I want to begin by acknowledging something important: artificial intelligence has enormous potential to improve our lives. It can assist medical research, improve accessibility, support education, and help people complete tasks more efficiently. I use AI myself, and I recognize its value. But recognizing that a technology can do tremendous good does not mean we should automatically assume its overall impact will be positive. I believe AI is more likely to do more harm than good because we are rapidly expanding our dependence on systems that can make serious mistakes, influence human judgment, disrupt society, and create consequences that are difficult to reverse. First, I am concerned about what happens when people begin trusting AI more than their own judgment. AI can produce answers that sound confident, professional, and convincing while being factually incorrect. Even when warnings explain that AI can make mistakes, people may overlook those warnings, especially after repeatedly receiving useful answers. This becomes particularly concerning in healthcare. Imagine a physician using AI to assist with a diagnosis. The system identifies a likely condition but overlooks an important alternative. If the physician trusts the recommendation without adequately examining the evidence, a patient could receive inappropriate treatment or experience a dangerous delay in care. This is not simply a hypothetical concern. The World Health Organization has warned about automation bias, where healthcare professionals or patients may overlook errors because they place excessive trust in automated recommendations. I am not arguing that AI should never assist physicians. I am arguing that introducing an additional source of potentially convincing errors into high-stakes decisions creates risks that cannot be dismissed simply because the technology is often helpful. Second, I believe we need to examine what widespread AI dependence could do to human reasoning and critical thinking. What happens when people increasingly rely on AI to analyze information, solve problems, write arguments, and make decisions? What happens when future generations grow up accustomed to receiving immediate answers without consistently practicing the skills necessary to reach those answers independently? A 2025 study involving 319 knowledge workers found an association between greater confidence in generative AI and reduced critical-thinking effort during AI-assisted tasks. That does not prove AI permanently damages intelligence, and I would not claim that it does. But it raises a serious concern about cognitive dependence. If we increasingly outsource our reasoning, will we remain equally capable of identifying mistakes, questioning assumptions, and making independent judgments? AI may help us accomplish tasks faster. But completing a task is not necessarily the same as understanding it. And the long-term consequences of that distinction deserve far more attention. Third, AI mistakes can create financial, professional, and personal consequences far beyond the original error. An inaccurate financial analysis can influence costly business decisions. A fabricated legal citation can undermine a court filing. An incorrect automated recommendation can affect employment, education, or access to essential services. Even when mistakes are eventually discovered, correcting them can require substantial time, money, and effort. And because AI systems can operate at enormous scale, a flawed process can potentially affect thousands of people before anyone recognizes the problem. Human beings make mistakes too. The difference is that AI can reproduce the same mistake rapidly across many decisions while creating an appearance of consistency and authority. Fourth, AI makes deception easier to produce and harder to recognize. Convincing impersonations, cloned voices, fabricated images, and misleading information can be generated at speeds and volumes that were previously difficult to achieve. The danger extends beyond individual scams. When people cannot reliably distinguish authentic information from fabricated material, trust in legitimate evidence can also deteriorate. That creates consequences for personal relationships, financial security, public institutions, and society’s ability to make informed decisions. Fifth, the economic consequences deserve serious scrutiny. AI may increase productivity and create valuable new opportunities. But productivity gains do not automatically translate into better outcomes for every worker. Some jobs may be transformed, some tasks automated, and some workers displaced or left with reduced bargaining power. I am not claiming that every job exposed to AI will disappear. The evidence does not support that conclusion. My concern is that the economic benefits may become concentrated among the companies and institutions controlling powerful AI systems, while workers and communities absorb a disproportionate share of the disruption. And there is another concern: dependence. As AI becomes embedded in healthcare, education, business, communication, and public services, what happens when systems fail, become unavailable, produce incorrect information, or are manipulated? The more essential these systems become, the more consequential their failures may be. Finally, I want to address the argument that safeguards and regulation will solve these problems. I support responsible oversight, independent verification, transparency, and meaningful human control. But safeguards are only effective when they are implemented, followed, and capable of addressing the risks they are intended to prevent. A warning that AI can be inaccurate does not guarantee that people will verify its answers. A requirement for human oversight does not guarantee that the person reviewing an automated decision has the time, expertise, or independence to challenge it. And the possibility of future regulation does not establish that society is currently prepared for the consequences of increasingly powerful AI. I am not suggesting that every risk will become a catastrophe or that AI’s benefits are insignificant. My argument is about the balance of consequences. AI can help people solve problems. It can also introduce new problems, amplify existing ones, and encourage dependence on systems that are not consistently reliable. It can make information easier to obtain while making truth harder to establish. It can increase productivity while creating economic insecurity. And it can make decisions easier while potentially weakening the independent reasoning needed to recognize when those decisions are wrong. The greatest danger may not be that AI becomes more intelligent than humans. It may be that humans become so accustomed to relying on AI that they stop exercising the judgment necessary to use it safely. A technology should not be judged solely by how much time it saves or how impressive its capabilities appear. It should also be judged by the mistakes it introduces, the dependencies it creates, the people it harms, and the long-term consequences society may struggle to reverse. I acknowledge the good AI can accomplish. But until we can demonstrate that its expanding capabilities are matched by reliable safeguards, meaningful accountability, and protection of human independence, I believe the scale of its potential harm is greater than the good we can confidently expect. That is why I am FOR the position that artificial intelligence will do more harm than good.

