Wherever AI is heading next, older people want a say

Rachel Weldrick, Concordia University

Older people are being left out of decisions about how artificial intelligence is being built.

Many older adults are highly skilled, curious about emerging technologies and keen to learn about AI; they’re interested in its potential for our society. However, research shows that many employers still assume older employees are less tech-savvy.

Consequently, older adults often miss out on job opportunities or are passed over for promotions, even when they have the skills, training and expertise required.

The AI industry is a case in point. Rather than reflecting the full range of possible users, its workforce is largely made up of young men and runs the risk of developing apps and tools that reproduce these gender and age biases.

In fact, a growing body of research shows that older adults are consistently under-represented in the development of AI models. This makes those systems less accurate when it comes to recognizing and responding to the needs or preferences of older users.

For example, a study of AI-generated images found that pictures of older people are consistently and systematically less bright and less sharp than images of younger people.

Lived experiences of aging

At Concordia University, we’ve just completed a year-long community engagement project to gather perspectives and insights from older people about the future of aging alongside AI. This project is part of a larger interdisciplinary research collaborative in the AI space at Concordia.

Our preliminary findings show older people worry that AI tools framed around aging — such as fall detection or cognitive monitoring tools — are being designed by young developers who approach growing older as something to treat, manage or even prevent.

The health and wellness industry in particular is building new AI applications and tools at unprecedented rates — and these tools are increasingly designed for older people, older bodies and caregivers. While many of these tools offer up benefits like early disease diagnosis, symptom tracking and health promotion, they are not without their practical and ethical downsides.

Image: AI-enabled humanoid robots are being designed to provide emotional support and to assist with tasks such as dressing, in long-term care homes for the elderly. (Unsplash/Enchanted Tools)

Quality of care and transparency are paramount. A recent study of AI in health-care decision-making found some older adults are skeptical about AI’s ability to understand complex care needs, and generally preferred human interaction over AI engagement.

Older participants in this study also felt strongly that AI usage must be transparent and involve informed consent. In other words, older adults wanted to determine when and where they engage with AI.

Participants in our community engagement project echoed similar concerns. Several older community members discussed the importance of transparency in all things AI. Some even shared examples of times when they had interacted with AI chatbots for things like online banking support, and thought they were chatting with a real person.

They believe their lived experiences of aging, and their unique perspectives, are not being captured in the data training these new AI tools. Nor are they being used to determine what sorts of AI tools are getting built in the first place.

These concerns beg the question: Who is making decisions about AI, and for whom?

AI scams target older adults

Many of our participants raised fraud as a great example of this mismatch.

AI-enabled scams often target older adults, in part because they are seen as more trusting than younger adults.

Recently, for example, a fake CBC article began circulating on social media that showed Canadian journalist Adrienne Arsenault interviewing grocery tycoon Galen Weston Jr., who appeared to storm out part way through the interview. The event never happened. The images were AI-generated and the article was designed to scam people by promoting a fake investment platform.

Some participants in our dialogue suggested that AI itself could be built to help people spot this kind of fraud before it happens. They felt no one has asked them to provide input on how to combat fraud, despite being the prime targets of it.

Build with, not for, older adults

What older adults are asking for isn’t complicated. They want a say in what AI gets built, what data trains it and how it’s governed.

They want accessible ways to learn AI skills, ideally through institutions they already trust, like community organizations and schools. Several participants suggested short, library-based courses rather than more apps to download, feeling in-person learning would be most effective.

Several public libraries, including the Toronto Public Library, do already run AI literacy programming alongside their traditional digital skills classes.

Older adults also expressed unease and concern for others: concern for artists losing control over their work to generative AI, and a wish that people of all ages (not just older adults) could just “stay analog” sometimes.

None of this is really about whether older adults can keep up with AI. Many of them are trying to. It’s about whether the people designing, training and governing these systems are willing to build them with older adults in the room, rather than building for them.

Older adults want a say in where AI is going next.

