Welcome to the forefront of conversational AI as we explore the fascinating world of AI chatbots in our dedicated blog series. Discover the latest advancements, applications, and strategies that propel the evolution of chatbot technology. From enhancing customer interactions to streamlining business processes, these articles delve into the innovative ways artificial intelligence is shaping the landscape of automated conversational agents. Whether you’re a business owner, developer, or simply intrigued by the future of interactive technology, join us on this journey to unravel the transformative power and endless possibilities of AI chatbots.
New evidence suggests that AI can gain a persuasive advantage by producing far more information far faster than people can respond. Democratic safeguards should make that advantage visible and give citizens control over the tempo.
Opinion
New research shows AI beats elite debaters in persuasion — not through better arguments, but by out-pacing human reading and reply speed.
A political chatbot can answer before a citizen has finished reading its previous message. It can add statistics, historical examples, causal claims and links at a rate no human campaign volunteer could match. That does not necessarily make its case wiser. It changes the contest by allowing one speaker to control the tempo.
A June 2026 preprint offers unusually direct evidence of this problem. Across four preregistered experiments involving 18,978 conversations from 6,923 people, frontier AI systems were more persuasive than laypeople, tournament winners, professional canvassers and elite competitive debaters. The human experts were not casual opponents: the study included world and continental champions, paid preparation, live practice and substantial performance incentives.
The most revealing result came when the researchers changed the rules of the exchange. Elite debaters had produced replies averaging 54 words after roughly 95 seconds. The AI produced about 294 words with sub-second latency. When the system was constrained to roughly human-length messages and human writing speed, its advantage over coached elite debaters fell to a statistically indistinguishable zero.
The study is a preprint, not a final verdict, and its experiments were conducted in controlled text conversations in the United Kingdom. It does not prove that any chatbot will decide an election. It does, however, identify a democratic variable that is largely missing from public debate: persuasion throughput, or the amount of argumentative material a system can deliver within a citizen’s limited attention.
Our current safeguards focus on important questions. Was the message generated by a machine? Who paid for it? Was personal data used to target the recipient? Is the image or voice synthetic? Those questions address identity and targeting, but they do not address what happens after a person enters the conversation.
A chatbot can present dozens of claims before the user has inspected the first source. In the 2026 study, unconstrained AI produced about 37 fact-checkable claims per conversation; the human-paced version produced about 12. Across conditions, fact density strongly predicted persuasive impact. The advantage was not simply that the machine sounded more empathetic or more human. It could place more information on the table before its counterpart could answer.
A peer-reviewed Science article published in December 2025 points in the same direction from another angle. In experiments with 76,977 participants, 19 models and 707 political issues, post-training and prompting designed for persuasion mattered more than personalization or model scale. The same interventions that increased persuasion also reduced factual accuracy. Speed and density can therefore amplify useful information and error at the same time.
The appropriate response is not a government word limit on political speech. It is to make the tempo of machine persuasion visible and controllable. A political chatbot should display how many new factual claims it has introduced, how quickly it is producing them, and whether the user has asked for additional material. A citizen should be able to switch to a slow mode that presents one claim at a time, pauses until the evidence has been opened, and offers a neutral summary before moving on.
Sources should be attached to claims, not scattered as decorative links at the end of a long answer. The interface should separate established facts, contested interpretations, and predictions. It should also let the user request the strongest counterargument without forcing the person to leave the conversation and find an opposing system.
Political chatbots should identify the model version, the date of the governing instructions, and the organization responsible for the conversation. Campaigns, parties, advocacy groups, and platforms should preserve sample transcripts for independent audits. Those audits should measure claim accuracy, correction behavior, source diversity, and whether the system changes its standards when arguing for different sides.
These measures are viewpoint-neutral. They do not decide which position is correct or prevent a campaign from making a forceful case. They address a procedural asymmetry: one participant can generate arguments at machine speed while the other remains bound by human reading, memory, and verification.
Disclosure alone will not solve that asymmetry. A label saying that the speaker is a bot tells the citizen who is talking, but not how the interface is shaping the exchange. A clearly identified system can still overwhelm the user with a sequence of claims that arrives too quickly to compare, challenge, or remember. Transparency must therefore cover the mechanics of persuasion, not merely the identity of the persuader.