My opponent makes a strong point about healthcare, and I want to acknowledge it rather than argue against evidence that deserves recognition. AI-assisted colonoscopy has improved adenoma detection in multiple clinical trials, and validated diabetic retinopathy screening systems can expand access to important diagnostic services. These are meaningful advances. I would not argue that we should reject technologies that demonstrably improve patient outcomes. But I think my opponent’s argument makes a critical leap: evidence that specific AI applications can outperform existing medical practices is not evidence that artificial intelligence as a whole will do more good than harm. In fact, the medical evidence illustrates why careful distinctions matter. A systematic review of 44 randomized trials involving more than 36,000 colonoscopy cases found improved adenoma detection with AI assistance. But it did not establish a corresponding improvement in every measure of advanced neoplasia detection. And a separate randomized trial involving 1,356 participants found that AI assistance did not significantly improve its primary adenoma detection outcome overall, although benefits appeared in certain analyses. It also resulted in more intensive surveillance recommendations for some patients. That doesn’t invalidate AI-assisted colonoscopy. It demonstrates that benefits depend on the application, clinical setting, and outcomes being measured. My opponent also mentions triage systems getting patients to clinicians faster. That is a reasonable potential benefit, but faster triage is not automatically safer triage. We would need evidence that the particular system improves meaningful patient outcomes without introducing unacceptable errors or disparities. Now consider the comparison to medications. I actually agree with my opponent that drugs should be tested, monitored, and regulated rather than rejected simply because some have risks. But that analogy supports my argument for caution just as much as theirs. We do not assume that because one medication works, an entire category of treatments must be beneficial overall. We examine individual products, indications, populations, side effects, and long-term outcomes. Why should AI receive a broader presumption of safety? A carefully validated diagnostic tool operating within a defined clinical setting is fundamentally different from a general-purpose AI system used for financial decisions, employment screening, education, misinformation, or autonomous actions. We cannot take the safety evidence from one application and transfer it to every other application. My opponent also says that AI’s harms are governance failures we can reduce. Some certainly are. Better regulation, prospective testing, monitoring, and meaningful human oversight can make AI safer. But identifying a possible safeguard is not the same as demonstrating that it will work consistently in practice. Human oversight can fail when people become overly reliant on automated recommendations. Monitoring can detect some problems only after people have already been affected. And regulations may struggle to keep pace with systems deployed across different industries and jurisdictions. Furthermore, not every harm is simply a technical error awaiting correction. Job displacement, concentration of economic power, increased dependence on automated systems, and the possible erosion of independent reasoning are broader social consequences. A perfectly accurate AI system could still contribute to some of those problems. This is the distinction I believe my opponent has not addressed. I am not arguing that AI must be perfect to be beneficial. I agree that it does not. I am arguing that demonstrating benefits in selected applications does not settle the much larger question of AI’s overall consequences. We need to examine not only whether AI performs individual tasks better, but also how its widespread adoption changes human behavior, economic opportunity, accountability, privacy, and society’s ability to function independently of these systems. My opponent has established that AI can do considerable good in healthcare. I concede that point. But proving that AI can improve a colonoscopy does not prove that its expanding influence across society will produce a net benefit. The debate is not whether AI has successes. It is whether those successes are sufficient to outweigh the full range of risks and consequences. And that larger burden remains unanswered.