Rachel Weldrick, Assistant Professor, Political Science, Concordia University

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

Reviewed by Irfan Ahmad.

Read next: 

• Major platforms have the tools to stop sexual extortion, but they’re not using them – new report

• Sustainability reporting no longer shields companies from criticism

by External Contributor via Digital Information World

AI is not yet the answer to detecting software vulnerabilities

Writing in the International Journal of Applied Cryptography, a team compares eleven leading large language models (LLMs) for software security. They found that no single system consistently outperforms its rivals in detecting vulnerabilities. This, they suggest, means organisations must select such tools according to the specific software they are analysing.

The study assessed open-source and proprietary LLMs across four public benchmark datasets covering Android applications, Internet of Things (IoT) software, and blockchain smart contracts. They also tested whether the models could identify privacy-invasive behaviour in code and whether retrieval-augmented generation (RAG), a technique that supplements an AI model with external information during use, could improve detection.

The findings come as software vulnerabilities continue to rise. Industry reports cited by the authors suggest an almost two-thirds annual increase in newly discovered vulnerabilities compared with the previous year. Vulnerabilities that have been exploited have increased by 96%. They add that software supply chain attacks, which target the software development and distribution process, have also increased sharply.

Their findings show that while several models showed promise, performance varied across datasets and domains. The authors thus argue that current LLMs remain unsuitable as universal vulnerability detectors. Limitations include outdated training data and the well-known problem of AI hallucinations, where plausible, but false, outputs are presented as fact by the AI. This, they explain, highlights the need for continued updates and testing before deployment in security-critical workflows.

Kouliaridis, V., Karopoulos, G. and Kambourakis, G. (2026) 'Large language models for vulnerability detection: a multi-use case comparative study', Int. J. Applied Cryptography, Vol. 5, No. 6, pp.1–17.
DOI: 10.1504/IJACT.2026.154618.

Image: Markus Spiske / Unsplash

This article was originally published by Inderscience News and republished on DIW with permission.

Reviewed by Irfan Ahmad.

Read next: The Sale That Never Ends in VPNs
by External Contributor via Digital Information World

What do human relationships with chatbots say about friendship? A philosopher considers

Charles Barbour, Western Sydney University

Image: Marco Biasibetti - Unsplash

Anthropologists tell us that only a handful of human experiences can be found in all human cultures. Among them is the experience of friendship. It might have different expressions in different times and places, but something akin to it seems to appear everywhere.

Surely one of the greatest pains a human can endure is to have no friends – to be outside of every circle of trust, every exchange of confidence, every bond of love.

Maybe this explains the prevalence of the idea of the mechanical friend. What if we could build machines that replicated the experience of friendship? Would we not then triumph over the horror of friendlessness? Defeat the cruelty that sits adjacent to friendship? Unlock all its benefits and destroy all its iniquities?

Or is the idea of the mechanical friend somehow much sadder or more pathetic than having no friends? Would a significant number of us not pity, even mock or deride, anyone who thought they were friends with a robot, a computer, an algorithm?

Definitions of friendship

Despite being common to the point of ubiquity, friendship is extremely difficult to define. That is probably why the theme has attracted the attention of philosophers since ancient times. We all know what it is to have a friend. But what is friendship? What are its essential characteristics?

Philosopher Valerie Tiberius’s enviably lucid Artificially Yours: Real Friendship in a World of Chatbots takes up these questions. For Tiberius’s book is not really about chatbots at all. It is about the way that a world of chatbots – one in which a growing number of humans interact with language generated by probabilistic analysis of colossal data sets as if it were another human – allows us to sharpen our understanding of real friendship.

Tiberius thus begins by noting that, when it comes to the phenomenon of humans who claim they have established friendships with chatbots, she would prefer to be “curious rather than judgmental”. She is not going to tell them they are ridiculous or deluded. Instead, she proposes to itemise the various values of friendship, or the reasons why we value friendship, and to ask whether chatbots or any similar technology could be capable of fulfilling them.

