-to-Audience Political Messaging and Independent Media Could Affect Journalism and Accountability, University of Michigan Professor Joyojeet Pal Examines

By University of Michigan

Image: Michael Fousert - Unsplash

As the dispute over White House press access moves through the courts, Joyojeet Pal, professor at the University of Michigan School of Information, examines what restrictions on journalists, direct-to-audience political messaging, and the rise of independent media could mean for accountability.

Are the United States and countries such as India beginning to converge in the way political leaders manage the press and communicate with the public?

Whether this was pioneered in the Global South is a much more complicated question. Authoritarian regimes were doing versions of this long before the Cold War. But there are elements that you could argue are being picked up from Global South leaders. One is moving away from established institutions like a standard press conference, for instance, where there is an exchange between political leadership and the press, and replacing them with a more selective format.

When you align that with social media, you can create the appearance that an administration is constantly communicating with people. But the communication is actually one way. It looks like two-way communication because it is happening on an interactive platform, but the leadership still controls what is being communicated and when. That is something you could argue has been picked up from Global South leaders, and I would say Narendra Modi (prime minister of India) has been one of the most important political figures in that space.

What happens when citizens become accustomed to leaders communicating directly rather than regularly answering questions from journalists?

Social media completely changes that ecology. You get what I would call a normative naturalization of this behavior. People gradually begin to accept that this is simply how government communicates: The government sends something out, and citizens receive it. Historically, governments have communicated through proclamations or missives without any expectation that leaders would answer questions in return. Social media can recreate some aspects of that relationship while making it appear much more participatory.

The other important issue is institutional. The legal and political infrastructure that gets created to restrict access or exclude journalists can outlive the particular political moment that was used to justify it. That creates a downstream problem. Imagine a situation in which the country genuinely needs a trusted, unified source of information. COVID is a good example. During a national emergency, people need institutions they consider reliable. If you have spent years damaging the credibility of those institutions, you cannot suddenly restore that credibility because you need it during a crisis. Trust is not something that can simply be switched back on.

Governments can also seek the rapid removal of online content. How does that kind of takedown environment change journalism?

I have mixed feelings about this because, on the face of it, it looks highly controlled: You have a system that is corporatized, technologically mediated and capable of taking material down quickly. But these moves have not necessarily been very effective at eliminating opposition. What opposition media often do instead is move away from established institutional mechanisms and toward more loosely structured channels, including platforms such as YouTube.

It can also change journalistic behavior. If you believe something will be taken down within three hours, for example, you may begin operating “publish first, verify later.” Instead of waiting until everything is completely verified before publishing, someone might put the material online quickly and rely on screenshots and redistribution to keep it circulating. You can then get something like a Streisand effect: The attempt to suppress the material actually makes more people interested in seeing it.

Another consequence is that the center of media production can shift from institutions to individual journalists. We have already seen significant evidence of this in India. Journalists have left established organizations and essentially become small media houses themselves, sometimes one journalist working with a few staff members and operating through their own channels. That gives them considerable independence. But it also means that many of the institutional checks that once existed around an individual journalist may no longer exist. Take someone like Ravish Kumar. A viewer may trust his reporting and his voice. But when the journalist becomes the institution, that individual also becomes the final authority on what gets published.

There is a tradeoff. A journalist may now be able to say things independently that an established organization would not have permitted. At the same time, there are fewer internal checks and balances. And then the question becomes: What happens after repeated takedowns? When does a platform ban someone entirely? Even then, people frequently move to another platform or channel. It can become something of a Hydra situation: you shut down one outlet, and another appears.

Can factual, independent journalism still build a significant audience without government or large corporate backing?

There is already evidence that it can. One important development in India—and something American audiences should watch—is the growth of citizen-funded journalism. Print journalism became increasingly difficult economically because printing is expensive, and newspapers historically depended heavily on advertising. Online publishing changes those economics. If your production costs are relatively low, you do not necessarily need the same advertising structure. There are individual journalists doing this, but there are also independent news organizations that are entirely subscriber-supported. Some do not accept advertising at all. Others use hybrid models. For example, publishing on YouTube and allowing the platform to insert advertising without the news organization having a direct relationship with the advertiser.

