Why Physical AI Is the Next Frontier | The a16z Show

以下是内容的中文翻译： Applied Intuition 公司，正如其首席执行官 Kasar 所阐述的，是一家“实体AI公司”，致力于将智能嵌入全球十亿台机器中。这项使命超越了优化广告或生成视频等典型的数字AI应用，而是专注于物理的、移动的实体，例如汽车、卡车、坦克、无人机和重型工业机械。该公司认为，智能革命最深远的影响，以及可能最大的经济参与者，将来自于那些能够改变物理世界的公司。 Applied Intuition 的业务范围广阔，并迅速超越了其最初在自动驾驶汽车领域的焦点。虽然汽车业务目前约占其业务的30%，但大部分已经是非汽车业务，涵盖国防、建筑、采矿、农业和物流等领域。这种多样化凸显了其坚信——几乎任何移动的物理系统都可以从自主化中受益。该公司拥有超过一千名工程师，并已融资超过十亿美元，专注于提供高质量产品而非仅仅依赖销售，在全球设有18个办事处。 与数字AI相比，实体AI面临独特的挑战。虽然数字AI常利用海量的互联网数据来构建基础模型，实体AI则需要广泛的专有数据采集，这通常涉及专业的车队，并需要应对复杂的国际法规以获取独有的数据集。安全性至关重要；与智能手机应用不同，一台重达数吨的自主机器需要严格的安全协议和验证。性能限制也更加严格，要求在实际物理环境中进行实时决策。 自动驾驶技术的发展已经从模仿学习转向了端到端强化学习，合成数据在加速开发中发挥着关键作用。Kasar 指出显著的进展，列举了特斯拉的全自动驾驶（FSD）和其他制造商的先进驾驶辅助系统。预计到2030年代初，广泛普及的个人自动驾驶汽车（L2++）将变得司空见惯，甚至免费，而Robotaxi（自动驾驶出租车）预计将在2030年至2033年间在主要城市投入使用。该公司将推广速度较慢归因于将这种先进、对安全至关重要的技术推向大众市场固有的成本和复杂性，并对比了 Waymo 定制化的、地理围栏式的方法与特斯拉经济高效的端到端模型。 一个重要的关注领域是长途货运，这是一个以经济效率为核心的“计算器业务”。Kasar 强烈驳斥了人们普遍担忧的失业问题，强调卡车司机严重短缺，因为这项工作不受欢迎的性质（例如，长时间远离家人、健康状况不佳、压力大）。采矿业（全球1%的劳动力却占8%的工伤死亡事故）和农业领域也存在类似的劳动力短缺问题，这使得自主化成为一种必需而非奢侈品，许多公司正在积极寻求在几年内实现完全无人驾驶运营。 为了实现实体AI开发的民主化，Applied Intuition 正在推出“Dana”平台，一个智能体平台，旨在降低创建自主系统的门槛。Dana 整合了 Applied Intuition 在过去十年中开发的所有工具和技术，包括完美模拟的真实世界环境、合成数据生成、预训练模型和高级仿真。愿景是让高中生也能像开发 iPhone 应用一样，开发出送货机器人或人形机器人等自主系统。该平台有望在各个领域激发创造力和创业精神的爆发，从而在医疗保健、建筑等领域催生新的应用。 该公司利用先进的“世界模型”——能够高保真地模拟并对现实世界做出反应的仿真系统，涵盖从基于物理的到基于神经网络的仿真。尽管努力实现模拟与现实的完美对齐是一个“不可能完成的难题”，但这一领域的进展使得实体AI的训练效率大大提高。Kasar 推测，未来的视频游戏，例如《侠盗猎车手7》，甚至可能建立在世界模型技术之上。 最终，Applied Intuition 将其工作视为对社会产生深远积极影响的力量，能够降低成本、提高安全性并解决关键的劳动力短缺问题。Kasar 反对技术恐惧症，强调实体AI能够释放的丰富性和效率，从更便宜的能源到更安全的交通。他强调社会必须拥抱技术进步，认识到创新“无人能挡”，而未能适应的国家将会落后。该公司通过其多元化的领导团队和国际业务所培养的全球视野，使其能够应对“主权AI”的复杂性，并与全球经济体展开合作。

Applied Intuition, as articulated by its CEO Kasar, is a "physical AI company" dedicated to embedding intelligence into a billion machines globally. This mission goes beyond typical digital AI applications like optimizing ads or generating videos, focusing instead on physical, moving entities such as cars, trucks, tanks, drones, and heavy industrial machinery. The company posits that the intelligence revolution's most profound impact, and potentially its largest economic players, will emerge from companies that transform the physical world. The scope of Applied Intuition's work is vast and rapidly expanding beyond its initial focus on self-driving cars. While automotive currently accounts for about 30% of its business, the majority is already non-automotive, covering sectors like defense, construction, mining, agriculture, and logistics. This diversification underscores the belief that almost any physical system that moves can benefit from autonomy. The company, with over a thousand engineers and having raised over a billion dollars, focuses on delivering high-quality products rather than relying solely on sales, boasting a global presence with 18 offices. Physical AI faces distinct challenges compared to its digital counterpart. While digital AI often leverages vast internet data for foundation models, physical AI requires extensive proprietary data collection, often involving specialized fleets and navigating complex international regulations to acquire unique datasets. Safety is paramount; unlike a smartphone app, an autonomous machine weighing many tons demands rigorous safety protocols and validation. The performance constraints are also much