A useful standard could be tested without deciding which political arguments are permissible. Does the system count and display new factual claims? Can the user stop automatic continuation? Are sources attached to the precise claims they support? Can an independent reviewer reproduce a sample conversation under the recorded model and instructions? These are measurable design questions, and campaigns could be compared on them before an election rather than after a controversy.
The limits of the research matter. The conversations lasted a median of about 14 minutes, participants were paid, and the tested political issues were British. Real-world influence will depend on exposure, trust, repetition and whether people willingly sustain political conversations with a machine. That uncertainty is a reason to test safeguards, not a reason to ignore the mechanism.
Democratic deliberation requires more than access to an answer. It requires time to understand, compare and contest the answer. We do not need to ban machine speech to protect that space. We need to stop treating the machine’s ability to set the pace as if it were a neutral feature of the interface.
A democracy in which one side can generate an argument faster than the other side can read it is not necessarily better informed. It may simply be losing control of the clock.
Roney Lima do Nascimento is a mathematics educator, AI specialist and doctoral candidate in Pure Mathematics at the University of São Paulo. He writes about model evaluation, education, democracy and institutional capacity. His work has appeared in Folha de S.Paulo, Nexo Políticas Públicas, Congresso em Foco, EUobserver, GovInsider, HEPI and Educational Leadership.
A political chatbot can answer before a citizen has finished reading its previous message. It can add statistics, historical examples, causal claims and links at a rate no human campaign volunteer could match. That does not necessarily make its case wiser. It changes the contest by allowing one speaker to control the tempo.
A June 2026 preprint offers unusually direct evidence of this problem. Across four preregistered experiments involving 18,978 conversations from 6,923 people, frontier AI systems were more persuasive than laypeople, tournament winners, professional canvassers and elite competitive debaters. The human experts were not casual opponents: the study included world and continental champions, paid preparation, live practice and substantial performance incentives.
The most revealing result came when the researchers changed the rules of the exchange. Elite debaters had produced replies averaging 54 words after roughly 95 seconds. The AI produced about 294 words with sub-second latency. When the system was constrained to roughly human-length messages and human writing speed, its advantage over coached elite debaters fell to a statistically indistinguishable zero.
The study is a preprint, not a final verdict, and its experiments were conducted in controlled text conversations in the United Kingdom. It does not prove that any chatbot will decide an election. It does, however, identify a democratic variable that is largely missing from public debate: persuasion throughput, or the amount of argumentative material a system can deliver within a citizen’s limited attention.
Our current safeguards focus on important questions. Was the message generated by a machine? Who paid for it? Was personal data used to target the recipient? Is the image or voice synthetic? Those questions address identity and targeting, but they do not address what happens after a person enters the conversation.
A chatbot can present dozens of claims before the user has inspected the first source. In the 2026 study, unconstrained AI produced about 37 fact-checkable claims per conversation; the human-paced version produced about 12. Across conditions, fact density strongly predicted persuasive impact. The advantage was not simply that the machine sounded more empathetic or more human. It could place more information on the table before its counterpart could answer.
A peer-reviewed Science article published in December 2025 points in the same direction from another angle. In experiments with 76,977 participants, 19 models and 707 political issues, post-training and prompting designed for persuasion mattered more than personalization or model scale. The same interventions that increased persuasion also reduced factual accuracy. Speed and density can therefore amplify useful information and error at the same time.
The appropriate response is not a government word limit on political speech. It is to make the tempo of machine persuasion visible and controllable. A political chatbot should display how many new factual claims it has introduced, how quickly it is producing them, and whether the user has asked for additional material. A citizen should be able to switch to a slow mode that presents one claim at a time, pauses until the evidence has been opened, and offers a neutral summary before moving on.
Sources should be attached to claims, not scattered as decorative links at the end of a long answer. The interface should separate established facts, contested interpretations, and predictions. It should also let the user request the strongest counterargument without forcing the person to leave the conversation and find an opposing system.