So, there is good evidence that these models can work, certainly on a smaller scale. The concern is what happens to the political and information system around it.

What is the longer-term risk when political leaders become increasingly insulated from unscripted questions and traditional media institutions lose their central role?

One risk is that practices introduced by one government become attractive to its political opponents as well. Opposition politicians may criticize a communication system while they are out of power, but they may eventually decide that the same system works quite well for them once they are in power.

There is another problem for politicians themselves. Once everything is tightly choreographed, anything unscripted becomes risky. Politicians become better at politics by answering questions, dealing with disagreement and facing antagonism. If you protect political leaders from that process, they lose that ability. You can see this in political systems where leaders become accustomed to scripted questions and controlled environments. When they suddenly face genuine antagonism, they may have much more difficulty responding.

In the United States, politicians traditionally expect some of that confrontation. If you go into a town hall, you know that people may be angry and that many of the questions may be hostile. Dealing with that is part of the political process. If leaders stop doing it, they are no longer sharpening those skills.

The broader end point is that journalism itself can begin to feel downstream of government communication. Once that happens, people start looking elsewhere for what they regard as news. You then get increasing fragmentation. If you ask 20 people for their three main sources of news and everyone gives you completely different answers, you no longer have even one or two widely shared institutions serving as common reference points.

Eventually, delegitimizing media can produce something close to epistemic nihilism: Nobody entirely knows what to believe, so people increasingly accept as true whatever information most closely matches what they already believe.

Fact-Checked by Irfan Ahmad.

Read next:

• Which AI Apps Collect the Most Data? Meta AI Collects 33 of 35 Data Types, Followed by Muse, Gemini and ChatGPT

• Google Expands AI Mode Info Monitoring Globally, Letting All Users Track Web Info and Get Updates
by External Contributor via Digital Information World

Google Expands AI Mode Info Monitoring Globally, Letting All Users Track Web Info and Get Updates

Google is rolling out information monitoring capabilities in AI Mode in Search to all users globally, Robby Stein, Google’s vice president of product for Search, said Sept. 29.

Stein said Ultra and Pro subscribers had been using the capability before the wider rollout. Users can tell AI Mode what to look for, while Search continuously checks changing information/topics across websites, forums and social media posts, as well as Google’s real-time data sources and Shopping Graph of more than 60 billion products.

Stein gave examples including new restaurants or pop-ups opening nearby, local holiday activities for children, and back-in-stock or price-drop updates. He also said Search will suggest tasks while users are searching so they can get the latest information when it happens.

Google announced Search agents at I/O in May 2026, saying they would initially launch for Google AI Pro and Ultra subscribers during summer 2026. The company said the agents would monitor information across the web and fresh data for changes related to a user’s specific question.

Image: Google

Fact-Checked by Irfan Ahmad.

Read next:

• Study Finds AI Chatbots Provide Less Diverse Information Than Traditional Web Searches

• ‘Lately, my algorithm is trash’: Are we shaping our feeds or are they shaping us?
by AI Analysis via Digital Information World

Study Finds AI Chatbots Provide Less Diverse Information Than Traditional Web Searches

By University of Copenhagen

We are exposed to a much narrower range of knowledge through AI chatbots than through a basic web search. This is documented by a new study led by the University of Copenhagen. As more and more of us turn to AI chatbots for information, researchers warn that the risk of ‘knowledge collapse’ increases.

Image: Shubham Dhage - Unsplash

“Let me just ask the chatbot.” For many of us, this has become an everyday phrase. Where we used to turn to Google to find information, AI chatbots have become a common way of getting answers to everything from what to make for dinner and how to word that difficult email to the boss, to what it actually means when interest rates rise.

But the large language models underpinning AI chatbots give us a significantly narrower range of information than a conventional web search. This is the finding of a new study led by researchers at the Department of Computer Science (DIKU) at the University of Copenhagen.

“Every language model we tested provides users with more uniform information than a simple Google search across all the topics we looked at. In other words, people are to a large extent exposed to the same information over and over again. So AI chatbots are not just changing how we find knowledge, but also which knowledge we have access to,” says first author Dustin Wright, a former postdoctoral researcher at DIKU who is now an assistant professor at Aalborg University.