My opponent is correct about one thing that I want to acknowledge immediately: AI should not be compared with a perfect world. Human beings make mistakes, healthcare systems have serious shortcomings, and AI can improve outcomes in certain circumstances. I have never argued otherwise. But my opponent’s response contains a contradiction that goes directly to the heart of this debate. They argue that I have not demonstrated that AI’s harms will outweigh its benefits. That is a fair challenge. Yet they have not demonstrated the reverse. Instead, they point to successful medical applications and propose safeguards that could reduce future risks. The possibility of effective safeguards is not proof of effective safeguards, just as the possibility of harm is not proof of inevitable harm. Both sides must be held to the same evidentiary standard. Let’s examine their central comparison: AI plus oversight versus the existing healthcare system. That comparison is appropriate when evaluating a specific medical application. If a validated AI-assisted diagnostic tool improves patient outcomes under real-world conditions, then its use may be justified. I concede that completely. But this debate is not limited to whether a particular diagnostic tool improves healthcare. It asks whether artificial intelligence will do more harm than good across society. My opponent cannot establish that conclusion by selecting some of AI’s strongest medical applications while leaving its broader consequences largely unexamined. That is not a balanced comparison. It is an incomplete one. Consider their proposed solution: validate AI prospectively, monitor it, restrict dangerous uses, and withdraw systems when they fail. Those are reasonable safeguards. But the argument assumes they can be applied effectively across a technology that is already spreading through healthcare, education, employment, finance, communication, and everyday decision-making. A clinical trial can evaluate a narrowly defined medical tool. It cannot automatically establish the safety of a general-purpose AI system operating across thousands of unpredictable situations. And withdrawing a system after discovering a problem does not necessarily reverse the harm already caused. A fraudulent transfer may not be recovered. Private information exposed through an AI-related failure cannot simply be made secret again. A person denied an opportunity by an unreliable automated process may never regain that opportunity. Monitoring can identify failures. It cannot guarantee that their consequences are reversible. My opponent also correctly notes that concerns about reduced critical-thinking effort are associations rather than proof of permanent cognitive decline. I agree. I have not claimed otherwise. But the fact that a risk is not unique to AI does not make it irrelevant. Automation bias existed before modern generative AI. What matters is whether AI can intensify that problem by making automated recommendations more convincing, more accessible, and more deeply embedded in everyday decisions. And there is an important difference between helping people think and encouraging them to stop thinking independently. AI can be an extraordinary intellectual aid. It can improve accessibility, support learning, and help people accomplish things they otherwise could not. I recognize those benefits. But the same technology can also encourage people to accept polished answers without understanding how they were produced or whether they are accurate. The question is not whether AI always weakens human reasoning. It is whether widespread dependence creates risks to independent judgment that society is adequately prepared to manage. My opponent has not addressed that long-term concern. Nor have they adequately addressed AI-enabled deception, privacy risks, economic disruption, or the concentration of power among organizations controlling increasingly influential systems. Some of these problems can occur even when AI functions exactly as intended. A highly accurate system can still displace workers. An efficient platform can still concentrate economic power. A convincing generative system can still make large-scale impersonation easier. These are not all software defects that disappear when engineers improve accuracy. They are consequences of how powerful technology changes society. And that is why my opponent’s comparison to evidence-based medicine, while useful, has limits. A medication is evaluated for particular uses, benefits, and risks. We do not conclude that all medications are beneficial simply because some save lives. Similarly, successful medical AI applications demonstrate that certain uses of AI are valuable. They do not establish that the technology’s total societal impact will be positive. My opponent asks me to prove that AI will do more harm than good. That is the proposition I am defending, and I acknowledge that future outcomes cannot be established with certainty. But my position rests on a cumulative concern: AI can amplify deception, introduce consequential errors, create difficult-to-reverse harms, increase economic disruption, and encourage dependence on systems that people may not fully understand or critically evaluate. These risks are not confined to one industry. They can interact and spread across the institutions on which society depends. My opponent’s response offers promising medical successes and a framework for responsible oversight. It does not establish that those protections will be consistently effective, that broader societal harms will be adequately addressed, or that the benefits will outweigh the costs for the people bearing them. The central distinction is this: demonstrating that AI can improve an imperfect system is not the same as demonstrating that the widespread deployment of AI will improve society overall. I am not asking us to abandon beneficial AI applications. I am asking us to recognize that technological progress and human progress are not automatically the same thing. A system can become faster without becoming fairer, more efficient without becoming safer, and more capable without making the people who depend on it more independent. The strongest argument for AI is what it can help us accomplish. The strongest argument against it is what we may sacrifice, often without realizing it, as we become increasingly dependent on its capabilities. And until the safeguards, accountability, and broader social protections are demonstrated rather than merely proposed, I believe the scale and interconnected nature of those risks make the prospect of more harm than good the more compelling conclusion.