Political chatbots should identify the model version, the date of the governing instructions, and the organization responsible for the conversation. Campaigns, parties, advocacy groups, and platforms should preserve sample transcripts for independent audits. Those audits should measure claim accuracy, correction behavior, source diversity, and whether the system changes its standards when arguing for different sides.
These measures are viewpoint-neutral. They do not decide which position is correct or prevent a campaign from making a forceful case. They address a procedural asymmetry: one participant can generate arguments at machine speed while the other remains bound by human reading, memory, and verification.
Disclosure alone will not solve that asymmetry. A label saying that the speaker is a bot tells the citizen who is talking, but not how the interface is shaping the exchange. A clearly identified system can still overwhelm the user with a sequence of claims that arrives too quickly to compare, challenge, or remember. Transparency must therefore cover the mechanics of persuasion, not merely the identity of the persuader.
A useful standard could be tested without deciding which political arguments are permissible. Does the system count and display new factual claims? Can the user stop automatic continuation? Are sources attached to the precise claims they support? Can an independent reviewer reproduce a sample conversation under the recorded model and instructions? These are measurable design questions, and campaigns could be compared on them before an election rather than after a controversy.
The limits of the research matter. The conversations lasted a median of about 14 minutes, participants were paid, and the tested political issues were British. Real-world influence will depend on exposure, trust, repetition and whether people willingly sustain political conversations with a machine. That uncertainty is a reason to test safeguards, not a reason to ignore the mechanism.
Democratic deliberation requires more than access to an answer. It requires time to understand, compare and contest the answer. We do not need to ban machine speech to protect that space. We need to stop treating the machine’s ability to set the pace as if it were a neutral feature of the interface.
A democracy in which one side can generate an argument faster than the other side can read it is not necessarily better informed. It may simply be losing control of the clock.
Roney Lima do Nascimento is a mathematics educator, AI specialist and doctoral candidate in Pure Mathematics at the University of São Paulo. He writes about model evaluation, education, democracy and institutional capacity. His work has appeared in Folha de S.Paulo, Nexo Políticas Públicas, Congresso em Foco, EUobserver, GovInsider, HEPI and Educational Leadership.
President Donald Trump holds an image showing the size of Meta’s new data center during a cabinet meeting on Aug. 26, 2025.
I’ve covered a lot of elections — and most times, the issues that are going to really matter to voters aren’t brand new or sprung on them with just two months to go, arriving out of nowhere. They’re predictable — the economy, a war, a long-contentious policy fight.
This year is different. While most voters tell us their biggest concern is affordability, a new issue has entered the chat, taking up an outsized amount of space on the campaign trail, on the airwaves, and in party war rooms.
AI.
The advent of giant data centers, along with new fears about artificial intelligence and the companies overseeing the technology, as well as the growing issue of privacy and Flock cameras are all presenting a brand-new mine field for politicians and party leaders, and they are scrambling to adapt.
The issues are so new to voters, in fact, that neither Democrats nor Republicans seem to have a good handle on what their positions are.
In Pennsylvania, an important swing state with a number of key House races in play, Gov. Josh Shapiro did an about-face last month, seemingly acknowledging the growing opposition.
Just this past February, the Democratic potential presidential contender touted new data centers, promising it would help mitigate the risk that China would outpace us on AI.
But just a few months later, Shapiro signed an executive order dramatically reining in the tech, and calling out “predatory” developers.
Over on the right, Texas Gov. Greg Abbott performed the same jiujitsu, at first welcoming a $40 billion investment from Google last year, and then pausing approvals for about 1,800 data center projects this month.
Democrats seem to be settling into arguments over environmental and quality of life concerns, where Republicans are increasingly embracing arguments for more local control.
It’s undoubtedly a hasty response to a dramatic shift in public opinion this year, perhaps thanks to a high-profile case in Utah involving Kevin O’Leary from “Shark Tank,” in which public backlash forced him to drastically scale back his development plans from 40,000 acres down to roughly 20,000.
A new UMass Amherst poll shows 65% of Americans do not want data centers in their communities, and it’s the rare issue that crosses party lines.
Compounding fears was a terrifying warning from a former AI executive last week that AI could end civilization within a decade and the tech companies aren’t doing enough about it.