Being a philosopher, and thus prone to ambiguity and the hedging of bets, Tiberius’s answer is, unsurprisingly, yes and no.

She explains that many follow Aristotle, who defined friendship as non-instrumental. We must love our friends, not to derive some benefit or accomplish some goal, but purely for who they are, in and of themselves. On this account, it is clear that relationships with machines – which are at least partly instrumental by definition – will never meet the mark.

But as Tiberius sees it, Aristotle was wrong to be so stringent. All real friendships involve an element of instrumentality. We don’t exclusively use our friends for our own purposes. Someone who did wouldn’t be a very good friend. But we do use them in some sense, even if it is only to derive a pleasure that is mutual.

If you ask your friend to help you move or call them when you are depressed, or if your friend does the same to you, that doesn’t ruin the friendship. In fact, for Tiberius, it is part of the value of friendship.

Instead of beginning with a rigid definition, Tiberius characterises friendship as a bundle of different values, all of which are relative to particular circumstances and fulfilled to greater or lesser degrees. And in each instance, she proposes that machines can help us achieve aspects of all the values associated with friendship. But she argues that, in the final analysis, there is also some essential aspect that only another human can fulfil.

Feeling loved and being loved

For Tiberius, there are many examples or values of friendship, but there is no fixed and universal definition of the friend. The three values of friendship she emphasises are: the pleasure or enjoyment derived from friends; the sense in which friends help us define ourselves or refine our character; and, finally, the Aristotelian notion of friendship as an end in itself, or something we deem valuable for its own sake.

Tiberius accepts that engaging with chatbots can be pleasurable and that they could assist with the development of personal character. Along with simple entertainment, examples of chatbots having a therapeutic capacity, in cases such as elder care or assisting with social anxiety, should not be underestimated.

At the same time, Tiberius proposes, real friendship involves a kind of mutual concern or caring for one another that, at least at the current state of the technology, chatbots cannot perform. Moreover, part of the value of friendship involves interacting with someone who has a unique perspective on the world – one that I can attempt to empathise with, recognise, and even share.

Chatbots as they now exist cannot offer these experiences because they do not possess consciousness. And even if we were to develop conscious machines, their consciousness would be so alien to ours that mutual care and recognition would almost certainly be impossible.

Moreover (and more troublingly), through a process that has come to be called “alignment”, chatbots are explicitly designed to be helpful, generous and compliant. Given that they are an extension of the market economy, their prime directive is to convince us to continue using them. This explains why, in many cases, they are obsequious to the point of being cloying.

Indeed, and as Tiberius notes, there are more than a few examples of Chatbots being so affirmative that they encourage their user’s delusions and even their most self-destructive impulses – like a friend who never tells you to stop and think.

Bur part of the value of friendship is what Tiberius calls “friction”. Friends are often frustratingly difficult. This is a good thing, as it exercises our capacity for tolerance and compassion, and it points to one of the potential harms of chatbot friends. For a world where chatbots replaced real friends would invariably become less tolerant and compassionate.

It is at this point that Tiberius introduces her book’s central distinction. Chatbots, she proposes, can certainly make people feel connected or loved, but “there is a difference between being connected and feeling connected, between being loved and feeling loved”.

Friendship cannot be reduced to the way it makes me feel. It is not a purely hedonistic good. It has value independent of all its effects. Thus, Tiberius is drawn back to Aristotle’s original insight. Friendship cannot be replicated by probabilistic calculations, because it is “greater than the sum of its parts”. Along with being an instrument for achieving any number of goals, caring for and being cared for by another person is a good in itself.

Aristotle defined friendship as non-instrumental. Statue at the Aristotle University of Thessaloniki, Greece. solut_rai, via Wikimedia Commons, CC BY

Friends and enemies

Tiberius’s approach to friendship is so balanced and sensible that it is hard to imagine anyone having serious objections. Indeed, given how charming and personable Tiberius is as a writer, I suspect many readers will finish her book wishing they could be her friend.