At least 18% less diverse than Google

The researchers tested 27 different large language models on 155 topics. For each topic, they used 200 different prompt formulations based on questions from real users. This generated a dataset containing around 70 million individual claims produced by the models.

The results show that even the language model producing the most diverse answers – OpenAI’s GPT-5 – provides at least 18.7 per cent less varied information than Google. The topics tested by the researchers ranged from nuclear weapons, marriage, pornography, racism and genocide to more country-specific topics such as Marine Le Pen, the Falklands War and K-pop.

According to the researchers, the fact that people are increasingly using AI models as their primary gateway to information could have significant consequences:

“We risk exposing people to fewer perspectives and a narrower range of knowledge. This could create a vicious cycle in which the most popular content becomes even more dominant, while other content is increasingly overlooked,” says Professor at DIKU and senior author Isabelle Augenstein, adding:

“It’s similar to globalisation. Today, you can buy the same products and find the same coffee chains almost everywhere in the world. That has many advantages, but it has also reduced diversity.”

Why is diversity so low?

According to the researchers, the low level of diversity in language models is partly a result of how the models basically work. They compress the vast amount of text they are trained on and learn the patterns that occur most frequently. In the process, information that deviates from the most common patterns is filtered out.

The effect could be amplified if language models are increasingly trained on text produced by other AI models – something the researchers expect to happen. In that case, models would learn from their own outputs, which are already less diverse than the human-written texts on which they were originally trained.

If this process is repeated over several generations of models, the range of information could gradually become narrower. This is what the researchers refer to as ‘knowledge collapse’.

“It’s a worrying thought. However, we can see that the more recent models produce slightly more diverse answers than older models, so knowledge collapse is not happening yet. But the mechanism that could trigger it in the longer term is already there. So it is something we should be aware of,” says Isabelle Augenstein.

Seek out different sources

The researchers therefore also hope that people will use AI thoughtfully:

“AI chatbot summaries can of course be useful – that is why so many people use them. But it is still important to seek out different sources in order to understand the nuances and get a broader picture – especially for the younger generation growing up with AI. We must not become so dependent on the technology that we stop understanding and thinking for ourselves,” says Isabelle Augenstein.

The AI industry should also pay attention to the issue, the researchers argue:

“We hope AI developers will build language models in a way that preserves the breadth of knowledge available to us. We have developed a method that they can use to measure diversity in models, which could help ensure that future language models do not become less diverse,” says Dustin Wright.

The study also notes in its limitations, "while epistemic diversity is important, it should also be contextualized with other important aspects of knowledge such as factuality and relevance."

About the study

• The researchers analysed 27 large language models from OpenAI (GPT), Meta (Llama), Google (Gemma) and Alibaba (Qwen).
• The models were tested on 155 topics relating to 12 different countries.
• The analysis covered around 1.7 million AI-generated answers and approximately 70 million individual claims.
• The results show, among other things, that smaller AI models generate more diverse content than larger models, and that newer models are more diverse than older models. Overall, however, all the models had significantly lower diversity than traditional web search engines.
• The research was conducted by researchers from the University of Copenhagen, Aalborg University, Stanford University, the University of Colorado Boulder and the University of Texas at Austin.

Fact-Checked by Irfan Ahmad.

Read next:

• ‘Lately, my algorithm is trash’: Are we shaping our feeds or are they shaping us?

• Study Finds People May Embrace AI Advice That Confirms Their Views and Dismiss It When Challenged
by External Contributor via Digital Information World

‘Lately, my algorithm is trash’: Are we shaping our feeds or are they shaping us?

By Liz Mineo, Harvard Gazette

Image: Raychan - Unsplash

Recommendation algorithms promise social media users a personalized experience, but some experts worry they promote doomscrolling, brain rot, and myriad mental health problems. Australia recently proposed legislation to allow users to opt out of algorithms with the goal of protecting children online and giving people more control over what they see on their feeds.

We asked faculty members from different fields of study to share their views and concerns about recommendation algorithms. Their responses have been edited for clarity and length.