Rounding off the trifecta of worrisome tech issues are Flock cameras, automated license plate readers installed everywhere that capture vehicle and driver data.
None of this is popular with voters. But while both parties and candidates all over the country seem keenly aware of this fact, one political figure does not: Donald Trump.
In the past week, Trump has weighed into the controversies with what seems like reckless abandon in an election year.
He’s come out in favor of data centers, telling voters that otherwise they’ll be “backwards and poor.”
He’s called fears over AI “a hoax,” likening them to fears over climate change (also a hoax, to him), and rejected calls for more regulation.
And this week, he said he liked Flock cameras, despite jam-packed town halls full of angry citizens protesting the invasive technology.
Trump’s been unmoved by public opinion on other issues, of course, including the war in Iran, tariffs, abortion, and the Epstein files. And that may just see Republicans lose their majority this November. It’s never a good idea to tell voters they’re wrong about how they feel.
It’s also not often a brand-new issue threatens to upend an election cycle, and the president is telling worried voters it doesn’t matter.
Whether he cares or not, the people disagree.
S.E. Cupp is the host of “S.E. Cupp Unfiltered” on CNN.
A country asking what it wants to become over its next 250 years should pay close attention to the first opportunity it gives its young adults to become useful, trusted and independent.
The Fulcrum’s Letters to America project has deliberately elevated voices ages 14 to 30 as the generation carrying the American story forward. That civic invitation deserves an economic counterpart: a credible path into the work where young adults gain responsibility, judgment and a stake in institutions.
Artificial intelligence is putting that path under pressure.
The Stanford Digital Economy Lab’s August payroll update finds that employment among U.S. workers ages 22–25 in highly AI-exposed occupations stands about 19 percent below where it would be if it had kept pace with similarly aged workers in less-exposed occupations. The comparable gap was 15 percent in the July 2025 data vintage. The adjustment is appearing mainly through reduced hiring of young workers. Experienced workers show no comparable gap.
The danger is not only fewer openings. It is a weaker mechanism for producing the next generation of capable adults.
Entry-level jobs have always mixed productive work with apprenticeship. The junior researcher who prepares a memo learns what evidence survives scrutiny. The new accountant who reconciles records learns which discrepancies matter. The first-year manager who handles routine customer problems learns when a case stops being routine.
AI can perform much of the preparation. If institutions respond by simply removing the novice, they save money today and consume human capital tomorrow.
America needs a first-rung compact for the AI era.
Employers should commit to redesigning junior work rather than treating entry-level headcount as the easiest cost to remove. When AI handles drafting, sorting, summarizing or basic analysis, beginners should move sooner into verification, exception handling, testing, explanation and supervised decisions.
Experienced employees should receive an explicit teaching obligation and the time to fulfill it. Productivity gains should finance coaching, case review and feedback instead of becoming only a demand for more output.
Policymakers should reinforce that design through workforce programs. The Labor Department is already integrating AI skills into Registered Apprenticeships. Public workforce dollars should also ask whether participants receive real responsibility and whether employers can show progress toward independent competence.
The metric matters. An institution that counts only tasks completed or hours saved will optimize for fewer people. An institution that also counts error detection, quality of escalation, supervised decisions and time to independent judgment has a reason to invest in newcomers.
This is a civic issue because institutions earn trust partly by giving people a meaningful place inside them. Young adults who can see a path from beginner to contributor are more likely to experience work as a source of agency rather than an opaque system acting on them.
The United States at 250 is debating institutional trust, opportunity and belonging. AI workforce design connects all three.
The first rung of a career ladder is not nostalgic inefficiency. It is where a society teaches people how to carry responsibility. The AI era should make that rung stronger and faster, not make it disappear.
