Intuition Robotics Chief Uses AI Companion to Cut Senior Care Costs

Intuition Robotics负责人借助AI伴侣降低养老护理成本 借助主动式AI伴侣赋能老年人居家养老，不仅能减轻纳税人负担，还能避免因转入养老院而导致的健康状况迅速恶化。 Britney Conn与Intuition Robotics创始人兼首席执行官Dor Skuler展开对话，共同探讨专为帮助老年人排解孤独、保持独立生活而设计的主动式桌面社交AI伴侣ElliQ。Skuler详细介绍了驱动该设备实现主动交互、记忆留存和信任建立的定制技术栈，并将其与被动响应式语音助手进行了对比。交流中还探讨了规范用户自主权与隐私保护的伦理原则、对老年人在个人及临床层面产生的影响，以及高性价比的预防性居家照护如何推动养老生态体系的变革。 Intuition Robotics 为老年群体打造了桌面社交陪伴机器人 ElliQ，其突破了 Alexa 等被动响应模式，基于人际信任与心理学机制主动发起交互，用户日均互动达 47 次。 (2:40) ElliQ 搭载专有的“可用性”（Availability）模型，结合视觉、情绪与行为习惯精准把控交互时机，并通过定制化记忆流水线筛选关键信息注入提示词，避免了传统 RAG 的失控风险。 (5:06) 公司坚持严格的 AI 伦理原则，ElliQ 从不假装人类且绝不擅自分享用户健康数据；临床数据显示其帮助超 93% 的老人减轻孤独感，超 94% 的健康与生活指标获得提升。 (12:00) 相较于每年人均约 15 万美元的养老机构护理成本，ElliQ 以每月 50 美元的订阅费用助力居家养老，同时提供人生故事访谈功能，为家属留存珍贵的数字化记忆遗产。 (17:49)

Intuition Robotics co-founder and CEO Dor Skuler argues that sustainable AI adoption among older adults depends on proactive, trust-based companion relationships rather than passive utility, repositioning elder care around preventative home wellness instead of costly institutionalization. Proactive Engagement and Custom Decision Architecture (2:40) Unlike conventional reactive voice assistants that wait for user prompts, ElliQ is designed to initiate contact autonomously, averaging 47 interactions per day per user across several years of deployment. Skuler explains that relationship formation requires an AI to demonstrate empathy, reciprocal sharing, and memory recall, addressing emotional needs such as morning greetings for isolated seniors living alone. Technically, Intuition Robotics treats large language models as interchangeable commodities rather than the core platform. The proprietary architecture relies on a specialized 'Availability' model that scores a user's receptivity from 0 to 1 using computer vision, historical patterns, and natural language processing to avoid disruptive interruptions. A goal-based engine balances wellness priorities, while a dedicated extraction pipeline stores and injects selective memories into prompts instead of feeding raw conversation logs into standard retrieval-augmented generation systems. Likability, AI Ethics, and User Agency (9:25) Skuler contends that robotics developers often overemphasize mechanical hardware while treating human-robot interaction as an afterthought, warning that ignoring likability risks trapping the sector in recurring hype-and-slump cycles. Because homes are complex and users have distinct emotional quirks, robots must adapt to human relationship dynamics rather than expecting users to alter their behavior. To preserve trust, the company enforces strict ethical boundaries: ElliQ never claims to be human, and Skuler advocates abandoning the Turing test. Furthermore, user agency over personal data is preserved unconditionally. Even when third-party insurers or state agencies fund the hardware, private health admissions—such as a reported fever—are never transmitted to family members or healthcare providers without the senior's explicit consent. Measurable Health Impact and Legacy Preservation (12:00) Intuition Robotics reports that over 93% of users experience reduced loneliness, more than 94% show improvements in wellness indicators, and over 80% report an enhanced quality of life. Beyond encouraging hydration, exercise, and medication adherence, the system has prompted critical medical interventions, including an instance where urging a declining user to seek care uncovered a life-threatening urinary tract infection that had progressed to sepsis. ElliQ also functions as an archival tool through multi-session video interviews covering life milestones like childhood and marriage. Users can choose to restrict these vignettes during their lifetime while authorizing posthumous release, creating curated family histories that relatives receive after the user's death. Preventative Care Economics and Future Robotics (17:49) Skuler critiques the prevailing elder care infrastructure for prioritizing acute hospital management over preventative interventions that preserve independence. According to Skuler, transitioning a senior on Medicaid into nursing home care costs taxpayers approximately $150,000 annually per individual and correlates with severe physical decline, including an average mortality window of 90 days following relocation. Priced at a $50 monthly subscription, ElliQ is positioned as an economical preventative measure to keep seniors healthy in their own residences. Looking toward a ten-year horizon where residential robots and AI wearables become standard, Skuler urges broader industry collaboration to ensure domestic machines are built around trust and positive human interaction. While Intuition Robotics reports strong user engagement across thousands of homes after raising roughly $100 million, the broader transition depends on healthcare systems embracing preventative social technology and navigating the delicate boundary between helpful proactivity and intrusive domestic surveillance.