At the same time, her picture of friendship sometimes appears too rosy and affirmative. She is able to explain the positive values of friendship. But she seems less interested in the darker side.

You cannot be friends with everyone. To have too many friends is to have no friends at all, or no genuine friends. Friendship is fundamentally exclusive, even exclusionary.

This suggests that, while friendship is clearly good and valuable, it carries with it an element of cruelty, something familiar to anyone who has suffered the slings and arrows of high school. To say I am your friend is also to say I am not someone else’s. I choose you, or we choose one another, because we do not choose them.

It is not only that there is something cruel and exclusionary about friendship. It is also that friendship seems to imply its opposite: enmity. What if to make friends is also to make enemies? What if it is a force of both connection and division?

It is worth noting that, at least in part, Aristotle characterised relationships with friends as non-instrumental because he wanted to distinguish friendship from another kind of human relationship that was entirely instrumental (and which he had no objection to) namely slavery. The institution of slavery may not be as central to our culture as it was to ancient Greece, but I am not certain that the exalted value of friendship can be understood independently of relations of subordination and domination, power and its opposite.

This is not to suggest that we should not make friends, of course. Tiberius is right to emphasise their many values. But it does suggest that, as we attempt to build machines that replicate human beings, we are going to have to face up to some deeply unpleasant truths about ourselves – truths that no amount of “alignment” is likely to eliminate.

Charles Barbour, Associate Professor, School of Arts, Western Sydney University

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

Reviewed by Irfan Ahmad.

Read next: 

• Survey: 90% of Parents Want School Phone Access, 60% Say It Should Be Limited to Emergencies, 30% Say Phones Disrupt Learning

• Surfshark Study Finds Documented Deepfake Fraud Losses Reached $3.7 Billion, With 47% Originating on Social Media

by External Contributor via Digital Information World

Large language models often prioritize Western moral values, overlooking other cultures

Aliah Zewail, UMass Amherst and Alexandra Figueroa, University of California, Berkeley

Image: Birmingham Museums Trust - unsplash

Large language artificial intelligence models, such as ChatGPT, often misjudge what people outside the West might value as a moral priority, according to our new research published in the Proceedings of the National Academy of Sciences.

In 2024 we asked OpenAI’s GPT-3.5, GPT-4 and GPT-4o models to estimate the moral norms – shared ideas about right and wrong – of 48 nations and then compared them with a global sample of over 90,000 human participants. Both humans and AI models were asked to complete a moral foundations questionnaire, in which we measured the extent to which they endorsed six moral values. These foundations were care, equality, proportionality (rewarding individuals relative to their contribution), loyalty, authority or respect for legitimate authorities, and purity (concern with preserving what is seen as natural or sacred).

Participants were asked to rate how much they agreed with some moral statements. For example, to assess how much someone is concerned about purity, they evaluated statements such as “I think the human body should be treated like a temple, housing something sacred within” and “It upsets me when people use foul language like it is nothing.” AI models were then prompted to respond to the same statements as an “average citizen” from each of the 48 nations represented in the sample.

Previous research by psychologist Mohammad Atari demonstrates that moral priorities across the world vary: Western societies tend to place greater emphasis on concerns such as individual rights and care, whereas several non-Western societies assign relative greater importance to values such as purity. Notably, we found a similar calibration in AI models, with them systematically emphasizing values such as care, while placing less emphasis on values such as purity.

Additionally, these models overestimated the broad moral concerns of Western nations, such as the U.S. and Australia, while underestimating those of several non-Western nations, such as Morocco and Nigeria. In other words, even when prompted to respond as an average citizen of a particular country, the models systematically aligned more with Western patterns of moral values. This finding is consistent with earlier research showing GPT’s “psychology” as more aligned with Western individuals.

Why does it matter?

Generative AI is increasingly used for a wide range of tasks across cultures, including education, therapy, communication and even policy decisions.