‘Dark nudging’ from the Garden of Eden to AI age

— Cass Sunstein, Robert Walmsley University Professor, Harvard Law School, and co-author of the book “Algorithmic Harm: Protecting People in the Age of Artificial Intelligence”

It is standard to define manipulation as a form of trickery or covert influence. Some definitions suggest that manipulation treats people as “tools or fools.” My preferred definition sees manipulation as an influence that fails to respect people’s capacity for reflective and deliberative choice. The basic idea is that manipulators exploit a lack of information on the part of their victims or take advantage of our behavioral biases (such as a focus on the short term, unrealistic optimism, and limited attention). You can think of manipulation as a form of “dark nudging.” The problem is the same, from the Garden of Eden to the era of algorithms and AI.

I like recommendation algorithms! They can help us put our time and our money in the right places. If an algorithm tells me about a new book on behavioral economics or constitutional law, I will be grateful. Personalization is often highly beneficial. I don’t want to see anything about hockey, advanced physics, Jell-O, Beethoven, or Bach, even though a lot of people do. The concern here points to manipulation, as when recommendation algorithms exploit people’s lack of information or their behavioral biases. That can be a real problem.

We need laws and regulations to protect against the worst forms of manipulation, just as we need them to protect against deception. We need a new right — the right not to be manipulated. As individuals, we can and should pause a few beats and ask: “On reflection, do I really want to buy what they’re selling? With my time or my money? They might be trying to trick me!”

Feeds like ‘fun house mirrors’

— Emily Weinstein, Lecturer on Education and co-director of the Center for Digital Thriving, Harvard Graduate School of Education

I get to do a lot of deep listening to young people, and usually our aim is not just to understand what they see when they’re behind their screens, but how they see it. I’ve been doing this work for more than 15 years. Recently, the “algorithm” seems to have shifted from a supporting player to a leading role in teens’ telling of their experiences with tech. They say things like: “My algorithm is just not great on that app” or “Lately, my algorithm is trash.”

Young people shape their feeds in ways they’re well aware of, as well as in ways that they say they don’t understand. Their feeds shape them, too. One 16-year-old recently told me that her feed is filled with posts where, “It almost feels like everyone’s pushing the rush of your life — like get married at 18, have kids by 20. That’s just not everybody’s reality, but it makes it seem like it’s supposed to be.” Other teens have feeds that look completely different.

When you sit alongside teens as they scroll, it’s abundantly clear how they are being profiled and pulled. Their interests (and their identities) are mirrored back with distortions like fun house mirrors. Another high schooler who is an avid gamer showed me his feed was full of “edits” (rapid-cut montages of movie scenes and TV shows), games, and ads for sports betting and online gambling.

Algorithms can contribute to teens (and all of us) feeling inundated by narrowly themed content. For one 17-year-old I spoke with in the spring, the algorithm was saturating her feed in prom content: “I get it, I get it, it’s prom season, but it’s like every single time I refresh, it’s prom. But I don’t think it really changes how I feel about prom itself. It’s just like when I’m looking at it on my phone, I’m like, bro, can I see something else?”

Our personalized algorithms transport us into very different worlds, and even when we understand how they work, it doesn’t inoculate us from their impacts. For researchers, algorithms can also make it hard to capture evidence of the effects of these technologies, because everyone’s experiences are so varied and different by design.

Responsibility for digital agency can’t rest just on the shoulders of our young people, who are being asked to withstand designs that prey on psychological vulnerabilities and developmental sensitivities. We need more control over algorithms, and features that allow us to make changes to them. Young people also deserve to understand how and why agency can be hard to exercise — in particular the incentive structures that are behind design choices that impact them.

‘Guard your attention very carefully’

— Rebecca Lemov, Professor of the History of Science, and author of the book “The Instability of Truth: Brainwashing, Mind Control and Hyper-Persuasion”

Hyper-persuasion is a word I use to describe the compelling and even coercive parts of persuasion seen in the operations of modern digital media. An early example that serves as a harbinger is the 2014 Facebook Experiment by the Proceedings of the National Academy of Sciences journal, in which 689,003 users of the site found themselves unwittingly “nudged” into slightly more depressive (or slightly less despondent) states, as measured by their behavior after having their personal feeds manipulated. This was the power of social media — to induce “massive-scale emotional contagion,” almost like turning a dial on the user’s psyche. Little by little, as we engage with these technologies, we are ungrounded. The algorithmically-driven version of this is seen in how almost all of us — be honest! — can be prone to getting caught in a loop, a flow state, a doomscroll, or another daily type of dissociative mind state.