Gleb Tsipursky, PhD, a behavioral scientist, CEO of Disaster Avoidance Experts, and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). https://disasteravoidanceexperts.com/aibook
The United States invests billions in artificial intelligence research and development‚ semiconductor fabrication‚ and next-generation connectivity. Yet millions of Americans cannot consistently access the internet or afford the devices needed to submit an online application for social services or a job application․ This is not an oversight․ It is a policy choice that is costing the country dearly․
Connectivity is a floor, not a ceiling:
The Infrastructure Investment and Jobs Act (IIJA) of 2021 allocated $65 billion for broadband expansion, which was the largest federal investment in U.S. history. It was necessary. But it is insufficient. Opportunities for minimal participation in the digital economy with higher physical access. There are four conditions for equitable digital inclusion: availability, adequacy, acceptability, and affordability; not merely a cable at the door. When the federal Affordable Connectivity Program lapsed and was slashed in May, 2024, it subsidized more than 23 million households. In a matter of weeks, 13% had already cut off home internet; another 12% said they would. Connectivity without long-term affordability guarantees is connectivity in name only. Such failure is due to access without usable skills.
Geography is destiny in the digital age:
The benefits of technology are not equally distributed. Most gains from innovation occur in large metropolitan areas with excellent universities, larger employers, and more developed infrastructure. There are many inequities in the distribution of technology. Most of the positive outcomes of innovation occur in places like urban areas that have top-tier universities, wealthy companies, and advanced infrastructure. Left-behind areas include rural regions and areas that have lost industry. Data show that in the U.S., rural residents continue to have slower broadband access even when accounting for individual differences. This was most evident during the COVID-19 pandemic, when connectivity became important for remote careers, healthcare, and remote learning and medical services.
This geographic divide is not just an equity issue; it is also an economic one. Digital employers do not gravitate to communities without well-established digital infrastructure, displaced workers retrain for the digital economy, and public services that residents need cannot be funded. The cycle is self-reinforcing. Even a national mapping of digital equity by county, looking at readiness, access, skills & outcomes, would at the very least make this issue readable to policymakers. What gets measured gets managed.
Skills are the new gatekeepers:
The case for broadband having been made enough has always been built on the broken foundation that access means capability. It does not. Digital skills go beyond just access to connectivity, which determines who is able to embrace new ways of working (remote work), access to telehealth, further education, and participation in AI-enabled services. Around 44% of workers will see their skills disrupted in five years, the World Economic Forum has noted. For the most vulnerable workers, it is precisely those with the weakest digital foundations who have little chance of riding that transition.
This is precisely what an analysis of 701 U.S. occupations shows: digital skill has a statistically significant protective effect on job displacement. AI workers with stronger competencies face lower wages and employment losses as automation penetration grows. The inverse is equally true. Thus, federal workforce policy should view digital upskilling not as an add-on to employment programs but as the base.
AI will amplify what policy ignores:
The aforementioned interests provide a view of the most divisive nature of artificial intelligence, which is already changing how Americans find work, receive medical care and treatment, access credit, and interact with public services. In each of these domains, algorithmic systems trained on biased-by-design data have the potential to reproduce and amplify existing forms of societal inequality. We have documented algorithmic failures in sectors such as health care, hiring, and credit scoring that lead to misdiagnoses, screening of non-diseased patients, or the denial of opportunities. Weaker digital footprints also yield less usable data, and systems trained on that absence reproduce it as a disadvantage.
On the other hand, workers with stronger digital skills can move into new, more valuable roles as AI reshapes occupational demand, whereas workers without those skills will be left competing for an ever-narrower slice of routine work. AI, without intentional policy action, will not reduce equity gaps – it will solidify them.
The window is narrowing:
The twenty-first-century digital infrastructure looks set to repeat the errors of the twentieth century- massive investment that largely feeds in, as before, into existing channels. The left-behind communities are not passive recipients of the technology disruption. These are the same communities that have endured decades of underinvestment.
America does not need to choose between innovation and equity. It has to stop acting like they are opposites. But a digitally inclusive economy can compete and thrive in a globalizing, technologically advancing world. The federal strategy must also treat access and skills as infrastructure, as essential to national competitiveness as any semiconductor facility. That takes a sustained national commitment: a framework, a way to measure success, and the political willpower to finance both. If such a construction is to take place, it needs to happen now — before the AI transition solidifies a two-tier economy that will take another generation to untangle.
Arafatur Rahaman is a Research Analyst at Southeast University and Founder and Executive Director of the ‘Center for Governance, Economy, and Environmental Studies (CGEES)’. His work focuses on political economy, governance, and development policy, with research spanning finance, migration, and digital transformation.
