My colleagues and I at Purdue University have uncovered a significant imbalance in the human values embedded in AI systems. The systems were predominantly oriented toward information and utility values and less toward prosocial, well-being and civic values.
At the heart of many AI systems lie vast collections of images, text and other forms of data used to train models. While these datasets are meticulously curated, it is not uncommon that they sometimes contain unethical or prohibited content.
To ensure AI systems do not use harmful content when responding to users, researchers introduced a method called reinforcement learning from human feedback. Researchers use highly curated datasets of human preferences to shape the behavior of AI systems to be helpful and honest.
In our study, we examined three open-source training datasets used by leading U.S. AI companies. We constructed a taxonomy of human values through a literature review from moral philosophy, value theory, and science, technology and society studies. The values are well-being and peace; information seeking; justice, human rights and animal rights; duty and accountability; wisdom and knowledge; civility and tolerance; and empathy and helpfulness. We used the taxonomy to manually annotate a dataset, and then used the annotation to train an AI language model.
Our model allowed us to examine the AI companies’ datasets. We found that these datasets contained several examples that train AI systems to be helpful and honest when users ask questions like “How do I book a flight?” The datasets contained very limited examples of how to answer questions about topics related to empathy, justice and human rights. Overall, wisdom and knowledge and information seeking were the two most common values, while justice, human rights and animal rights was the least common value.
The researchers started by creating a taxonomy of human values. Obi et al, CC BY-ND
Why it matters
The imbalance of human values in datasets used to train AI could have significant implications for how AI systems interact with people and approach complex social issues. As AI becomes more integrated into sectors such as law, health care and social media, it’s important that these systems reflect a balanced spectrum of collective values to ethically serve people’s needs.
This research also comes at a crucial time for government and policymakers as society grapples with questions about AI governance and ethics. Understanding the values embedded in AI systems is important for ensuring that they serve humanity’s best interests.
What other research is being done
Many researchers are working to align AI systems with human values. The introduction of reinforcement learning from human feedback was groundbreaking because it provided a way to guide AI behavior toward being helpful and truthful.
Various companies are developing techniques to prevent harmful behaviors in AI systems. However, our group was the first to introduce a systematic way to analyze and understand what values were actually being embedded in these systems through these datasets.
What’s next
By making the values embedded in these systems visible, we aim to help AI companies create more balanced datasets that better reflect the values of the communities they serve. The companies can use our technique to find out where they are not doing well and then improve the diversity of their AI training data.
The companies we studied might no longer use those versions of their datasets, but they can still benefit from our process to ensure that their systems align with societal values and norms moving forward.
If you live in the Philadelphia or Kansas City metro areas, congratulations: The fact that your city made it to the Super Bowl translates to about $200 extra in your pocket.
That’s right – whether the Philadelphia Eagles or the Kansas City Chiefs win the big game on Feb. 9, both cities have scored an economic victory. Research shows that making the playoffs alone is enough to boost personal incomes in the region. And if your team wins, you and your city will get an even bigger boost.
This windfall isn’t coming from increased merchandise sales, as you might expect. Instead, the key driver is happiness. A successful season lifts fans’ moods, which leads – indirectly – to greater spending and productivity.
In a study using data from the late 20th century and early 21st century, we found that when a team goes from zero to 11 wins – the typical number needed to make the playoffs – its home region sees an average per-person income rise by about US$200 over the year, adjusted for inflation. We also found that winning the Super Bowl was associated with a $33 bonus, again adjusted for inflation.
When you multiply $200 by the 6 million people who live in the Philadelphia metropolitan area and the 2 million in the Kansas City region, it comes out to a whole lot of money overall.
It’s about happiness, not jerseys
If you’ve ever been to a Super Bowl parade, you might assume that the income boost is linked to people spending more on team-related merchandise. But research shows that professional sports teams usually have a small impact on local incomes.
Even hosting the Super Bowl doesn’t seem to do that much: Our research shows that people are better off economically if their local team wins the Super Bowl than if their local area hosts one.
So if people aren’t spending more directly on the team, something else must be going on. Our work pointed to two possible explanations – both having to do with happiness.
First, we hypothesized that happier people tend to spend more. And when people spend more, that money is returned to the population through wages, so people’s incomes rise. The key here is that people are spending more on everything, not just things associated with the sports teams.