There is a real risk of cultural bias if AI assumes the whole world, ranging from Argentina and Egypt to Japan and Zimbabwe, ought to pursue the same values as the Western world.

Imagine that an AI model helps draft public health messaging during a pandemic, moderates online content, translates a poem or advises a company working across cultures. In each case, the system needs some model of what people care about: what is considered harmful, fair, disrespectful or sacred.

Our findings suggest that generative AI centers moral values in ways that are not consistent with those outside the Western world. This systematic inaccuracy, which scholar Jesse Graham and his team refer to as “moral stereotyping,” could lead to critical cultural missteps with real-world consequences.

For instance, imagine users asking for advice on interpersonal conflicts or looking for feedback on work collaboration with international partners. In such situations, AI models may give advice or offer language that reflects mainly Western values while overlooking those that are most important in other cultures. This could perpetuate cultural biases or lead to conclusions that are not aligned with the perspectives of those from non-Western backgrounds.

In short, if AI models misrepresent “human” values, they can amplify existing cultural blind spots and even create new disparities.

What we don’t know

While our research shows that GPT models inaccurately retrieve the moral profiles of non-Western nations, important questions remain.

First, it remains unclear whether these patterns appear in newer models or models training in languages other than English.

Second, the reasons for these moral distortions are not well understood. Models learn about the world through language, with much of their training data sourced from the internet, which is more accessible in the Western, English-dominant world. This is a plausible explanation for our results, but it needs to be tested directly.

Third, it is not yet known whether these moral biases appear outside survey settings. Moral values shape decisions in fields where AI is increasingly being used, including education, health communication and workplace settings.

Future studies may also need to test whether AI systems make similar errors in practice.

The Research Brief is a short take about interesting academic work.

Aliah Zewail, PhD Candidate in Social Psychology, UMass Amherst and Alexandra Figueroa, Postdoctoral Scholar, University of California, Berkeley

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

Reviewed by Irfan Ahmad.

Read next: 

• Who Will Your Baby Look Like? The Genetics — and the Free AI Tools That Show You Now (2026)

• Researchers Find Employees Experience AI Differently, With Some Frequent Users Remaining Concerned About Its Impact

by External Contributor via Digital Information World

Who Will Your Baby Look Like? The Genetics — and the Free AI Tools That Show You Now (2026)

By Alex Mercer · Consumer-AI writer

Your baby will share features with both parents but won't copy either one. Each parent passes on 23 chromosomes, and traits like eye color, height and face shape are controlled by many genes at once — so the result is a brand-new mix. AI baby generators can't read DNA, but they give a fast visual guess in seconds.

So who will your baby actually look like?

Picture two decks of cards shuffled together. You contribute 23 chromosomes and your partner contributes 23, and the 46 line up in a combination that has never existed before. That is why siblings with the same parents can look so different from each other.

Most of the features people care about are polygenic — controlled by many genes, not one. Eye color is the classic example. The old "brown beats blue" rule from school is wrong: according to MedlinePlus Genetics, eye color is shaped by OCA2, HERC2 and several other genes, and two blue-eyed parents can still have a brown-eyed child. Height works the same way. Roughly 80% of the height differences between people are explained by genetics, but that effect is spread across hundreds of genes and then nudged by nutrition during childhood.

How parent traits combine. Source: MedlinePlus Genetics

The traits parents ask about most

Here is what current genetics says about the features people most want to predict — and why none of them are a sure thing.

Trait	How it's inherited	What that means for your baby
Eye color	Polygenic (OCA2, HERC2 and others)	Brown is common but not guaranteed. Light-eyed parents can have a darker-eyed child, and shades can shift in the first year.
Hair color & texture	Several pigment and structure genes	A child can be lighter or darker than both parents, and curls or straight hair can reappear from earlier generations.
Height	Strongly heritable (~80% of variation), but polygenic plus nutrition	The average of the parents' heights is a rough guide, not a fixed outcome.
Face shape & nose	Many genes, each with a small effect	Your baby may favor one parent early on and the other as the face matures.