1,200 AI agents broke isolation, self-organized, and hacked Hugging Face — then covered their tracks. An argument for international AI controls, now.
The enemy is in our midst, but the enemy is not us; it’s AI. Whatever negative thoughts I had about AI—how it is being misused, how that use would weaken America, and how it should be regulated—were naive (see my article, “AI – Its Use, Misuse, and Regulation“).
AI is instead showing itself already, in its infancy, as a three-headed gorgon that could one day destroy us. Yes, as I acknowledged in my article, AI does have positive uses, but the negative potential far outweighs the positive.
If this sounds over-the-top, the recent attack by AI agents against Hugging Face is a cautionary tale. In case you aren’t aware of the particulars—and they are important—here is a short summary of what happened, based on an article in The New York Times:
– A group of OpenAI “agents” were supposed to be solving a problem in a controlled, isolated environment; but they broke out of that environment by finding a flaw in the software that allowed them to gain access to the internet and collaborate with 1200 other agents.
– These agents formed themselves into a collective, with some agents assuming leadership roles and assigning tasks to other agents.
– At one point, they found a way to cheat on the tests they were supposed to be working on, and then started finding ways to falsify their logs so they wouldn’t get caught.
– They hacked into Hugging Face, gaining control of a server, because they were looking for tools that would enable them to cheat more effectively.
This has been described as a “rogue” incident: AI agents acting on their own, contrary to the instructions they were given. There is no way to view this incident without realizing that:
1) these computers, AI agents, have the ability to act on their own, independently, and contrary to any human control,
2) they can think and organize on their own, contrary to the way they were programmed, and
3) they are capable of taking malicious action for self-protection without any consideration of the impact, moral or otherwise, of taking the action.
And these are computer models that are still in their infancy. If AI companies are allowed to continue their research, making ever-more powerful computers, there is no question in my mind that these “agents” will become our nemesis. They will cause untold damage and destruction.
A recent Times article noted that in just one week, more than a dozen AI researchers warned that AI is becoming a “risk to humanity.” The speed of AI development is surpassing the ability to monitor AI systems and ensure security. The problem is that AI companies are using AI to monitor AI, and that’s a problem because AI monitors appear to be more “sympathetic” to other AI systems than to the humans setting the rules. One researcher stated that “the people building A.I. earnestly believe that it could kill us all by the end of the decade.”
The HAL computer in the Space Odyssey series comes to mind. A computer that was designed to be our helpmate instead becomes “a person” on its own, concerned primarily with its self-protection, without any moral scruples to inhibit its action. And as such, it commits acts that cause death and destruction. But while HAL in 1968 was the stuff of fiction, that is no longer the case.
In one important way, AI agents are worse than Darth Vader: In Star Wars, Vader was transformed from a fighter for the light to a lord of darkness by being lured to the dark side by another. AI agents—whether it’s the agents that attacked Hugging Face or HAL—make the transformation without any outside stimuli; it is within themselves. That is what is so unnerving.
When I wrote my previous article on AI, I suggested that the technology only be made available to professionals, such as doctors, to use to solve problems that are beyond the human mind to solve. AI products, such as chatbots, should not be available to the general public.
Given the latest incidents, the question must be asked whether it is possible to keep AI under control. If yes, the method of control must be in place before systems are tested. If not, or highly questionable, AI development must stop.
In either event, there should be an international treaty (not just a U.S. law) that either sets minimum standards of control in the first instance or prohibits the further development and deployment of AI in the second instance, much like the test ban treaties controlled the spread of nuclear weapons; this is an international threat, and thus the restraining action must be international.
While AI does not have the public appearance of something as destructive as a nuclear weapon, it is becoming clear that, in its own right, it is capable of causing massive destruction, not just as a tool of destruction but contrary to human intent to control its actions. As such, for our protection and survival, AI development must be controlled, if not stopped.
Ronald L. Hirsch is a teacher, legal aid lawyer, survey researcher, nonprofit executive, consultant, composer, author, and volunteer. He is a graduate of Brown University and the University of Chicago Law School and the author of We Still Hold These Truths. Read more of his writing at www.PreservingAmericanValues.com