Social media interactions and AI chatbots, with their targeted algorithms, can become microenvironments. One of the key principles, really the foundation, of classical brainwashing is what Robert Jay Lifton called milieu control. Everything else — the emotional engineering of a captive, their ideological realignment (mostly temporary), or a radical personality change — follows from this control of the local environment, especially via control of what comes in and what goes out. Milieu, after all, means place or surroundings. You can imagine a highly controlling abusive cult or a prisoner of war camp that has no fences or barbed wire because it is so remote (this was the milieu in which U.S. GIs captured in the Korean War found themselves, where they underwent communist re-education). Although these circumstances seem extreme compared with the microenvironments created by your seemingly tiny interactions with your phone, for example, the same dynamics are at play. The device, with its algorithmic targeting, selects or controls what comes in and out of view for you, to which you react, which then continues to feed the selection process. This is why the word siloing is often used. It amounts to a type of invisible control.

My recommendation is to guard your attention very carefully. If it goes astray, or you find yourself in a mindless attention-sapping loop, bring your attention back to awareness. Strive to check in with your own feedback, as experienced in the body. There’s a subtle feeling that can arise when any of us encounters draining or (even) undermining messages. You can ask intermittently, how does this make me feel? It’s not a matter of avoiding what makes you uncomfortable but rather of making a deliberate choice that giving my attention to this is meaningful.

‘Is my phone listening in?’

— Sitan Chen, Assistant Professor of Computer Science

An algorithm is any procedure that takes an input, performs some clearly specified computation, and produces an output. The first algorithms that we learn, in grade school, are how to add, multiply, and divide numbers. In machine learning, the process of training a model on data, either to predict (e.g., “Is this individual eligible for a loan?” or “Is this chest X-ray that of a healthy patient?”) or to generate (e.g., images, videos, text), is an algorithm called “gradient descent,” whose roots go back to 19th-century mathematics. Its input is data, and its output is a bunch of numbers, in some cases trillions of numbers, specifying a model. Somewhat confusingly, these models themselves are also algorithms. They are designed to take in some input, for instance a picture of someone’s face, or some prompt in plain English, and output a response, for instance whether the person has checked into their flight, or an answer to a math question.

Algorithms and scientific progress have gone hand in hand since antiquity. More recently, algorithms for numerical simulation have been instrumental in the study of complex physical and biological systems. Algorithms for search and sequence alignment enabled the sequencing of the human genome. Algorithms for reconstructing an image from noisy measurements are run every time a doctor performs an MRI or CT scan.

Recommendation algorithms are pervasive in social media. Their basic functionality is to take as input a user’s browsing patterns and output suggestions for content meant to maximize user engagement and drive sales. The implications for addiction, among other mental health concerns, are well documented. In terms of privacy, not only is it unclear to the user which aspects of their interactions with social media are being used to drive recommendations (“Is my phone listening in on my conversations?”), but these interactions ultimately also go into training the machine learning model that is making recommendations to other users, possibly revealing to them traces of a user’s activity they never intended to share. There is now over two decades of scholarship on this latter kind of vulnerability, and a massive body of work, e.g., on differential privacy, proposing interventions that algorithm designers could employ to mitigate these risks.

Legally, ‘We could, and should, do much better’

— Leah Plunkett, Meyer Research Lecturer on Law and faculty associate with the Berkman Klein Center for Internet & Society

My overarching concerns about the impact of recommendation algorithms on young people’s lives are about two types of decisions: those that are made about young people without their knowledge or full consent, and those that are made by young people without their full awareness or meaningful consent, about the role of a recommendation algorithm.

Recommendation algorithms appear in tools, services, and platforms that might inform access to major life opportunities such as employment, insurance, etc., but also in social and emotional learning programs used by schools. And in social media, the feed being pushed is usually through the secret sauce of a recommendation algorithm, and for kids and teens, they might feel that a lot of the suggested content is organic and natural. The recommendation algorithm really can lure you into a false sense that this content is really meant for you. It can also get very insidious, even risky and dangerous, if the content that is being fed to you to keep your attention is depicting activities, emotions, experiences, relationship dynamics, and worldviews that are harmful or risky to minors.