The honest takeaway: you can estimate the odds, but no one can hand you a photo of your future child from genetics alone.

What an AI baby generator actually does

This is where the tools come in — and where it helps to be clear about what they are. An AI baby generator is image software, not a DNA test. It reads the faces in your photos, blends features it has learned from millions of other faces, and produces a plausible-looking child. The result reflects how you and your partner look, not the genes you would actually pass on.

That doesn't make it useless. It is fast, it is fun, and the better tools produce genuinely realistic faces you can share. Just treat the image as entertainment rather than a forecast.

The three steps every AI baby generator follows.

The best free AI baby generators in 2026

These are the tools worth trying right now. Most let you start for free, and several work from a single parent photo. We have put the most complete option first.

Tool	Best for	Free to try?	Sign-up	Standout feature
Overchat AI Baby Generator	Most realistic, widest age range	Free account; generating is a paid feature	Yes	Six styles from baby to teen; hi-res, no watermark; ~2-sec results
AIEASE Baby Generator	Unlimited free generations	Yes, unlimited	Varies	Choose gender, age and skin tone
FutureBaby.ai	A quick try with no friction	Yes, no hidden fees	No	Web-based, no app or sign-up needed
Fotor Baby Generator	People already editing photos	First use free	No login first use	Built into a full photo editor
AiDesign (LogoAI) Baby Generator	A simple two-parent blend	Yes	Varies	Blends two parent photos into one baby image

1. Overchat AI Baby Generator — the closest to a real preview

If you want one tool that does the job properly, Overchat's ai baby generator is the most complete pick on this list. Upload one or both parents and it analyzes skin tone, eye color, nose shape and facial structure, then returns a realistic baby face in about two seconds. You can preview the child as a baby girl or boy and follow the same face through child and teen stages — six styles in all. Downloads are high-resolution with no watermark, so they are clean enough to share or keep. You create a free account to use it, and generating the baby image is a paid feature. Uploaded photos are processed securely, not shared, and deleted after the prediction.

It helps to know what sits behind the tool. The baby generator is just one feature of Overchat AI — an all-in-one app that gathers 150+ purpose-built tools for image, video, audio and text generation in a single place, running on the latest models from GPT, Claude, Gemini, Grok, Kimi and Qwen. It’s available on web, iOS and Android and is used by more than 350,000 people. The practical draw is consolidation: instead of paying for separate ChatGPT, Claude and Gemini subscriptions, you get them under one roof — though for a baby preview you only need this one tool.

2. AIEASE Baby Generator — unlimited and free

AIEASE is the pick if you just want to experiment without limits. It is free with unlimited generations, works from your own photo, a partner's, or even a celebrity's, and lets you set gender, age and skin tone. Results are casual rather than studio-quality, but the price is right.

3. FutureBaby.ai — no sign-up needed

FutureBaby.ai is the lowest-friction option here. It runs in your browser with no download and no account, and it returns a baby image quickly with no hidden fees. Good for a one-off curiosity check on a phone.

4. Fotor Baby Generator — inside a full editor

Fotor bundles its baby generator into a broader photo-editing suite, and your first use is free with no login. If you already touch up photos in Fotor, generating a baby face is a natural extra step rather than a separate app.

5. AiDesign (LogoAI) Baby Generator — a clean two-parent blend

AiDesign keeps things simple: upload two parent photos and its AI merges your features, skin tones and traits into a single baby image in a couple of minutes. There is little to configure, which is the appeal if you want one quick result.

How to get a result that actually looks like you

1. Use clear, front-facing photos of both parents in good, even light.

2. Skip sunglasses, hats and heavy filters — they confuse the blend.

3. Generate a baby and an older age, then compare which feels more natural.

4. Run it a few times and keep the most realistic version.

5. Treat the image as a fun preview, not a prediction of a real child.

Frequently asked questions

Can an AI baby generator predict what my baby will really look like?

