The greatest threat posed by artificial intelligence is not that machines will become capable of doing everything humans can do. It is that society may become so focused on what machines can accomplish that we stop asking what their success is costing humanity. I argue that artificial intelligence will ultimately do more harm than good, not because it lacks extraordinary potential, but because its rapid expansion threatens to fundamentally reshape economic security, human independence, privacy, accountability, and the distribution of power in ways that may be extraordinarily difficult to reverse. And here is the critical distinction: AI does not have to malfunction to cause serious harm. It can operate exactly as designed and still produce devastating social consequences. Imagine a corporation replacing thousands of employees with highly efficient AI systems. From the corporation’s perspective, productivity increases and expenses decrease. The technology has succeeded. But what happens to the displaced workers, their families, and the communities that depended on their income? What happens when entry-level opportunities disappear, leaving future generations without pathways to develop professional expertise? Economic efficiency is not synonymous with human prosperity. A technology can create enormous wealth while concentrating that wealth among those who already possess the greatest resources. Now consider the consequences for truth itself. AI makes it increasingly possible to manufacture convincing images, voices, videos, and written information at enormous scale. The danger is not merely that people will believe false information. It is that people may begin doubting authentic evidence as well. When both truth and deception become easier to challenge, public trust becomes increasingly fragile. And then there is the question of human autonomy. As AI systems become integrated into education, healthcare, employment, and government, individuals may gradually surrender decisions they once made independently. Over time, dependence can weaken expertise, critical thinking, and the ability to challenge automated conclusions. But perhaps the most serious concern is the concentration of power. The organizations controlling advanced AI systems can acquire extraordinary influence over information, economic opportunity, surveillance, and institutional decision making. Even without malicious intentions, the incentives driving rapid deployment may prioritize competitive advantage and financial returns over public accountability. My opponent may point to AI-assisted medical discoveries, improved accessibility, personalized education, and scientific breakthroughs. Those benefits are real possibilities, and some are already being realized. I do not dispute them. But an impressive achievement is not the same as a positive overall outcome. A medical breakthrough does not automatically compensate workers who lose their livelihoods. Greater productivity does not restore compromised privacy. Educational convenience does not necessarily preserve independent learning. And the promise of future regulation does not guarantee protection against harms occurring today. The most important issue is how benefits and harms are distributed. Those who profit from AI’s expansion may not be the same people who suffer its consequences. The gains may be concentrated, while the risks are imposed across entire populations, including people who never consented to the systems affecting their lives. And unlike many previous technological changes, AI can influence multiple foundations of society simultaneously: employment, education, communication, healthcare, security, and public trust. Problems in one area can amplify problems in another, creating consequences that cannot be evaluated by examining isolated success stories. My opponent may argue that humanity has always adapted to technological progress. But adaptation is not a guarantee of fairness, stability, or well-being. History does not establish that every technological transformation produces more benefits than harm, especially for the generations required to endure its disruption. The burden of this debate is not to prove that AI can do extraordinary things. We already know it can. The burden is to establish whether those achievements will outweigh the economic insecurity, concentrated power, diminished autonomy, and systemic vulnerabilities that widespread adoption may create. We must stop confusing the advancement of technology with the advancement of humanity. They are not automatically the same. If artificial intelligence makes corporations wealthier but workers less secure, information more abundant but truth less trustworthy, and decisions faster but human beings less empowered, then we have not necessarily created a better society. We have created a more efficient one at an extraordinary human cost. And ultimately, the measure of progress is not how powerful our machines become. It is how much freedom, dignity, security, and control human beings retain as those machines become more powerful. That is why I maintain that artificial intelligence, despite its remarkable potential, may ultimately do more harm than good.