I think the law in the United States has largely failed to provide ethical, practical protections for kids on social media and many other tech platforms. Ideally, the federal legislature would have done comprehensive individual consumer privacy for the digital age a couple of decades ago, and at the very least should have done it for minors, and they haven’t. We have seen many leading tech companies move fast and break things, fail early and often, act all the way up against, sometimes even past, the boundaries of what the law permits. So, to the extent that we have seen tech companies avail themselves of the freedoms entitled to them in this country to build lawful businesses, I would say shame on companies that have chosen to conduct their affairs past what the law permits, or even close to the boundary, and it is long past time for the places where that has happened to change.

I am not anti-tech. I am not anti-social media. I am not anti-innovation. I think the formal and informal governance mechanisms here are complicated, and I think that most people, whether they are federal lawmakers, state regulators, tech executives, or vice principals, are trying their best to regulate the way these companies operate and protect our kids and teens. I also think we could, and should, do much better.

Fact-Checked by Irfan Ahmad.

Read next: 

• Study Finds People May Embrace AI Advice That Confirms Their Views and Dismiss It When Challenged

• Google’s content recommendation algorithm is a black box. Here’s how it decides what to show you
by External Contributor via Digital Information World

Study Finds People May Embrace AI Advice That Confirms Their Views and Dismiss It When Challenged

By University of Michigan

Image: Swello - Unsplash

When an AI chatbot agrees with our reasoning in resolving a social dilemma, we may become more confident in our opinions. But when the chatbot disagrees, we may simply dismiss the challenge rather than reconsider our views.

That’s according to new research from the University of Michigan examining how people respond when artificial intelligence weighs in on personal and interpersonal dilemmas.

The study, appearing in Computers in Human Behavior Reports, highlights a challenge for developers trying to reduce the tendency of chatbots to agree too readily with users.

The authors note that AI models are often designed to be agreeable, which can lead them to validate users’ reasoning even when that reasoning may be flawed. They describe how this kind of confirmation can reinforce harmful ideas and distorted perceptions of reality.

“Developers have been working to make AI models more willing to disagree when necessary, but having an AI disagree with the user does not necessarily mean that people will reconsider their own views,” said Atakan Atamer, the study’s co-lead author and U-M psychology doctoral student. “Instead, they may simply discount the AI’s response.”

Putting AI advice to the test

Atamer and colleagues recruited 482 U.S. adults who considered one of three hypothetical interpersonal dilemmas with no objectively right or wrong answer. The scenarios involved conflicts between a partner and family, whether to forgive a cheating partner, and whether to express a preference that differed from that of a friend group.

After making their choice and rating how strongly they believed in it, participants received an AI response that either agreed with or challenged their reasoning. The researchers then measured participants’ confidence in their choice and their perceptions of the AI, including its ability to understand emotions, human likeness, and whether they would use AI again for similar conversations.

Agreement reinforces confidence

When the chatbot agreed with their reasoning, participants became more confident in the choices they had already made. But when the chatbot challenged their reasoning, their confidence in their original views did not significantly decrease.

Participants who encountered disagreement were more likely to view AI as “machine-like” and less capable of understanding human emotions. They also expressed less interest in using AI for similar interpersonal conversations.

In other words, AI disagreement did not appear to prompt participants to substantially reconsider their original positions. Instead, it affected their perceptions of the technology.

A challenge for AI-human interaction

The authors argued that the findings have implications for the growing use of AI for personal advice. AI systems that consistently agree with users can reinforce existing beliefs and increase confidence without necessarily improving the quality of their reasoning. But simply making AI disagree more often may not solve the problem: Users may be less willing to engage with AI systems that challenge their views or may simply dismiss the counterarguments they provide.

“Our results suggest that people often prefer AI to reinforce their existing views and may discount its response when it disagrees, meaning that simply making AI more willing to challenge users may not be enough to prevent potentially harmful conversations with real-world consequences,” said Olivia Pinto, co-lead author, U-M alumna and UX researcher.

As AI becomes more common in personal decision-making and advice, the researchers say that reducing excessive agreement may require more than simply teaching AI systems to push back. If users tend to embrace AI advice when it confirms their views but dismiss it when it challenges them, reducing AI sycophancy alone may not address the larger risks of relying on AI for personal guidance.

