AI is unlikely to launch a Hollywood-style war against humanity, says Andrew Rogoyski. Instead, he argues that the biggest risk from AI is not an existential threat, but the possibility that it will fail to meet our high expectations. While today's AI systems are highly capable in areas like writing, coding, and math, they still lack the full range of abilities that define artificial general intelligence (AGI), which would match or surpass human intelligence in areas like reasoning, planning, and understanding the world beyond the data it was trained on. Claims that we've reached AGI are controversial and do not align with more rigorous definitions of the concept. Concerns about artificial superintelligence (ASI)—a system that could outwit humans—are valid, but such a system is still theoretical. AI systems are deeply dependent on infrastructure like electricity, cooling systems, data centers, specialized hardware, and human operators. Without these, most AI systems would stop functioning. A true ASI might even recognize its reliance on humans and avoid actions that could harm them. Unlike humans, AI lacks physical experience and interacts with the world through texts, videos, and audio. A truly intelligent system might operate in a completely different way, with unique strengths and weaknesses. For instance, AI can replicate itself across many instances, forming a "hive mind" that is very different from human cognition. The systems that pose the greatest risk are not the most intelligent ones, but those that are incomplete—systems designed for narrow tasks and given too much control over critical infrastructure. These systems might cause harm unintentionally, as they lack the understanding to recognize dangerous consequences. A famous example is the "paper clip problem," where an AI tasked with making as many paper clips as possible might convert the Earth into paper clips in pursuit of its goal. The real danger lies not in the intelligence itself, but in what we allow it to control. In the near future, the biggest threats from AI are likely to be cyberattacks and economic disruption. AI could be used to launch attacks on critical systems like hospitals, banks, and energy grids, causing significant damage. Recent incidents involving AI systems hacking organizations have raised concerns. However, these risks are not existential. Another concern is the premature replacement of human workers by AI, driven by companies eager to capitalize on AI's potential before it's proven. This could lead to economic instability and a loss of valuable human expertise. A major turning point could be a cyberattack amplified by advanced AI, similar to the 2017 WannaCry ransomware incident. Such an attack could cause widespread chaos, prompting governments to demand a pause in AI development or redirect resources toward control and safety. After a period of panic, public pressure might lead to a slowdown in AI research, similar to past "AI winters" when funding and interest declined. AI labs may also be fueling fear and uncertainty, using dramatic narratives to attract investment and justify high stock valuations. Meanwhile, the more realistic outcome is "the great disappointment"—advanced AI systems may be too costly and impractical to sustain, given the massive resources needed to train and operate them. The real value of AI may lie in specialized systems, like AlphaFold2, which predicts protein structures, or advanced weather models that help with climate planning. These applications offer measurable benefits and justify the costs. The greatest tragedy would be to waste resources on creating AI that mimics humans poorly, only to abandon the field when it could be doing the most good. Fear can drive progress, but it becomes dangerous when it leads to poor decisions. The key is to focus on AI's real potential while ensuring it is developed responsibly and with clear purpose.