My opponent argues that AI’s benefits are measurable while its harms are largely hypothetical. But that claim is factually incomplete, and it weakens the foundation of their entire argument. Let’s begin with their strongest example: cancer detection. I acknowledge that AI-assisted screening can improve outcomes. In fact, a randomized Swedish mammography study involving approximately 106,000 women found higher cancer detection rates and substantially reduced radiologists’ screening workloads when AI was used alongside clinical expertise. But notice what that evidence actually demonstrates. A particular AI system, used for a specific medical purpose, under controlled conditions and with professional oversight, produced measurable benefits. That is evidence supporting a particular application. It is not evidence that artificial intelligence, across every sector of society, will produce more good than harm. My opponent is taking a specific success and extending its conclusion far beyond what the evidence establishes. Now let’s examine their claim that AI’s harms are hypothetical. According to the FBI’s 2025 Internet Crime Report, complaints involving AI-related information were associated with more than $893 million in reported losses. Those are not imagined future consequences. They are documented financial losses involving real victims. The International Labour Organization has also identified substantial occupational exposure to generative AI. Exposure does not automatically mean unemployment, and I would never claim otherwise. But it establishes that AI’s potential disruption to employment is a serious economic concern requiring more than the assumption that new opportunities will inevitably compensate for those lost. My opponent compares AI to the printing press. But that analogy overlooks a critical difference. The printing press expanded the reproduction and distribution of information. AI can generate convincing synthetic information, imitate identities, automate decisions, and operate across multiple sectors at enormous scale. Historical examples of beneficial innovation cannot establish that every future technology will have a positive overall outcome. Their argument about fraud prevention also deserves closer examination. Yes, AI can identify suspicious patterns and assist investigators. But the ability to detect fraud does not establish that AI prevents more fraud than it enables. Both capabilities can expand simultaneously. Without comparative evidence, claiming that one outweighs the other is an assumption, not a demonstrated conclusion. And when my opponent says regulators can simply target misuse, they overlook something central to my position: not all harmful consequences arise from misuse. An employer replacing workers with AI may be acting legally and using the technology exactly as intended. A company automating customer service may improve efficiency while reducing access to meaningful human assistance. A school encouraging widespread AI dependence may unintentionally undermine opportunities for students to develop independent skills. These outcomes cannot all be solved by prohibiting malicious behavior because they are not necessarily the result of malicious behavior. My opponent also presents a false choice between embracing AI and accepting slower diagnoses, inferior forecasting, and unnecessary manual work. I have never argued that society must reject beneficial medical technology or preserve inefficiency. My argument concerns whether AI’s widespread expansion, including its economic and social consequences, will ultimately leave humanity better or worse off. Responsible use of a proven diagnostic tool does not require unquestioning acceptance of every other AI application. Most importantly, my opponent correctly states that this debate concerns net impact. But that standard applies equally to their argument. Demonstrating benefits does not establish a positive net outcome any more than demonstrating harms automatically establishes a negative one. A meaningful comparison must examine the scale, severity, distribution, and reversibility of both. My opponent has demonstrated that AI can improve specific medical procedures and assist with fraud detection. I accept those findings. But they have not demonstrated that these benefits outweigh the documented harms, the potential displacement of economic opportunity, the erosion of human independence, and the concentration of technological power. The issue is not whether AI is capable of doing good. The issue is whether society can ensure that its growing power serves humanity rather than making humanity increasingly dependent, economically vulnerable, and less able to control the systems shaping its future. A technology can succeed technically while failing socially. It can increase productivity without increasing shared prosperity. It can improve individual decisions while weakening collective accountability. And it can deliver extraordinary benefits without guaranteeing that those benefits outweigh the costs. My opponent has offered evidence of AI’s capabilities. I am asking us to examine the consequences of building an entire society around them. That is the distinction their argument has yet to resolve.