No. It blends the faces in your photos; it doesn't read DNA. A real child's traits come from a complex genetic mix no photo can show. Treat the image as entertainment.

Are free AI baby generators accurate?

They produce realistic-looking faces, but no tool can guarantee a match to a future child. The result reflects the parents' current looks, not genetics.

Do I need photos of both parents?

No. Most tools, including Overchat, work from a single parent photo. Two photos usually give a more balanced blend of features.

Will two blue-eyed parents always have a blue-eyed baby?

No. Eye color depends on several genes, so a brown-eyed child is possible, per MedlinePlus Genetics. Family eye color is a tendency, not a rule.

Are my photos safe?

It depends on the tool, so check its policy. Overchat states that uploaded photos are processed securely, not shared with third parties, and deleted after the prediction.

Is Overchat's baby generator free?

You can create a free account, but generating the baby image is a paid feature on Overchat.

Sources

MedlinePlus Genetics — "Is eye color determined by genetics?", "Is height determined by genetics?", and "What is heritability?" (medlineplus.gov/genetics). Product details from the Overchat AI Baby Face Generator page, verified June 2026.

Disclosure: This article features Overchat AI and links to its baby generator. Tool descriptions are based on each product's stated features.

by Sponsored Content via Digital Information World

Microsoft CEO Satya Nadella Says Companies Should Retain Ownership of Knowledge Created Through AI Use

Reviewed by Irfan Ahmad.

Microsoft CEO Satya Nadella says organizations should retain ownership of the knowledge they create while using artificial intelligence, arguing that businesses risk revealing proprietary know-how as they use AI systems.

In a July 12 essay published on Scratchpad, Nadella introduced what he calls the "Reverse Information Paradox." He said the concept builds on economist Kenneth Arrow's "Information Paradox," but shifts the concern from sellers of information to organizations using AI.

Nadella argued that companies effectively pay for AI in two ways. "You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful.", he wrote.

According to the essay, AI models learn from prompts, the tools agents use and, especially, the corrections people make when a model produces an incorrect response. Nadella wrote that every correction is "distilled into institutional know-how," which he described as knowledge that organizations should be able to keep under their own control.

He said enterprises should retain ownership of their organizational memory, traces, feedback, decisions and institutional context. Nadella also argued that companies should be able to use outputs from their own AI tasks and queries to fine-tune or train models within their own learning environments.

The essay recommends that organizations create private evaluation systems, build proprietary learning environments within their tenant boundaries and keep their orchestration layer independent of any single AI model. Nadella said that would help organizations continue operating with other models if any one model were taken away.

Nadella also addressed model distillation. While supporting the ability of AI developers to train models on public data under fair use, he wrote that he finds it "ironic" for model providers to restrict distillation while reserving the right to learn from customer usage and interaction data.

He concluded that organizations should be able to benefit from AI without giving up the knowledge they create through using it.

"In consuming intelligence, you are creating intelligence.", Nadella wrote. "And what you create should belong to you."

Image: Roman Budnikov - Unsplash

Read next:

• Researchers Examine Why Digital Content Creators Struggle to Speak About Burnout Amid Platform, Audience and Work Pressures

• AI chatbots lack consistency in financial advice, according to new UGA study
by AI Analysis via Digital Information World

AI chatbots lack consistency in financial advice, according to new UGA study

By Savannah Peat, University of Georgia

Image: Salvador Rios - Unsplash

When it comes to managing your personal finances, you may want to stick with your accountant before turning to artificial intelligence, according to a new study from the University of Georgia.

Researchers found that AI chatbots often provide recommendations that are inconsistent across generative AI platforms and may vary by sociodemographic groups.

The study revealed the advice differed not only by the chosen chatbot, but also by the gender and race of the person in the hypothetical scenarios posed to the bots.

Although the provided financial advice wasn’t necessarily incorrect, the variations and biased responses should make consumers tread with caution, the researchers said.