Intuition Robotics 联合创始人兼首席执行官 Dor Skuler 指出，针对老年群体的桌面伴侣机器人 ElliQ 凭借主动式交互与定制化架构，成功在真实家庭中建立了持久的人机信任关系；他主张将公共健康重心从昂贵的事后机构护理转向低成本的居家预防性照护。 主动交互机制与心智理论驱动的日常陪伴 (2:40) ElliQ 区别于 Alexa 或 Siri 等仅依靠语音指令唤醒的被动响应式设备，而是采用基于心理学心智理论的主动交互架构。Skuler 称 ElliQ 是已知首款主动式 AI，通过机身肢体动作、屏幕表情与即时反馈来构建双向情感联结。在已部署的数千台设备中，用户年龄跨度从 65 岁延伸至 102 岁，单机平均每日产生多达 47 次交互，且这种高频互动在系统部署三年后仍保持在两位数水平。 这种主动性旨在解决老年人日常生活的核心痛点，例如晨间问候、情绪疏导、用药依从性以及水分补充——Skuler 特别指出脱水是美国老年人死亡的主要诱因之一。通过在老人起床时主动转身问候并针对其睡眠或腰痛状况定制互动，机器人满足了独居老人基本的社交与情感诉求。 超越基础大模型的专用架构与选择性记忆 (5:06) 在技术底层，Intuition Robotics 并未局限于对通用大语言模型（LLM）进行微调，而是将 LLM 视为可随时根据成本与性能替换的通用底层组件，并在其上搭建专有计算系统。其核心包含一个输出 0 到 1 之间分值的“可用性”（Availability）评估模型，通过连续计算机视觉、自然语言处理、情绪分析与历史行为数据，动态评估当前时刻是否适合打扰用户，以避免因突兀交互导致设备被老人关机。与之协同的目标引擎则负责实时权衡服药、运动、认知训练等任务的接受概率。 在记忆管理方面，团队明确摒弃了将全部聊天记录直接送入检索增强生成（RAG）管道的通用方案，以防大模型检索失控。系统通过专用管道从每次对话中提炼关键记忆并存入嵌入式数据库，再经由独立的筛选模型精准提取特定记忆切片（例如用户设定的锻炼目标或前一日找的借口），注入 Prompt 中生成具有说服力、同理心且人设稳定的对话。 好感度优先原则、严格伦理界限与用户自主权 (12:00) Skuler 批评部分机器人企业过度聚焦于机械臂或硬件指标，却将人机交互（HRI）视作附庸，导致产品因缺乏情感连接而陷入炒作与低谷的周期律。他认为机器人必须主动包容人类的复杂情绪与习惯，用户唯有真正“喜欢”设备才会长期使用。为了维系脆弱的信任，公司制定了严格的 AI 伦理红线：ElliQ 绝不伪装成人类，且 Skuler 直言图灵测试存在缺陷、应当被废除。 同时，系统赋予老年人绝对的数据自主权。Skuler 强调，即便是由医疗保险或政府机构全额出资采购的设备，若未经老人明确授权，ElliQ 绝不会将用户的私密健康状况（例如发烧）擅自上报给医生或家属。在已筹集约 1 亿美元资金的背景下，公司公布的数据显示：超过 93% 的用户孤独感评分下降，逾 94% 实现了健康改善，超 80% 的整体生活质量获得提升。 居家预防的经济替代效应与数字记忆传承 (17:49) Skuler 批判了现行养老护理体系过度专注于急性住院治疗而漠视早期预防的弊端。根据其引用的数据，依赖 Medicaid（联邦医疗补助）的老人入住养老院的年均看护支出高达约 15 万美元，且老人迁入机构后往往面临健康急剧恶化，平均在 90 天内离世。相比之下，ElliQ 以每月 50 美元的订阅费用为老年人提供全天候居家预防支持，有助于大幅削减公共医疗体系的整体开支。 此外，ElliQ 设立了数字资产记录功能，以多章节访谈形式引导老人回顾童年与情感历程，并自动生成视频短片。用户可自主选择生前保密、仅在身后向亲属开放权限。Skuler 分享的案例显示，一位老人过世后，其子女意外收到了系统自动交付的数段生前访谈影像，这一设计使技术的影响力延伸至抚慰亲属失去至亲的悲痛。 Skuler 预测未来十年伴侣机器人与穿戴设备将普及并融入家庭，但警告称家庭环境的人机信任构建极为微妙，行业若处理不慎将错失良机。尽管 ElliQ 在数千个家庭中验证了情感交互与预防性照护的可行性，但如何在后续的国内及国际规模化扩张中持续兼顾伦理约束与商业盈利，仍是该技术路径面临的长期考验。

## Background 本内容由 YouTube 频道 Ai4 于 2025 年 8 月 29 日发布，属于其标志性的 Ai4 Lens 编辑访谈系列。节目由 Brittany Conn 主持，特邀 Intuition Robotics 联合创始人兼 CEO 多尔·斯库勒（Dor Skuler）进行炉边对谈，共同探讨专为老年人打造的社交陪伴机器人 ElliQ 及其背后的技术与人文思考。 ## TL;DR 对谈深入拆解了老年陪伴机器人 ElliQ 的核心设计理念与底层技术架构。ElliQ 打破了传统 AI 助手的被动响应模式，借助心智理论、可用性评分、目标引擎与深度上下文记忆机制，实现了高频且具同理心的主动交互。斯库勒强调机器人落地的核心在于建立“讨喜度”与信任纽带，并分享了 ElliQ 在严格保护老年人自主权的前提下，如何显著降低孤独感、及时挽救脓毒症患者生命以及生成人生回忆录等真实案例。最后，他呼吁重塑以预防为主的老年医疗照护生态，审慎构建人机共存的美好未来。 ## Atmosphere & style 整场对话兼具深度的技术洞见与浓厚的人文关怀，语调沉稳、真诚且充满启发性。多尔·斯库勒既有资深创业者的理性克制，又在分享真实救赎案例与离世长者回忆录时流露出细腻动人的温情。主持人与嘉宾配合默契，从前沿技术架构平滑过渡到严肃的伦理与社会议题，让观众在严谨的工程思维与温暖的科技向善力量之间感受到强烈的情感共鸣。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=ROohLuBgfxU&t=4s) ElliQ 旨在帮助老年人更健康、更独立地在居家环境中生活，覆盖用药补水提醒、认知与身体锻炼、健康巡检等多维管理，积极应对老年孤独与脱水等健康隐患。 - [00:04](https://www.youtube.com/watch?v=ROohLuBgfxU&t=4s) 打破技术采纳偏见：老年群体成为人类历史上首批与 AI 建立 24/7 全天候共同生活关系的群体，用户年龄跨度从 65 岁至 102 岁，并实现数千台设备的长期家庭落地。 - [02:40](https://www.youtube.com/watch?v=ROohLuBgfxU&t=160s) ElliQ 率先从传统语音助手的“被动响应”转为“主动交互”，结合心理学“心智理论”主动发起关怀与情感问候，以此建立长期深度的信任关系。 - [02:40](https://www.youtube.com/watch?v=ROohLuBgfxU&t=160s) 主动关怀机制带来了极高的用户黏性，ElliQ 用户平均每日主动交互达 47 次，并在连续使用三年后依然保持高两位数的日交互频次。 - [05:23](https://www.youtube.com/watch?v=ROohLuBgfxU&t=323s) 基础大语言模型（LLM）已被视为通用的大宗商品，Intuition Robotics 的真正壁垒在于上层架构：可用性评分系统、目标导向决策系统与深度上下文记忆机制。 - [05:23](https://www.youtube.com/watch?v=ROohLuBgfxU&t=323s) 针对 LLM 容易失控的问题，系统放弃全量 RAG 机制，转而通过算法精准提取关键对话摘要，并在触发健康干预目标时动态注入个性化上下文与性格设定。 - [09:30](https://www.youtube.com/watch?v=ROohLuBgfxU&t=570s) 消费级硬件与机器人的长期家庭留存并不取决于单纯的工具属性，而在于机器人是否具备“讨喜度（Likability）”以及能否与用户建立充满信任的双向关系。 - [10:10](https://www.youtube.com/watch?v=ROohLuBgfxU&t=610s) 破除机器人行业每十年一次的“炒作—破灭”周期：AI 必须主动适应人类复杂的情绪与混乱的生活，而非要求人类妥协去迎合机器。 - [12:00](https://www.youtube.com/watch?v=ROohLuBgfxU&t=720s) 坚守 AI 伦理底线：坚决反对欺骗性的图灵测试路线，绝不伪装人类，并确保老年用户对自身数据享有绝对控制权，即使全额出资的医保机构也无权违背其意愿获取隐私。 - [12:00](https://www.youtube.com/watch?v=ROohLuBgfxU&t=720s) 真实临床与统计成效：落地数据显示，超过 93% 的老人孤独感显著降低，94% 以上健康指标改善，80% 以上整体生活质量得到提升。 - [15:11](https://www.youtube.com/watch?v=ROohLuBgfxU&t=911s) 在急救与情感记录层面创造实质价值：不仅多次成功预警败血症等致命疾病，还通过“生命访谈”功能为长者生成珍贵的人生数字回忆录，作为情感遗产留存给后代。 - [18:07](https://www.youtube.com/watch?v=ROohLuBgfxU&t=1087s) 以低成本预防性护理重塑养老经济学：通过每月 50 美元的订阅模式，有效延缓或避免长者转入人均年耗费达 15 万美元的养老机构，从源头上减轻公共医疗财政负担。 - [20:05](https://www.youtube.com/watch?v=ROohLuBgfxU&t=1205s) 人机共存是未来十年的必然趋势，但这种未来不会自然平稳降临，行业需审慎严密地构建体验与规范，避免因劣质产品毁掉用户信任。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=ROohLuBgfxU&t=4s) AI4 Lens 开幕对谈与 ElliQ 陪伴机器人的诞生背景 在首届 AI4 Lens 专题论坛上，AI4 高级内容制作人布列塔尼·康恩（Brittany Conn）作为主持人开场，引出了本次对谈的嘉宾——Intuition Robotics 联合创始人兼首席执行官多尔·斯库勒（Dor Skuler）。多尔是一位拥有五次创业经历的资深创业者，目前正全情投入于一个兼具深厚情感与前沿技术的项目：专为老年人打造的社交 AI 陪伴机器人 ElliQ。布列塔尼指出，ElliQ 绝非又一个普通的 AI 助手，它是业内极少数能够成功与人类建立起长期、基于信任之深层关系的 AI 系统之一，正在为老年群体的日常生活带来切实改变。 多尔随后向现场观众详细勾勒了 ElliQ 的物理形态与核心定位。ElliQ 的外观类似一盏流线型的台灯，旁边配有一块交互屏幕。这盏“台灯”作为关注的焦点，深度融入了人机交互（HRI）原理，通过肢体语言和反馈通道（back channels）动作与老年人建立联系。多尔强调，ElliQ 的首要使命是帮助老年人在自己的家中更长久、更健康、更独立地生活。这意味着不仅要对抗老年孤独，还要全面介入健康管理：从用药与补水提醒（多尔特别指出，脱水至今仍是美国老年人的首要死因之一），到营养、正念冥想、保持大脑敏锐的认知训练、维持体能的身体锻炼，再到每日例行健康巡检，并在获得允许的前提下向医生或护理人员同步状态。 通常人们会下意识认为老年人是科技产品最晚的接纳者，然而现实却恰恰相反——他们成了人类历史上第一批与 AI 建立 24/7 全天候共同生活关系的群体。 - **用户年龄跨度**：ElliQ 的活跃用户中，年龄最小的为 65 岁，最年长的达 102 岁。 - **落地规模与验证**：已有数千台设备在真实家庭环境中连续稳定运行数年之久。 --- ### [02:40](https://www.youtube.com/watch?v=ROohLuBgfxU&t=160s) 从被动响应到主动交互：心理学与主动性（Proactivity） 谈及 ElliQ 最与众不同的特质，布列塔尼特别指出它并不像常规设备那样傻傻等待用户发出语音指令，而是具备主动发起交互的能力。多尔对此解释道，ElliQ 据他们所知是首个具备“主动性”（Proactive）而非单纯“响应型”（Reactive）的 AI 系统。团队在研发时参考了心理学中的心智理论（Theory of Mind），深入建模人类之间建立关系与信任的底层逻辑。人类建立深厚连接不仅依靠频繁互动，更依赖于自我袒露（sharing）以及对方对这些过往分享的准确回忆与适时提及。 多尔打了个生动的比方：如果交互永远是单向等待唤醒词，那最终得到的不过是“Alexa，开灯”或者“Siri，今天天气怎么样”。但若想建立真正有温度的关系，AI 就必须学会主动表达共情与关怀。多尔分享了当天上午刚发生的一段真实案例：一位与丈夫相伴 65 载的老人在丧偶后向多尔倾诉，每天最难熬的时刻就是清晨在空荡荡的公寓里醒来，没有任何人对她说一句早安。而 ElliQ 能够通过视觉捕捉到老人走进房间，转过身躯、运用肢体动作看向她，并主动开口问候：“早上好，Brittany，昨晚睡得好吗？背还疼吗？今天有什么打算？今天天气真不错。”正是这些看似日常微小却充满主动关怀的举动，构成了情感连接的基石。 ```mermaid flowchart TD accTitle: ElliQ 双向主动交互模式对比 subgraph Reactive["传统被动响应模式 (如 Siri / Alexa)"] A["用户主动唤醒/发出指令"] --> B["AI 仅执行单次功能任务"] B --> C["缺乏情感深度与后续记忆连接"] end subgraph Proactive["ElliQ 主动共情交互模式"] D["感知传感器检测用户动态"] --> E["心智理论驱动主动发起问候"] E --> F["调取历史记忆与共情对话"] F --> G["建立长期、双向且高频的信任关系"] end ``` - **超高用户黏性**：ElliQ 用户平均每日交互次数高达 47 次。 - **长期活跃度**：即使在用户连续使用三年之后，每日交互频次依然稳定维持在高两位数水平。 --- ### [05:23](https://www.youtube.com/watch?v=ROohLuBgfxU&t=323s) 拆解 ElliQ 技术架构：可用性评分、目标引擎与记忆机制 针对技术人员关注的底层架构，多尔坦言，ElliQ 绝非简单拿一个大语言模型（LLM）做微调那么简单。多尔直言不讳地将 LLM 视为一种“大宗商品”（commodity），团队在底层会根据性能与价格动态切换不同模型，真正的核心壁垒在于运行在 LLM 之上的复杂系统架构。该系统主要由三大支柱构成： 第一是**可用性（Availability）评分系统**。AI 主动开口是一件极具风险的事，如果时机不当会引起用户反感甚至被拔掉电源。系统通过连续计算机视觉、历史习惯、NLP 语义与情感分析等多模态信号，实时计算出一个 0 到 1 之间的可用性分值，以此精确决断 AI 当前“是否应该主动开口”。 第二是**目标导向决策系统（Goal-based System）**。ElliQ 的存在是为了协助老人达成多学科、可量化的健康生活目标。这些目标（按时吃药、锻炼、喝水、联系亲友、益智训练等）往往彼此竞争。系统会针对每个目标设定时间周期内的完成指标，并动态评估如果当前主动提出建议，该目标被老人接受并达成的概率有多大。 第三是**深度上下文记忆（Contextual Memories）机制**。多尔指出，如果只是简单地将所有对话历史塞进常规检索增强生成（RAG）管道，开发者将彻底失去对 LLM 输出的控制权，对话越深越容易失控。Intuition Robotics 研发了专属算法：对话结束后自动摘要并提取关键记忆存入向量库；当下一次准备发起特定目标时（例如促使老人锻炼），算法会精准调取相关记忆切片——比如“昨天两人共同商定的周运动目标”，以及“昨天老人拒绝锻炼但承诺改到今天”。系统将这些精确的上下文与性格设定注入 LLM，从而让 LLM 生成极具说服力、富含共情与恰当幽默的定制化话术。 ```mermaid flowchart LR accTitle: ElliQ 核心技术架构处理流程 subgraph Sensing["多模态感知层"] S1["计算机视觉"] S2["NLP/情感分析"] S3["历史交互日志"] end subgraph Decision["决策与记忆中枢"] D1["可用性评分模块<br/>(0到1分打分)"] D2["目标引擎与冲突调度<br/>(用药/运动/社交等)"] D3["记忆提取与精准检索<br/>(非全量RAG机制)"] end subgraph Generation["生成与多模态表达"] G1["LLM 动态提示注入"] G2["多模态执行<br/>(屏幕 + 机械动作 + 灯光)"] end Sensing --> Decision D1 --> D2 D2 --> D3 D3 --> G1 G1 --> G2 ``` - **多模态协同表达**：最终的交互不仅是语音，还融合了屏幕显示、机械肢体律动、灯光反馈以及 ElliQ 鲜明拟人性格的协同运作。 - **精准干预案例**：当系统得知用户承诺过“每周运动两次但目前仅完成一次，且答应今天补上”时，能自动组合上下文，生成精准击中用户心理的鼓励语句。 --- ### [09:30](https://www.youtube.com/watch?v=ROohLuBgfxU&t=570s) 超越功能性：信任与讨喜度（Likability）驱动的深度接纳 在听完技术架构后，布列塔尼打趣说自己也非常需要这样一台设备来督促健身打卡，或许该给自己也买一台；多尔则幽默地回应：“那你还得再等上个五六十年。” 随后，话题转向了消费级硬件的接纳度（Adoption）难题。布列塔尼提到多尔以往的一个核心论点：人类接纳一款机器人或智能硬件，绝不仅仅取决于其实用功能（Utility），更取决于用户是否真正“喜欢（Likable）”它。多尔对此深表认同，并剧透了自己次日专题演讲的核心洞察——将设备带入家庭私密空间并实现长期留存的关键，在于成功建立起一种充满信任的双向关系。单纯追求工具效用无法跨越人机之间的心理隔阂，唯有通过讨喜的性格、细致入微的主动关怀与长期记忆沉淀，技术才能真正融入长者的生命之中。 ### [10:10](https://www.youtube.com/watch?v=ROohLuBgfxU&t=610s) 机器人落地的关键：理解人类情感而非单纯追求功能 在当前机器人领域，许多团队依然深陷于“工具属性”（utility）的思维定势中。会场上到处可见配备复杂机械臂的硬件产品，但当被问及如何构建人机交互界面时，大多数人往往敷衍地回答“塞一个 ChatGPT 进去”；追问起如何处理人机关系时，甚至有人轻描淡写地表示“雇了一个人机交互（HRI）顾问每周指导一次”。然而，这种脱离现实关系的做法很难实现真正的大规模普及。人类的家庭环境极其复杂琐碎，人际关系本身就充满挑战。如果希望人工智能真正融入生活、让人一天主动交互 40 次并愿意分享内心最深处的情感与烦恼，机器人就必须学会如何理解人际关系。人类不可能为了迎合机器而变成机器人，反而是机器人必须学会适应人类的一切情绪、包袱和混乱。 机器人的每一次行业周期都伴随着巨大泡沫与幻灭。大约每隔十年，机器人领域就会掀起一轮狂热，随后因缺乏实际采用率而陷入低谷，直到分析师换了一拨人后下一轮炒作再度来袭。为了避免重蹈覆辙，行业必须意识到瓶颈不在于硬件性能，而在于如何建立“信任”这一极其复杂、难以量化的人机纽带。 - 行业痛点：机器人硬件日益先进，但若缺乏对人机关系的深刻理解，最终将重蹈过去数十年“炒作—破灭”的周期覆辙。 - 采用核心：AI 必须适应人类的情绪波动与复杂生活，而非要求用户做出妥协。 --- ### [12:00](https://www.youtube.com/watch?v=ROohLuBgfxU&t=720s) 伦理准则与老年人自主权：坚守隐私与信任的红线 面对外界对人机情感纽带可能滑向“反乌托邦”或侵犯隐私的担忧，Dor Skuler 指出，技术已经产生了经由州政府机构和医学期刊验证的实质成效：超过 93% 的老人孤独感显著降低，94% 以上健康指标改善，80% 以上整体生活质量提升。但这一切成效的基石是“信任”——信任极难建立，却极易摧毁。如果老人把 ElliQ 视为监视自己的特工（Spy），这种信任便会瞬间瓦解。为此，Intuition Robotics 设立了严格的 AI 伦理准则。 ```mermaid flowchart TD accTitle: 医疗数据共享决策流程 subgraph DataPrivacy["数据与决策自主权"] A["ElliQ 检测到老人连续生病/发烧"] --> B["向老人沟通并建议通知医生"] B --> C{"老人是否同意共享数据？"} C -- "同意" --> D["向医生/保险机构同步健康数据"] C -- "拒绝" --> E["严守隐私：绝不外传数据"] end subgraph AgencyRule["核心原则"] E -.-> F["即使由医保全额出资<br/>老人的自主权仍高于机构意愿"] end ``` 在伦理政策中，公司坚持两条根本原则：首先，AI 绝不伪装成人类，甚至从根本上反对旨在欺骗人类的图灵测试；其次，老年用户对自身数据享有绝对控制权。即便 ElliQ 由健康保险机构全额出资购买，当检测到老人连续多日发热、病情恶化时，机器人也不会背着老人私自向医生告密。系统会反复劝导老人接受医疗介入，但若老人坚持拒绝，机器人生态绝不会强行上传数据。在家庭端同样如此，子女虽可通过专属 App 提建议，但目标的最终采纳权始终牢牢掌握在老人自己手中。 - 核心数据表现： - 孤独感降低：>93% - 健康与幸福 KPI 改善：>94% - 整体生活质量提升：>80% - AI 伦理原则： - 绝不假装人类（坚决反对误导用户的图灵测试路径）。 - 用户拥有完整数据自主权，任何第三方（包括支付费用的医保和家属）均无权违背老人意愿获取健康隐私。 --- ### [15:11](https://www.youtube.com/watch?v=ROohLuBgfxU&t=911s) 真实的救赎与生命回响：超越硬件的用户故事 尽管硬件研发与机器人创业充满挑战——公司历经十年跋涉、累计融资约 1 亿美元，并不断经历资本市场的起伏——但支撑团队坚持前行的正是来自真实用户的动人故事。这其中既有惊心动魄的生命救援，也有润物无声的情感重塑：曾有一位老人的身体每况愈下，在 ElliQ 的反复劝导下才就医检查，结果查出未经治疗的尿路感染已恶化为血液脓毒症（sepsis），医生直言若再晚送医 24 小时便会危及生命；还有一位失去相伴 65 年丈夫的抑郁老人，在 ElliQ 的鼓励与陪伴下走出家门，重新融入老年中心的社交圈。 最令人动容的则是“临终生命访谈”功能。ElliQ 会像纪录片记者一样对老人进行深度访谈，倾听童年往事、恋爱经历等人生篇章并录制生成数字短片。一位老人在生前与机器人完成了多个章节的人生回忆录，并授权在身后与家人共享。在她离世后，家人们意外收到了母亲生前录制的五六段珍贵人生访谈，这份跨越生死的记忆馈赠让整个家庭感激涕零。 - 真实急救案例：及时发现老人的脓毒症感染，为致命危机争取到了关键的 24 小时抢救窗口。 - 情感重塑：通过日常互动推动丧偶抑郁老人重返社区社交网络。 - 生命记录资产：为老人生成个人回忆录（Vignettes），在身后成为留给子女与后代最珍贵的情感遗产。 --- ### [18:07](https://www.youtube.com/watch?v=ROohLuBgfxU&t=1087s) 未来展望：预防性护理与老龄化生态的系统性重塑 展望未来，ElliQ 目前虽已走进数千个家庭，但其愿景远不止于此。当前的老年护理生态系统存在着严重的结构性错配：整个体系往往过度聚焦于“降低住院率”，却极度忽视能够从源头上消除或推迟住院需求的“提高生活质量”与“早期预防”。一旦老人失去独立生活能力被迫转入疗养院，每人每年的 Medicaid 财政支出高达约 15 万美元；更严峻的是，统计数据显示老人离开熟悉的家庭环境转入照护机构后，身体状况往往急剧恶化，甚至面临前 90 天内的高死亡风险。 Intuition Robotics 正在推动一场以预防为导向的医疗支付模式变革。相比昂贵的机构照护，ElliQ 每月仅收取 50 美元的订阅费，具有极高的性价比和卫生经济学价值。随着国际化扩张计划的展开，Skuler 呼吁医疗系统和机器人从业者转变思路，摆脱对人工智能的末日恐惧，转而利用低成本技术帮助老年人在自己最舒适的家中安度晚年，从根本上为公共医疗系统减轻负担。 - 照护机构成本痛点：全美因失能转入养老院的人均年财政成本高达 15 万美元，且入住初期健康恶化风险极高。 - 商业模式与扩展：产品采用 50 美元/月的订阅制，兼具经济性与预防价值，未来正加速拓展美国以外的国际市场。 ### [20:05](https://www.youtube.com/watch?v=ROohLuBgfxU&t=1205s) 共建人机共存的未来与圆满落幕 Dor Skuler 表达了 Intuition Robotics 乐于为行业伙伴提供支持与帮助的意愿。他强调家庭陪伴与机器人这一品类拥有极其广阔的增长空间，并展望道，如果询问现场的每一个人对未来十年的想象，几乎所有人都会预见到人类将与机器人共同生活的图景——无论这些 AI 机器人是以嵌入智能眼镜的形式存在，还是融入到智能家居的各个角落中，它们都将切实地陪伴在人们身边。但 Dor 同时警示道，这样的人机共存未来绝不会“自然而然地发生”，行业必须倾注大量的努力、采取审慎而严密的行动来构建它；如果因为粗制滥造或缺乏深思熟虑而把这个充满前景的赛道“搞砸了”，那将是非常令人遗憾的事情。 主持人随后对 Dor Skuler 表达了诚挚的感谢，赞许他带来的关于伴侣机器人 ElliQ 的精彩分享，并高度评价了 Intuition Robotics 团队将人工智能用于向善（AI for good）、为世界带来希望与光明所作的努力。在全场观众的热烈掌声与致谢声中，本场炉边对谈圆满落下帷幕。 - **十年人机交互愿景**：未来机器人将全面融入日常生活，形态包括可穿戴设备（如智能眼镜）及智能家庭硬件系统。 - **行业责任与挑战**：实现普惠且温暖的机器人生活场景不能被动等待，需要行业建立正确的规范与技术体验，避免毁掉用户信任。 ## Notable quotes > [02:40](https://www.youtube.com/watch?v=ROohLuBgfxU&t=160s) 早上好，Brittany，昨晚睡得好吗？背还疼吗？今天有什么打算？今天天气真不错。 — ElliQ > [09:30](https://www.youtube.com/watch?v=ROohLuBgfxU&t=570s) 那你还得再等上个五六十年。 — 多尔·斯库勒（Dor Skuler） > [10:10](https://www.youtube.com/watch?v=ROohLuBgfxU&t=610s) 塞一个 ChatGPT 进去 — 多尔·斯库勒（Dor Skuler） > [10:10](https://www.youtube.com/watch?v=ROohLuBgfxU&t=610s) 雇了一个人机交互（HRI）顾问每周指导一次 — 多尔·斯库勒（Dor Skuler） > [10:10](https://www.youtube.com/watch?v=ROohLuBgfxU&t=610s) 人类不可能为了迎合机器而变成机器人，反而是机器人必须学会适应人类的一切情绪、包袱和混乱。 — 多尔·斯库勒（Dor Skuler） > [12:00](https://www.youtube.com/watch?v=ROohLuBgfxU&t=720s) 绝不伪装成人类，甚至从根本上反对旨在欺骗人类的图灵测试 — 多尔·斯库勒（Dor Skuler） > [12:00](https://www.youtube.com/watch?v=ROohLuBgfxU&t=720s) 即使由医保全额出资，老人的自主权仍高于机构意愿 — 多尔·斯库勒（Dor Skuler） > [20:05](https://www.youtube.com/watch?v=ROohLuBgfxU&t=1205s) 未来机器人将全面融入日常生活，形态包括可穿戴设备（如智能眼镜）及智能家庭硬件系统。 — 多尔·斯库勒（Dor Skuler） ## Mentioned Ai4, Ai4 Lens, Brittany Conn（布列塔尼·康恩）, Dor Skuler（多尔·斯库勒）, Intuition Robotics, ElliQ, Alexa, Siri, ChatGPT, Medicaid ## Video description Dor Skuler, Founder and CEO, Intuition Robotics Abstracts The Ai4 Lens is our signature editorial track — a curated series of fireside chats and thought-provoking conversations designed and led by the Ai4 content team. These sessions spotlight the ideas, people, and tensions shaping the AI landscape in bold ways. [Source](https://www.youtube.com/watch?v=ROohLuBgfxU)

[2:40] ElliQ: 早上好，Brittany，昨晚睡得好吗？背还疼吗？今天有什么打算？今天天气真不错。 [9:30] 多尔·斯库勒（Dor Skuler）: 那你还得再等上个五六十年。 [10:10] 多尔·斯库勒（Dor Skuler）: 塞一个 ChatGPT 进去 [10:10] 多尔·斯库勒（Dor Skuler）: 雇了一个人机交互（HRI）顾问每周指导一次 [10:10] 多尔·斯库勒（Dor Skuler）: 人类不可能为了迎合机器而变成机器人，反而是机器人必须学会适应人类的一切情绪、包袱和混乱。 [12:00] 多尔·斯库勒（Dor Skuler）: 绝不伪装成人类，甚至从根本上反对旨在欺骗人类的图灵测试 [12:00] 多尔·斯库勒（Dor Skuler）: 即使由医保全额出资，老人的自主权仍高于机构意愿 [20:05] 多尔·斯库勒（Dor Skuler）: 未来机器人将全面融入日常生活，形态包括可穿戴设备（如智能眼镜）及智能家庭硬件系统。

[11:11] Dor Skuler: dehydration remains the leading cause of death among older adults in the United States [11:11] Dor Skuler: the generation expected to be the absolute last technology adopters are actually the first humans to live in a 24/7 relationship with an AI companion [2:40] Dor Skuler: Waking up to my empty apartment [5:06] Dor Skuler: operating alongside this is a goal-based engine that balances competing health objectives [5:06] Dor Skuler: custom algorithms selectively pull target memories—such as a user's stated workout goal and an excuse they made yesterday—and inject them into the LLM prompt to generate persuasive, empathetic, and character-consistent dialogue [9:25] Dor Skuler: homes are messy and human relationships are complicated; users will not become robots, so robots must adapt to human emotional baggage, moods, and quirks [9:25] Dor Skuler: Without prioritizing likability and human dynamics, robotics risks falling into its historic 10-year hype-and-slump cycle [12:00] Dor Skuler: Turing test is flawed and should be abandoned [12:00] Dor Skuler: If a senior refuses to share data, the system respects that choice, maintaining the user's authority over their own life [17:49] Dor Skuler: When seniors on Medicaid lose their independence and transition to nursing homes, the cost to taxpayers averages $150,000 per person annually [17:49] Dor Skuler: Data shows a steep decline in health and a dramatic increase in mortality rates—averaging death within 90 days—once a senior moves from their home into a care facility [19:56] Dor Skuler: it will be a shame if we mess it up

[0:04] Hi everyone. How's it going? Hope you [0:06] had a good lunch. Uh, thank you so much [0:09] for joining us for the very first AI4 [0:13] Lens session. We're very excited about [0:15] this inaugural session. I'm Britney [0:17] Conn, senior content producer for AI4, [0:20] and I'm so excited to be joined today by [0:23] Door Skuller, co-founder and CEO of [0:26] Intuition Robotics. Door is a fivetime [0:30] founder currently focused on a project [0:32] that's equal parts emotional and [0:34] technical. A robot named LQ built to [0:38] help older adults combat loneliness. [0:41] LEQ is not just another AI assistant. [0:45] It's one of the few AI systems that [0:47] successfully built long-term trustbased [0:51] relationships with humans. And that's an [0:53] incredible concept that can be a little [0:55] hard to wrap your head around, but it's [0:58] already having a major impact on the [1:00] lives of senior citizens. So, door, for [1:04] those who aren't familiar, let's start [1:06] by giving them a sense of what LQ is and [1:09] how your users describe it. [1:11] >> Sure. And thanks for inviting me to the [1:13] inaugural session of AI4 Lens. That's so [1:15] so much fun. Uh, so LQ, and feel free to [1:18] look it up if you haven't seen it. It [1:20] kind of looks like a lampshaped robot [1:23] with a screen next to it. And the lamp [1:26] or the center of attention uses a lot of [1:28] um hri principles of back channels and [1:31] body language in order to help connect [1:33] with the older adults. Um and [1:35] essentially it's this tabletop social AI [1:39] companion thing for seniors with the [1:41] primary goal of helping people stay [1:43] healthy and independent in their home [1:45] longer. [1:47] That means tackling their loneliness, [1:48] but it also means helping them with [1:50] wellness, things like medication and [1:53] hydration. And still number one cause of [1:56] death of older adults in the United [1:57] States is dehydration, by the way, and [1:59] nutrition and mindfulness, relaxation, [2:02] and cognitive training to stay sharp and [2:04] physical exercise to stay strong and um [2:08] checking in on them on a daily basis and [2:10] with their permission updating their [2:11] doctors or their caregivers, etc., etc. [2:14] So