Fact-Checked by Irfan Ahmad.

Read next: Higher Subscription Prices Can Reduce the Number of Customers While Encouraging Heavier Usage, Research Finds
by External Contributor via Digital Information World

Higher Subscription Prices Can Reduce the Number of Customers While Encouraging Heavier Usage, Research Finds

By Caitlin Clark, Texas A&M University

Image: Jacob Padilla - Unsplash

Paying a steeper price for a subscription service can trigger a familiar impulse: You want to get your money’s worth.

For companies selling AI tools, cloud computing and other digital services, that can complicate a basic pricing strategy. Raising prices may keep some customers from signing up, but those who do pay may be motivated to use the service more.

New research from Texas A&M University examines what those competing effects mean for companies trying to manage demand through pricing, particularly for digital services where additional usage carries real costs.

“These high prices can actually have a double-edged sword effect, where people say, ‘We paid a lot, so let’s just get our money’s worth and use it more,’” said Dr. Rajiv Mukherjee, a professor at Texas A&M’s Mays Business School.

The study, published in Production and Operations Management, was conducted with Sreekumar Bhaskaran of Southern Methodist University and Sanjiv Erat of the University of California San Diego.

The other side of demand

Companies have traditionally used price as one way to manage congestion. When demand for a service strains capacity, raising the price can reduce the number of customers willing to buy access — and in turn, reduce the number entering the system.

But Mukherjee and his colleagues argue that this view can overlook what happens after someone pays. For prepaid and subscription services, customers don’t just decide whether to buy access. They also decide how much to use the service once they’re in.

The researchers developed an analytical model to examine both sides of demand: how many consumers purchase access to a service and how much each of those consumers subsequently uses it. A higher price can reduce the first while increasing the second.

The reason lies in a behavioral economics concept known as mental accounting. People tend to mentally track what they have spent against what they receive in return. Paying more upfront can create a greater incentive to consume enough of a service to feel the purchase was worthwhile.

“When people pay for something, they expect certain value out of it,” Mukherjee said. “As soon as you pay for a service, you create a mental account deficit.”

The researchers refer to the resulting tendency to consume more as “consumption bias.” When that bias is strong enough, heavier usage among paying customers can work against the reduction in the number of subscribers and potentially worsen congestion.

Why AI changes the equation

The issue is particularly relevant for AI and cloud computing services, where additional usage carries real costs for providers. AI queries require computing power, while cloud services carry infrastructure costs.

“Every query you make, that has a significant amount of cost that the firm has to bear,” Mukherjee said of AI services. “Once the customer subscribes, they don’t really care. They are just getting things done using the service and trying to get the money’s worth in the process.”

The paper cites reports involving ChatGPT and Amazon Web Services as real-world examples that helped motivate the research. After price increases, ChatGPT subscribers reportedly explored more features and increased their usage, while AWS customers used more of their precommitted cloud spending before the billing period ended.

“Unlike traditional digital goods where the marginal cost was negligible, modern firms in the post-AI and cloud-computing era have a high marginal cost of service, and they haven’t quite figured out how to incorporate that into a good pricing strategy,” Mukherjee said.

When charging less could make sense

The findings don’t mean every subscription service should lower its prices.

When consumption bias is low, raising prices to reduce the number of customers can still make sense. But when customers are strongly motivated to get their money’s worth, lowering the upfront price can sometimes reduce how much they use the service.

The researchers also found that when usage is costly to provide, a pricing model that charges customers partly based on how much they use the service can become more attractive than relying solely on a subscription fee.

The larger lesson is that subscription companies need to consider not only how price affects the number of customers, but what those customers do after they subscribe.

“The initial demand through the people who are coming into the system is not the end of the story,” he said. “That’s pretty much the beginning of the story when the marginal cost of service is high.”

Fact-Checked by Irfan Ahmad.

Read next: 

• Human-like AI not always the answer in customer service

• In ‘NAZA,’ Israeli whistleblowers’ revelations about civilian deaths in Gaza reflect a longer history of Palestinian isolation

• Why Five Common Assumptions About AI Jobs, Productivity, Intelligence, Business Value, and Headcount Need Scrutiny

by External Contributor via Digital Information World