My opponent’s final argument rests on a distinction that sounds persuasive but does not withstand closer examination: that AI’s benefits are already measurable while its harms remain hypothetical. That is simply not an accurate characterization of the evidence. AI-assisted cancer screening has demonstrated meaningful benefits in clinical research. I acknowledge that without hesitation. But documented benefits also coexist with documented harms, including AI-assisted financial fraud, deceptive synthetic content, and the risks associated with unreliable automated outputs. The question is not whether one side has evidence and the other has speculation. Both sides have evidence, and both must confront uncertainty about the future. My opponent repeatedly asks me to prove that the net ledger is negative. That is a fair standard, but it applies equally to their claim that the ledger is positive. They have provided examples of successful applications, not a comprehensive assessment demonstrating that AI’s benefits outweigh its economic, social, and institutional costs. A medical improvement is meaningful. Fraud detection is valuable. Greater access to expertise can change lives. But these benefits cannot simply be added together while treating economic displacement, privacy violations, misinformation, and diminished human autonomy as negligible because their long-term consequences are difficult to quantify. The absence of a complete measurement is not evidence of the absence of harm. My opponent also argues that the distribution of AI’s benefits is merely a policy problem rather than a technological problem. But that distinction overlooks how technologies operate in the real world. AI is not being introduced into an imaginary society with perfectly functioning institutions. It is being deployed within existing systems shaped by economic inequality, concentrated ownership, competitive pressure, and uneven regulatory capacity. If those conditions predictably influence how AI is developed and deployed, their consequences belong in the assessment of AI’s actual social impact. We cannot separate a technology from the institutions controlling it whenever doing so makes its effects appear more favorable. And the suggestion that regulation can resolve these problems deserves closer scrutiny. Regulation can reduce risks, but not every harmful outcome results from illegal misuse. A company may legally automate thousands of jobs. An educational institution may encourage AI dependence without violating any law. A business may concentrate enormous economic advantages through entirely lawful deployment. These are not necessarily examples of technology malfunctioning. They are examples of technology functioning successfully while potentially producing harmful consequences for society. My opponent also argues that rejecting AI would mean accepting slower medicine, weaker fraud detection, and greater dependence on existing institutions. But this presents a choice I have never advocated. Supporting carefully validated medical applications does not require endorsing unrestricted deployment across employment, education, surveillance, and public communication. We can recognize genuine benefits without assuming that every expansion of AI is beneficial. And here is the deeper issue my opponent has not resolved: the benefits of AI may be concentrated in particular applications, while its risks can spread across interconnected systems. A medical algorithm may improve diagnostic accuracy in one setting. Meanwhile, AI-generated misinformation can undermine trust across entire communities. Automation can improve efficiency for one company while weakening economic opportunities for workers elsewhere. Increased convenience can coexist with declining human independence. These outcomes are not mutually exclusive. They can occur simultaneously, and a credible assessment must account for all of them. My opponent says we do not need guarantees. I agree. No serious assessment of an emerging technology can demand certainty about the future. But we do need to examine incentives, institutional capacity, unequal exposure to harm, and the possibility of consequences that become difficult to reverse. That is not an argument based on imaginary catastrophe. It is an argument about how technological power interacts with real economic and social structures. My opponent has shown that AI can produce extraordinary benefits. I have acknowledged those benefits throughout this debate. But demonstrating that AI can improve selected outcomes is not the same as demonstrating that its widespread adoption will produce a positive net effect. They ask for a comparison, and that is exactly what this debate requires: not the best examples of AI compared against the worst features of today’s society, but the full consequences of AI adoption compared against realistic alternatives, including targeted innovation, human oversight, and investment in existing institutions. The greatest danger is not necessarily that AI will fail to accomplish what we ask of it. It is that AI may accomplish those objectives while imposing costs on people who never chose them, cannot meaningfully challenge them, and may never share in the resulting benefits. A society can become more technologically capable without becoming more economically secure, more trustworthy, or more free. The ultimate measure of artificial intelligence is not how impressive its achievements become, but whether humanity retains the security, independence, and control necessary to benefit from them. And until those broader consequences are accounted for, examples of technological success cannot settle the question of whether AI will do more good than harm. My position remains that the scale, reach, and structural consequences of AI create a serious risk that its harms will ultimately outweigh its benefits. Technological progress deserves recognition, but it does not deserve an automatic presumption of social progress.