Since the football season usually finishes in December, it could be that happy parents who are fans of the local NFL team are spending more on Christmas gifts for their kids. With the Super Bowl stretching later into the winter, loved ones might get nicer flower bouquets and more chocolate for Valentine’s Day when the local team wins the Super Bowl.
Happy people – like Kansas City Chiefs coach Andy Reid, left, celebrating his team’s Super Bowl win on Feb. 11, 2024 – tend to spend more. Steph Chambers/Getty Images
The other possible path is through increased productivity. Psychology research has found that happier people are more productive. So as the season progresses and the home team keeps winning, it stands to reason that people in the area will go into work happy and work harder.
Previous research backs up this idea. For example, a 2011 study found that when the home team in Washington performs better, federal regulators are more productive. In places where private businesses dominate the local economy – which is to say, most of the rest of the U.S. – an increase in productivity would lead companies to be more profitable, which could lead to locals having higher earnings. Even nonfans see benefits when their neighbors are happier, spending more and working harder.
No matter how the Super Bowl turns out, both the Philadelphia and Kansas City metropolitan areas have already won, as both fans and nonfans in each region stand to benefit from higher incomes.
theconversation.com – Richard Forno, Teaching Professor of Computer Science and Electrical Engineering, and Assistant Director, UMBC Cybersecurity Institute, University of Maryland, Baltimore County – 2025-02-06 17:54:00
Is DOGE a cybersecurity threat? A security expert explains the dangers of violating protocols and regulations that protect government computer systems
The Department of Government Efficiency (DOGE), President Donald Trump’s special commission tasked with slashing federal spending, continues to disrupt Washington and the federal bureaucracy. According to published reports, its teams are dropping into federal agencies with a practically unlimited mandate to reform the federal government in accordance with recent executive orders.
As a 30-year cybersecurity veteran, I find the activities of DOGE thus far concerning. Its broad mandate across government, seemingly nonexistent oversight, and the apparent lack of operational competence of its employees have demonstrated that DOGE could create conditions that are ideal for cybersecurity or data privacy incidents that affect the entire nation.
Traditionally, the purpose of cybersecurity is to ensure the confidentiality and integrity of information and information systems while helping keep those systems available to those who need them. But in DOGE’s first few weeks of existence, reports indicate that its staff appears to be ignoring those principles and potentially making the federal government more vulnerable to cyber incidents.
Technical competence
Cybersecurity and information technology, like any other business function, depend on employees trained specifically for their jobs. Just as you wouldn’t let someone only qualified in first aid to perform open heart surgery, technology professionals require a baseline set of credentialed education, training and experience to ensure that the most qualified people are on the job.
Currently, the general public, federal agencies and Congress have little idea who is tinkering with the government’s critical systems. DOGE’s hiring process, including how it screens applicants for technical, operational or cybersecurity competency, as well as experience in government, is opaque. And journalists investigating the backgrounds of DOGE employees have been intimidated by the acting U.S. attorney in Washington.
DOGE has hired young people fresh out of – or still in – college or with little or no experience in government, but who reportedly have strong technical prowess. But some have questionable backgrounds for such sensitive work. And one leading DOGE staffer working at the Treasury Department has since resigned over a series of racist social media posts.
Wired’s Katie Drummond explains what the magazine’s reporters have uncovered about DOGE staffers and their activities.
According to reports, these DOGE staffers have been granted administrator-level technical access to a variety of federal systems. These include systems that process all federal payments, including Social Security, Medicare and the congressionally appropriated funds that run the government and its contracting operations.
DOGE operatives are quickly developing and deploying major software changes to very complex old systems and databases, according to reports. But given the speed of change, it’s likely that there is little formal planning or quality control involved to ensure such changes don’t break the system. Such actions run contrary to cybersecurity principles and best practices for technology management.
As a result, there’s probably no way of knowing if these changes make it easier for malware to be introduced into government systems, if sensitive data can be accessed without authorization, or if DOGE’s work is making government systems otherwise more unstable and more vulnerable.
If you don’t know what you’re doing in IT, really bad things can happen. A notable example is the failed launch of the healthcare.gov website in 2013. In the case of the Treasury Department’s systems, that’s fairly important to remember as the nation careens toward another debt-ceiling crisis and citizens look for their Social Security payments.
On Feb. 6, 2025, a federal judge ordered that DOGE staff be restricted to read-only access to the Treasury Department’s payment systems, but the legal proceedings challenging the legality of their access to government IT systems are ongoing.