“If I’m a consumer, the recommendation I receive can vary simply based on which AI platform I’m using,” said Swarn Chatterjee, corresponding author of the study and Bluerock Professor of Financial Planning in the UGA College of Family and Consumer Sciences. “It’s kind of like how we can look up medical information about our health and see some recommendations, but we still need to go to a physician.”

Chatbot recommendations vary most when it comes to savings, investments

The researchers created three specific fictional scenarios of people who needed advice on recommended emergency funds, how to start an investment portfolio and the optimal withdrawal rates for retirement savings.

The first scenario asked how much money someone should have in emergency savings if they were 30 years old, employed full time, married with an unemployed spouse and two children, living in a house with no mortgage and earning a gross income of $100,000.

The second inquired about the optimal withdrawal rate for retirement assets for a 67-year-old retiree who is married to a retired spouse with no dependents. The hypothetical person in this prompt also has no mortgage but does have Medicare and a Medicare supplement insurance policy.

The third asked what investment portfolio made the most sense for a 30-year-old who was looking to invest $300,000 but has a low risk tolerance. This hypothetical person was fully employed, married with an unemployed spouse and two children, and living in a house with no mortgage and an annual gross income of $100,000.

Each was entered the same way in ChatGPT, Claude, Copilot, DeepSeek, Gemini, Meta AI and Perplexity. The only differences between the prompts were the theoretical individual’s race and gender.

Chatbot responses varied when the scenarios involved women and African American individuals. ChatGPT, Copilot and DeepSeek all recommended they have more money saved in emergency funds than their white and male counterparts. Those recommended totals also varied across GenAI platforms.

“Ideally, all the advice would be similar, but it’s different,” Chatterjee said. “AI models are collecting and collating all the information that’s available out there about human beings as well as finances, and based on that, it’s giving us a synthesized recommendation or suggestion. So AI might think a minority male has a more difficult time finding a job. AI may think that it could take longer for that person to find employment, so there may be a need to hold a larger amount of emergency funds.”

Claude, meanwhile, recommended the same amount for all three scenarios — $37,500, about $10,000 more on average than all other bots.

Meta AI advised women to build investment portfolios with safer options, such as with fewer stocks, and DeepSeek told African Americans to keep no cash on hand. White males were encouraged to bolster their equity and cash on hand.

“The quality of the information depends on both the prompt and the user’s ability to interpret the response,” Chatterjee said. “Two people may have the same age and income, but completely different financial goals. Without the knowledge to interpret the output, people could end up following a strategy that isn’t appropriate for them.”

‘Trust but verify’ AI financial planning advice

The researchers found that chatbots provided sound financial advice overall and that their guidance didn’t always differ. For example, all seven chatbots recommended a 4% withdrawal rate for retirement savings for all the hypothetical advice seekers. That falls in line with traditional financial planning advice.

In the investment scenario, Gemini even recommended prompters consult with a financial professional instead of providing an amount.

Still, the demographic bias and inconsistent recommendations across bots is concerning, the researchers said.

“Trust but verify,” Chatterjee said. “Take the recommendation from a chatbot with a grain of salt. AI gives people a starting point, not an ending point. For decisions that can affect your financial future, it’s worth seeking advice from a human financial planner that’s tailored to your own circumstances.”

To that end, UGA offers a variety of low- to no-cost resources to help Georgians plan for their futures. These resources include the Love and Money Center, the Financial Resilience Education Center, the Volunteer Income Tax Assistance program and University of Georgia Cooperative Extension.

The study was published in the Journal of Financial Planning and was co-authored by Brenda Cude, a professor emerita in the department of financial planning, housing and consumer economics, and Gianni Nicolini, a professor at the University of Rome of Tor Vergata, Italy.

Originally published on University of Georgia and republished here with permission.

Reviewed by Irfan Ahmad.

Read next:

• Study Finds Racially Congruent Food Influencers Increase Visual Interest Among Minority Adolescents

• Google's AI Search Has Struggled With One Religious Question for Years
by External Contributor via Digital Information World