tighter, requiring real-time decision-making in actual physical environments. The evolution of self-driving technology has seen a shift from imitation learning to end-to-end reinforcement learning, with synthetic data playing a crucial role in accelerating development. Kasar notes significant progress, citing Tesla's Full Self-Driving (FSD) and other manufacturers' advanced driver-assistance systems. While widespread personal autonomous vehicles (L2++) are predicted to become routine and even free by the early 2030s, robotaxis are expected to be available in major cities by 2030-2033. The company attributes the slower rollout to the inherent cost and complexity of bringing such advanced, safety-critical technology to mass market, contrasting Waymo's bespoke, geofenced approach with Tesla's cost-effective, end-to-end model. A significant area of focus is long-haul trucking, which presents a "calculator business" where economic efficiency is key. Kasar strongly debunks the common fear of job displacement, highlighting the severe shortage of truck drivers due to the undesirable nature of the job (e.g., long periods away from family, poor health outcomes, high stress). Similar labor shortages in mining (where 1% of the global labor pool accounts for 8% of work-related fatalities) and agriculture make autonomy a necessity rather than a luxury, with many companies actively pursuing fully driverless operations within a few years. To democratize physical AI development, Applied Intuition is launching "Dana," an agentic platform designed to lower the barrier to entry for creating autonomous systems. Dana integrates all the tools and techniques developed by Applied Intuition over the past decade, including perfectly simulated real-world environments, synthetic data generation, pre-trained models, and advanced simulation. The vision is to enable a high school student, akin to creating iPhone apps, to develop autonomous systems like delivery robots or humanoids. This platform is expected to foster an explosion of creativity and entrepreneurship in diverse fields, leading to new applications in healthcare, construction, and beyond. The company leverages advanced "world models" – simulations capable of representing and reacting to the real world with high fidelity, ranging from physics-based to neural simulations. While striving for perfect alignment between simulation and reality is an "impossibly difficult problem," progress in this area makes physical AI training much more efficient. Kasar speculates that future video games, like Grand Theft Auto 7, might even be built on world-model-based technologies. Ultimately, Applied Intuition views its work as a profound positive force for society, driving down costs, increasing safety, and addressing critical labor shortages. Kasar argues against technophobia, emphasizing the abundance and efficiency that physical AI can unlock, from cheaper energy to safer transportation. He stresses the necessity for society to embrace technological progress, recognizing that "no hand can block the sun" of innovation, and that countries that fail to adapt will be left behind. The company's global perspective, cultivated through its diverse leadership and international presence, positions it to navigate the complexities of "sovereign AI" and work collaboratively with economies worldwide.