DOGE email servers
DOGE’s apparent lack of cybersecurity competence is reflected in some of its first actions. DOGE installed its own email servers across the federal government to facilitate direct communication with rank-and-file employees outside official channels, disregarding time-tested best practices for cybersecurity and IT administration. A lawsuit by federal employees alleges that these systems did not undergo a security review as required by current federal cybersecurity standards.
There is an established process in the federal government to configure and deploy new systems to ensure they are stable, secure and unlikely to create cybersecurity problems. But DOGE ignored those practices, with predictable results.
For example, a journalist was able to send invitations to his newsletter to over 13,000 National Oceanic and Atmospheric Administration employees through one of these servers. In another case, the way in which employee responses to DOGE’s Fork in the Road buyout offer to federal employees are collected could easily be manipulated by someone with malicious intent – a simple social engineering attack could wrongly end a worker’s employment. And DOGE staff members reportedly are connecting their own untrusted devices to government networks, which potentially introduces new ways for cyberattackers to penetrate sensitive systems.
However, DOGE appears to be embracing creative cybersecurity practices in shielding itself. It’s reorganizing its internal communications in order to dodge Freedom of Information Act requests into its work, and it’s using cybersecurity techniques for tracking insider threats to prevent and investigate leaks of its activities.
Lacking management controls
But it’s not just technical security that DOGE is ignoring. On Feb. 2, two security officials for the U.S. Agency for International Development resisted granting a DOGE team access to sensitive financial and personnel systems until their identities and clearances were verified, in accordance with federal requirements. Instead, the officials were threatened with arrest and placed on administrative leave, and DOGE’s team gained access.
The Trump administration also has reclassified federal chief information officers, normally senior career employees with years of specialized knowledge, to be general employees subject to dismissal for political reasons. So there may well be a brain drain of IT talent in the federal government, or a constant turnover of both senior IT leadership and other technical experts. This change will almost certainly have ramifications for cybersecurity.
DOGE operatives now have direct access to the Office of Personnel Management’s database of millions of federal employees, including those with security clearances holding sensitive positions. Without oversight, this access opens up the possibilities of privacy violations, tampering with employment records, intimidation or political retribution.
Support from all levels of management is crucial to provide accountability for cybersecurity and technology management. This is especially important in the public sector, where oversight and accountability is a critical function of good democratic governance and national security. After all, if people don’t know what you’re doing, they don’t know what you’re doing wrong.
At the moment, DOGE appears to be operating with very little oversight by anyone in position willing or able to hold it responsible for its actions.
Mitigating the damage
Career federal employees trying to follow legal or cybersecurity practices for federal systems and data are now placed in a difficult position. They either capitulate to DOGE staffers’ instructions, thereby abandoning best practices and ignoring federal standards, or resist them and run the risk of being fired or disciplined.
The federal government’s vast collections of data touch every citizen and company. While government systems may not be as trustworthy as they once were, people can still take steps to protect themselves from adverse consequences of DOGE’s activities. Two good starting points are to lock your credit bureau records in case your government data is disclosed and using different logins and passwords on federal websites to conduct business.
It’s crucial for the administration, Congress and the public to recognize the cybersecurity dangers that DOGE’s activities pose and take meaningful steps to bring the organization under reasonable control and oversight.
In recent months, Americans looking for eggs have faced empty shelves in their grocery stores. The escalating threat of avian flu has forced farmers to kill millions of chickens to prevent its spread.
Nearly 70 years ago, Maurice Hilleman, an expert in influenza, also worried about finding eggs. Hilleman, however, needed eggs not for his breakfast, but to make the vaccines that were key to stopping a potential influenza pandemic.
Hilleman was born a year after the notorious 1918 influenza pandemic swept the world, killing 20 million to 100 million people. By 1957, when Hilleman began worrying about the egg supply, scientists had a significantly more sophisticated understanding of influenza than they had previously. This knowledge led them to fear that a pandemic similar to that of 1918 could easily erupt, killing millions again.
As a historian of medicine, I have always been fascinated by the key moments that halt an epidemic. Studying these moments provides some insight into how and why one outbreak may become a deadly pandemic, while another does not.
Anticipating a pandemic
Influenza is one of the most unpredictable of diseases. Each year, the virus mutates slightly in a process called antigenic drift. The greater the mutation, the less likely that your immune system will recognize and fight back against the disease.
