Why is AI so bad at design?

演讲者首先承认 Linear 的网站是卓越设计和品味的典范，以至于去年被许多新网站广泛模仿。尽管这种设计值得称赞，但演讲者认为，一个仅仅复制 Linear 网站的模型，即使看起来很先进，也不代表人工智能能力上的真正“惊人飞跃”。这为更深入地讨论设计与软件工程之间的根本区别，特别是新颖性和抽象所扮演的角色，奠定了基础。 据演讲者所述，新颖性在设计中的重要性远超在软件工程中的重要性。在软件工程中，人们偏好于“过度依赖已知模式”——为了可靠性、效率和可维护性，依赖既定、可预测的解决方案。这造就了基于经验证的方法论的健壮系统。然而，设计遵循着不同的原则；它需要“随机性和新颖性”的元素。优秀的设计不仅仅关乎功能；它关乎创造新鲜、引人入胜、有时出乎意料的用户体验，这些体验能够脱颖而出并在情感层面产生共鸣。这种对新颖性和原创性的内在需求，正是当前人工智能模型经常力不从心之处，它们能够模仿现有风格，但却难以生成真正新颖且有意义的设计。 演讲者阐述道，核心挑战在于理解和操作“抽象概念”，而不仅仅是表面层面的美学。为了说明这一点，他使用了公司品牌重塑的类比。这种工作的“浅层版本”将涉及对数百个独立元素进行费力的、逐组件的更新——例如，改变263个不同组件的颜色、字体或布局。这项任务即使是当前的自动化工具或基本人工智能也可能处理，因为它主要涉及对视觉属性的直接操作。 然而，品牌重塑的“深层版本”超越了这些表面变化。它关乎理解设计选择背后潜在的“语义”——即其含义和意图。它关乎理解为什么两个表面上不同的元素实际上可能共享一个共同的潜在目的或交互模式。演讲者举例说：“它们都在列表中，具有这种风格，向用户传达这种交互模式。”这指向了一种更深层次的概念理解，其中设计元素不仅仅是视觉资产，而是传达特定的功能、层级或行为。它关乎交互模式的一致性、认知负荷，以及向用户提供的直观指导，而不考虑表面风格变化。 这种基于深刻语义联系和抽象设计原则进行理解和生成的能力，而不仅仅是复制视觉模板，仍然是当前技术的一个重大障碍。演讲者最后指出，这种对设计抽象的“深层”理解，以及从中生成新颖性的能力，对当代人工智能和设计工具来说，仍然感觉“有点遥不可及”。这凸显了人工智能在创意领域发展的一个前沿，其目标不仅仅是复制，而是根植于语义理解的真正创新。

The speaker begins by acknowledging Linear's website as a prime example of outstanding design and taste, so much so that it was widely imitated by new websites last year. While such a design is laudable, the speaker posits that a model merely replicating Linear's website, though seemingly advanced, would not represent a true "incredible leap" in AI capabilities. This sets the stage for a deeper discussion about the fundamental differences between design and software engineering, particularly concerning the role of novelty and abstraction. According to the speaker, novelty holds significantly greater importance in design than it does in software engineering. In software engineering, the preference is to "over-index known patterns" – to rely on established, predictable solutions for reliability, efficiency, and maintainability. This leads to robust systems built on proven methodologies. Design, however, thrives on a different principle; it necessitates an "element of randomness and novelty." Great design isn't just about functionality; it's about creating fresh, engaging, and sometimes unexpected user experiences that stand out and connect on an emotional level. This inherent need for newness and originality is where current AI models often fall short, capable of mimicking existing styles but struggling to generate genuinely novel and meaningful designs. The core challenge, the speaker elaborates, lies in understanding and manipulating "abstractions" rather than just surface-level aesthetics. To illustrate this, he uses the analogy of a company rebrand. The "shallow version" of such an undertaking would involve the laborious, component-by-component update of hundreds of individual elements – for example, changing colors, fonts, or layouts across 263 distinct components. This is a task that even current automation tools or basic AI could potentially handle, as it primarily involves direct manipulation of visual attributes. However, the "deep version" of a rebrand transcends these superficial changes. It's about comprehending the underlying "semantics" – the meaning and intent – behind design choices. It's about understanding *why* two seemingly different elements might actually share a common underlying purpose or interaction pattern. The speaker gives an example: "They're both in lists that have this style that convey this interaction pattern to the user." This points to a deeper conceptual understanding where design elements are not just visual assets but communicate specific functionalities, hierarchies, or behaviors. It's about the consistency of interaction patterns, the cognitive load, and the intuitive guidance provided to the user, irrespective of superficial style changes. This ability to grasp and generate based on these profound semantic connections and abstract design principles, rather than just copying visual templates, remains a significant hurdle for current technology. The speaker concludes by noting that this "deep" understanding of design abstractions, and the ability to generate novelty from them, still feels "a little bit out of reach" for contemporary AI and design tools. This highlights a frontier for AI development in creative domains, where the goal isn't just replication, but true innovation rooted in semantic understanding.