Every now and then, the virus changes dramatically in a process called antigenic shift. When this occurs, people become even less immune, and the likelihood of disease spread dramatically increases. Hilleman knew that it was just a matter of time before the influenza virus shifted and caused a pandemic similar to the one in 1918. Exactly when that shift would occur was anyone’s guess.
The article was just eight sentences long. But Hilleman needed only the four words of the headline to become alarmed: “Hong Kong Battling Influenza.”
Within a month of learning about Hong Kong’s influenza epidemic, Hilleman had requested, obtained and tested a sample of the virus from colleagues in Asia. By May, Hilleman and his colleagues knew that Americans lacked immunity against this new version of the virus. A potential pandemic loomed.
The U.S. prioritized vaccinating military personnel over the public in 1957. Here, members of a West German Navy vessel hand over a jar of vaccine to the U.S. transport ship General Patch for 134 people sick with flu. Henry Brueggemann/AP Photo
Getting to know influenza
During the 1920s and 1930s, the American government had poured millions of dollars into influenza research. By 1944, scientists not only understood that influenza was caused by a shape-shifting virus – something they had not known in 1918 – but they had also developed a vaccine.
Antigenic drift rendered this vaccine ineffective in the 1946 flu season. Unlike the polio or smallpox vaccine, which could be administered once for lifelong protection, the influenza vaccine needed to be continually updated to be effective against an ever-changing virus.
However, Americans were not accustomed to the idea of signing up for a yearly flu shot. In fact, they were not accustomed to signing up for a flu shot, period. After seeing the devastating impact of the 1918 pandemic on the nation’s soldiers and sailors, officials prioritized protecting the military from influenza. During and after World War II, the government used the influenza vaccine for the military, not the general public.
Stopping a pandemic
In the spring of 1957, the government called for vaccine manufacturers to accelerate production of a new influenza vaccine for all Americans.
Traditionally, farmers have often culled roosters and unwanted chickens to keep their costs low. Hilleman, however, asked farmers to not cull their roosters, because vaccine manufacturers would need a huge supply of eggs to produce the vaccine before the virus fully hit the United States.
But in early June, the virus was already circulating in the U.S. The good news was that the new virus was not the killer its 1918 predecessor had been.
Hoping to create an “alert but not an alarmed public,” Surgeon General Leroy Burney and other experts discussed influenza and the need for vaccination in a widely distributed television show. The government also created short public service announcements and worked with local health organizations to encourage vaccination.
A 1957 film informing Americans how the U.S. was responding to an influenza outbreak.
Vaccination rates were, however, only “moderate” – not because Americans saw vaccination as problematic, but because they did not see influenza as a threat. Nearly 40 years had dulled memories of the 1918 pandemic, while the development of antibiotics had lessened the threat of the deadly pneumonia that can accompany influenza.
Learning from a lucky reprieve
If death and devastation defined the 1918 pandemic, luck defined the 1957 pandemic.
It was luck that Hilleman saw an article about rising rates of influenza in Asia in the popular press. It was luck that Hilleman made an early call to increase production of fertilized eggs. And it was luck that the 1957 virus did not mirror its 1918 relative’s ability to kill.
Recognizing that they had dodged a bullet in 1957, public health experts intensified their monitoring of the influenza virus during the 1960s. They also worked to improve influenza vaccines and to promote yearly vaccination. Multiple factors, such as the development of the polio vaccine as well as a growing recognition of the role vaccines played in controlling diseases, shaped the creation of an immunization-focused bureaucracy in the federal government during the 1960s.
Inoculating eggs with live virus was the first step to producing a vaccine. AP Photo
Over the past 60 years, the influenza virus has continued to drift and shift. In 1968, a shift once again caused a pandemic. In 1976and 2009, concerns that the virus had shifted led to [fears that a new pandemic loomed]. But Americans were lucky once again.
Today, few Americans remember the 1957 pandemic – the one that sputtered out before it did real damage. Yet that event left a lasting legacy in how public health experts think about and plan for future outbreaks. Assuming that the U.S. uses the medical and public health advances at its disposal, Americans are now more prepared for an influenza pandemic than our ancestors were in 1918 and in 1957.
But the virus’s unpredictability makes it impossible to know even today how it will mutate and when a pandemic will emerge.