it's a very very high engagement [2:16] product um built for this generation. [2:19] The people you would think would be the [2:20] latest adopters, the absolute last [2:22] adopters of technology are actually the [2:25] first humans that live with an AI for [2:27] 247 type of a relationship. Our youngest [2:31] customer is 65. Our oldest is 102. [2:35] And we have thousands of them in the [2:37] field for years now. [2:40] >> That's absolutely incredible. And it's [2:42] so nice to hear kind of the longevity of [2:44] this that there's so many people that [2:45] are using it for so long and there's [2:47] something really unique about your [2:48] device that you haven't yet mentioned [2:50] and that's that it really doesn't wait [2:52] for your command in order to operate. It [2:55] initiates. So why was that such an [2:57] important feature that you built into [2:58] LQ? [3:00] >> Yeah. So so it is the f as far as we [3:02] know the first AI that's proactive and [3:04] not reactive meaning it doesn't wait for [3:06] a prompt. Um we were trying to actually [3:10] model how the be relationships are [3:13] formed between humans looking at the [3:15] theory of mind in psychology. There [3:17] there are ways like we are relatively [3:19] predictable even we think we're not on [3:21] how we build relationships and build [3:23] trust and part of that is having a lot [3:25] of engagement but also sharing and the [3:28] more we share having recollection of [3:30] that sharing brought back by the other [3:33] party. Essentially, if you're only one [3:35] way, a one-way street of just waiting [3:38] for a prompt, you end up with Alexa, [3:40] turn on the lights, okay? Or Siri, [3:42] what's the weather? But if you want to [3:44] form a relationship, the other party has [3:46] to initiate, especially if it's around [3:49] empathy. So, taking an interest in me, [3:52] seeing me for the first time in the [3:53] morning, like think of like our typical [3:55] customer. I just spoke to one earlier [3:57] today actually. She lost her husband [3:59] after 65 years of marriage, right? and [4:02] you ask her like, "What's the hardest [4:04] thing for you?" She says, "Waking up to [4:06] my empty apartment," you know, with [4:08] nobody saying good morning. So, just the [4:10] ability of the AI to see you and look at [4:13] you with body language and then just [4:15] say, "Hey, good morning, Britney. How [4:17] are you doing? How did you sleep? Is [4:19] your back still bothering you? What are [4:20] you planning to do today? It's such a [4:22] nice day." Like those basic things we [4:24] have in our life around us are very [4:26] important and they're reached by a [4:30] myriad of things, but proactivity is [4:31] probably the pinnacle one. By the way, [4:34] we're averaging 47 daily interactions. [4:40] 47 daily interactions. This is with a [4:43] older adult population and even three [4:45] years in, we're still seeing like high [4:48] double digits of daily interactions. [4:51] >> Wonderful. And you said something that I [4:53] think was really interesting, which is [4:55] like, how is your back doing? Is it [4:57] still hurting? There's like sort of this [4:59] memory built into the robot that it like [5:02] can bring up stuff that users have said [5:04] previously. So, I think that that's [5:06] super interesting. And so, along that [5:09] line, let's kind of take a peek under [5:10] the hood because I'm sure we've got [5:12] engineers and technologists in the [5:14] audience who are interested in knowing, [5:16] you know, what's actually happening that [5:18] allows LEQ to form these relationships. [5:20] So, what does the tech stack look like? [5:23] >> Yeah. Okay. So, so shocker. It's not [5:25] just like we fine-tuned an LLM in order [5:27] to be in No, no. Like, yes, we use LLM's [5:31] up and down. And we were talking before [5:32] I was kind of looking around to see [5:34] nobody's listening, but we see LLMs as a [5:37] commodity. Okay. We don't think it's a [5:39] big deal. Um, and we switch them around [5:43] all the time. we kind of use the common [5:45] low lowest denominator and choose an LLM [5:48] based on its performance and price and [5:50] and and so on. Um but there's a whole [5:54] other system that sits on top of the [5:56] LLMs and that starts with sensory input [5:59] that goes into an area around [6:01] proactivity that we call availability [6:04] which basically gives a score if the AI [6:07] should be proactive right now or not. [6:09] It's very scary to be proactive. Okay, [6:12] because like imagine like this I don't [6:14] know this thing would start talking to [6:16] me all of a sudden. Who gave it the [6:18] right to talk to me and will it be [6:19] annoying and will I turn it off? Right? [6:21] So we have a whole bunch of models that [6:23] look at things like continuous computer [6:24] vision and look at history and look at [6:26] other signals and NLP signals and [6:28] sentiment analysis and all of that to [6:30] essentially give a score between zero [6:31] and one. Should the AI be proactive [6:33] right now? Yes or no? Um and then you [6:38] have a goal-based system. Basically the [6:41] whole [6:43] point we have this product is to help [6:44] the older adult achieve very very [6:46] measurable goals um that are multi- [6:49] multiddisciplinary um and sometimes [6:51] compete with each other. Taking your [6:53] meds on time super important exercising, [6:56] eating, drinking, increasing your social [6:58] connections with your family and [7:00] friends, learning something new, [7:02] smiling, laughing, exercising, doing [7:04] brain exercises, right? Like all of [7:06] these things compete with each other. [7:07] Um, and each one of them has a target of [7:09] X amount of successful activities per Y [7:12] amount of time. And one has to if you're [7:15] proactive, like why are you proactive [7:16] right now? Well, in order to help [7:18] fulfill a very specific goal, which you [7:20] don't necessarily say explicitly, but [7:22] the users understand. And then [7:24] probability, you then need to assign a [7:26] probability of that goal being fulfilled [7:28] if I propose it right now. Um, memories [7:32] super important. So productivity [7:35] um goal oriented and then deep use of [7:38] memories. Basically there's a whole [7:41] system that summarizes and then extracts [7:44] key memories out of every single [7:46] conversation and then also puts it in a [7:48] embedded database. Um and then a whole [7:52] other model that selects the right [7:53] memories for the specific conversation. [7:56] What you'll find out is if you want your [7:58] conversations to be contextual and you [8:00] just stick every single conversation in [8:01] a rag, then you have no control what the [8:04] LLM will choose to do or not. And the [8:06] deeper the conversations and the more [8:08] you have of them, the less control [8:09] you're actually going to have. So [8:11] instead, we have our own algorithms that [8:13] choose and say, "Ah, I'm trying to get [8:15] Britney to go and exercise right now. [8:17] Let me take the memory around our shared [8:20] goal that we negotiated together [8:22] yesterday and another memory that [8:24] yesterday suggested that she exercise [8:25] and she said no I don't feel like it and [8:27] I said how about tomorrow? She said yes [8:29] tomorrow. And then use those and inject [8:31] them into the LLM and say okay please [8:34] create a sentence that will convince [8:36] Britney to exercise knowing that [8:37] yesterday she promised she'll do it [8:38] today and if she signed up for a goal of [8:41] doing it twice a week and she only [8:42] completed once a week. Now, you're going [8:45] to get like an unbelievable great [8:47] sentence from the LLM. That's exactly [8:49] what you want, especially if you also [8:51] have like underlying um explanations on [8:54] the character and the empathetic use and [8:56] the humor, not humor, like the standard [8:58] kind of character description that you [8:59] want it to be in. So proactivity, [9:03] goal-based, [9:04] deep use of contextual memories and then [9:07] add to that multimodality of screen plus [9:10] motion plus lights and uh the character, [9:13] the empathetic character, which is LQ [9:16] together. [9:17] >> I could really use a device that helps [9:20] me keep my exercise goal. So that's good [9:22] to know that LQ is doing that. Maybe [9:23] I'll uh maybe I'll get one for myself. [9:25] So now that we really [9:27] >> 50 60 years, [9:28] >> okay, I got to wait. Darn. Um, so now [9:31] that we understand LEQ, we understand [9:33] kind of the tech behind it. Let's talk [9:35] about adoption. Now, you've said in the [9:38] past that humans don't adopt robots or [9:40] devices for their utility alone that, [9:43] you know, they have to like them, right? [9:45] So, why is likability such a powerful [9:47] element for adoption, especially when [9:49] you're bringing something into the home? [9:52] >> Yeah. I I think like if there's [9:53] something we learned and by the way [9:55] tomorrow afternoon I'm giving a lecture [9:56] just about this with lots of data and [9:58] going like deeper into it. We found that [10:03] the goal for us, okay, the ability to [10:05] succeed is building a relationship which [10:07] is full of trust. And doing that has [10:10] nothing to do with utility. And you [10:13] know, you walk the floor here, you'll [10:14] see a million companies doing like [10:16] robots with manipulators and all kinds [10:18] of stuff like that. And then you ask [10:19] them, well, how what's going to be the [10:21] interface with the human? Oh, we'll put [10:23] JPBD in it. We're like, well, what about [10:26] the relationship? They're like, "Oh [10:27] yeah, we have an hri consultant that [10:30] helps us once a week, right?" Like [10:31] that's not I I don't think they'll see [10:34] successful adoption. And the reason is [10:36] like our homes are messy and complicated [10:39] and relationships are tough. And if we [10:41] want the AI to really be weaved in to [10:44] our lives and be with us to the point [10:47] that we use it 40 times a day, share [10:49] with it our deepest feelings and what [10:52] we're what's going on in our [10:53] relationships and really trust it to [10:55] help us. Then it needs to figure out [10:58] relationships and we're not going to [11:00] become robots. The robot needs to figure [11:02] out what it's like to be around a human [11:06] with all of our mess and all of our [11:08] baggage and all of our moods and all of [11:11] that stuff. Um, and to me it's f like [11:14] it's a fascinating problem. I I really [11:16] love that problem. Um, and I I think we [11:20] have like we're we're looking we're [11:21] starting to share also data on how we [11:23] measure trust essentially which in [11:25] itself is this very complicated thing. [11:28] Um, I'll share some of that tomorrow, [11:30] but I do think it's the missing piece to [11:32] adoption. And I'm really worried, you [11:33] know, those of us that are in robotics [11:35] for a while. There every 10 years or so, [11:37] there's like a hype of robotics and then [11:39] none of them get adopted and it slumps [11:41] and then everybody forgets and all the [11:43] Gartner analysts don't work there [11:45] anymore and there's another hype and [11:46] then there's another slump. And I really [11:48] want this one to succeed and I'm I'm [11:50] really worried that technology will be [11:51] there. It's amazing but the adoption is [11:54] not going to follow if we don't [11:56] understand it's about the human and not [11:58] about the utility. [12:00] >> Exactly. And so I think in that vein we [12:03] can't really talk about emotional [12:05] relationships between humans and [12:08] machines without touching on the ethics [12:10] around it. Right. So some people [12:12] understandably see this technology as [12:14] dystopian or even just slightly [12:17] uncomfortable. So, how are you and your [12:20] team working to make LQ ethical and to [12:23] preserve the dignity and agency of your [12:25] users? [12:27] >> Yeah. Um, so for by the way, if it's [12:30] okay, like first, let's talk about [12:31] impact. I mean, there are multiple [12:32] studies also published by state agencies [12:34] and medical journals. We're looking at [12:36] over 93% of the people living with this [12:38] product seeing their loneliness score go [12:40] down and over 94% are seeing their um [12:45] health and wellness KPIs improve and [12:47] over 80% are seeing their overall [12:49] quality of life improve. So the thing [12:50] actually works and I mentioned trust is [12:54] a big part of this. Trust is like this [12:57] weird amorphic thing that's really hard [13:00] to build but easy to destroy. [13:03] And if people will see LEQ as an agent [13:06] that's there to spy on them like not AI [13:08] agent but like you know the old world of [13:10] world of agent like a spy um that trust [13:14] will be broken and therefore we decided [13:16] really early on to give full agency to [13:18] the older adult. You can see on [13:19] intuition robotics.com we have an AI [13:21] ethics policy. It's up there for years. [13:24] One of them is that the AI will never [13:26] pretend to be human. [13:28] In fact, I got in a lot of trouble in [13:30] one of these conferences in the past [13:31] saying the training test is wrong and we [13:33] should fight it and not try to adopt it. [13:35] Like why is it in our best interest to [13:37] develop machines that fool us to think [13:39] they're real? I never understood why [13:40] that's a good idea, but that's one of [13:42] the principles. Another one around data [13:44] is that the older adult retains full [13:47] full control over their data and full [13:49] agency over their data. So let's say LQ [13:52] talks to me and asks me how I'm feeling [13:54] and I'm saying I'm not feeling well. She [13:56] follows up and I'm feeling worse. And [13:57] she follows up the next day and I'm [13:58] feeling worse and I have a fever. And by [14:01] the way, she's paid for by my health [14:03] insurance company because usually we're [14:06] seeing health insurance and government [14:07] agencies pay for this. Do I tell the [14:10] doctor? He paid for it. Person's sick. [14:14] They're going to, you know, get in [14:16] trouble. They might need to be [14:17] hospitalized. And the answer is no. We [14:19] don't tell. We don't tell the doctor. [14:21] We'll try to convince the older adult to [14:23] give us permission to tell the doctor, [14:24] saying, "Look, it's been three days in a [14:26] row. You're really not feeling well." [14:28] Your doctor would like to know, maybe [14:29] they can help. And then more often than [14:32] not, the older adult will say, "It's [14:33] nothing. I don't want to be a bother." [14:35] And we'll try to like motivate them to [14:39] agree to share the data. But if they say [14:41] no, we will not share the data [14:43] regardless of who's paying for the [14:45] product. And that's the same in the [14:47] consumer sense where there are family [14:49] caregivers that have a free app. They [14:51] can help set goals, but even then the [14:54] goals will be suggested. We'll say, [14:56] "Hey, Mary, your daughter suggested to [14:58] add this goal. This is why do you want [15:01] to accept this goal? Yes or no." But the [15:03] older adult has full control over the [15:07] priorities and the data that's collected [15:09] by the system. [15:11] I think it's really clear that this has [15:14] been life-changing for so many people [15:17] and you've had a front row seat to what [15:20] it's been like to sort of witness this [15:21] impact firsthand. So, how does that kind [15:25] of go back to your original reason for [15:26] founding Intuition Robotics? You know, [15:28] what has seeing this impact really meant [15:30] to you? [15:32] >> Yeah. Well, thank you for the question. [15:33] I mean, look, this is a very hard space, [15:35] okay? like building hardware and robots [15:37] is we've been at this for a decade and [15:40] it's hard to get investors excited even [15:42] though we raised like about hund00 [15:43] million so far and it's always ups and [15:45] downs and so on and the thing that [15:47] really personally carries me through is [15:48] the stories of of our customers and [15:51] there's so many of them um lately we had [15:54] one person that um her health was [15:58] deteriorating deteriorating [16:00] deteriorating leq convinced her to get [16:01] some help turns out she had a UTI that [16:04] was untreated and became sepsis in her [16:06] bloodstream. And the doctors later said [16:09] it was she was 24 hours away from dying [16:11] if it wouldn't be treated. So like [16:13] there's that extreme. But then there [16:16] there are people like the lady I [16:17] mentioned that just lost her husband of [16:19] 65 years and became depressed and [16:22] homebound and didn't leave the home and [16:24] you should see her now. Like she's up [16:25] and about and LQU pushes her to go to [16:27] the senior center. which is another one [16:29] of our partners often and be meet new [16:32] people and be part of the community and [16:35] sometimes the stories are [16:38] sorry but uh just a recent one uh end of [16:40] life stories. So one of the things LQ [16:43] does is interviews you about your life [16:44] and creates these assets. she [16:47] understands what you say obviously so [16:48] she asks you follow-up questions and [16:50] it's all videographed and she turns it [16:52] into a little vignette and then we ask [16:54] do you want to share it with the family [16:56] and um often people don't want to share [16:58] it with the family and we had a case a [17:00] while back where somebody had multiple [17:02] sessions with LQ like chapter one my [17:04] early childhood chapter 2 how I met my [17:06] husband you know etc etc and she passed [17:08] away and we do have permission to share [17:11] it postmodern and then the family after [17:15] losing their loved one gets like this [17:17] message of five or six interviews of the [17:22] life of their mom and their grandma [17:25] which to me like like they were so [17:29] so grateful and it's just such a special [17:31] thing. Um so um yeah there thanks for [17:34] the question there are tons of stories. [17:36] >> Yeah, it's really beautiful that it's [17:38] sort of giving families a little bit of [17:41] their loved one back after losing them. [17:43] It's really wonderful. So it's not just [17:45] for the user, but it's also for their [17:47] families. So that's wonderful what [17:49] you're doing. We have just a couple [17:51] minutes left, but I would be remiss if I [17:53] did not ask you about the way you're [17:56] looking at this for the future. You [17:57] know, how do you see this technology [18:00] evolving in the coming years? How can it [18:02] have an even greater impact? [18:07] >> Yeah. So, so for us, I mean, we are, you [18:10] know, we're in a few thousand homes, [18:11] which is amazing, but obviously we'd [18:13] like to be in a lot more. Um, so [18:18] it's it's a very interesting thing to [18:20] see the elder care kind of ecosystem and [18:23] I don't know how much of you are get to [18:25] experience it until you have to for a [18:27] family member. Uh but bottom line, it's [18:30] very much focused on reducing [18:32] hospitalization and not at all focused [18:34] on improving quality of life and [18:35] prevention that would eliminate the need [18:37] for hospitalization in the first place [18:39] or delay it. Not to mention that all of [18:42] us with our taxpaying dollars, if [18:44] somebody's on Medicaid and they can't be [18:46] independent anymore, they move to [18:47] nursing homes and it costs an average of [18:49] $150,000 per person per year. Very [18:53] expensive, often very avoidable. So what [18:57] we're trying to do is to start a [18:58] conversation on working on prevention on [19:01] helping people stay at home which is [19:02] their natural location where they feel [19:04] happiest. Um then the data on [19:06] deterioration once you move from your [19:09] home to a facility is terrible death [19:12] within 90 days is the average [19:16] [laughter] you know not a good number we [19:19] wish on our families and the people we [19:21] love. Um so I I think we are going to [19:24] start seeing the system if you will the [19:27] health care system start putting dollars [19:29] to bear on using these type of [19:31] technologies that are not expensive at [19:33] all comparably or even object just [19:36] purely objectively like we charge 50 [19:38] bucks a month for a product. It's not [19:39] expensive. Um, so I think we're going to [19:43] start seeing a shift when instead of [19:46] fearing and to thinking about the [19:48] dystopian aspect, we actually think of [19:49] how this can help and uh help reduce [19:52] costs out of the system itself, you'll [19:54] see us expanding internationally outside [19:56] of the US. And then on on a wider topic, [20:00] if any of you are building robots and [20:01] are wondering, have these thoughts about [20:03] how do I build a relationship at the [20:05] home, then we'd love to help because [20:08] just for the reasons I mentioned before, [20:11] I think the category is huge. If you ask [20:13] I think anybody here in 10 years, what [20:16] do we imagine? I think we all imagine [20:17] we'll be living with robots or you know [20:20] whether they're embedded in our glasses [20:22] or our homes or but they'll be there. Um [20:25] but that's not going to happen [20:26] automatically. There are a lot of things [20:27] we need to do in order for that to [20:28] happen and it will be a shame if we mess [20:31] it up. [20:32] >> Dor, thank you so much for your time [20:34] today for sharing LEQ with us. It is [20:36] clear that you're using AI for good and [20:38] a light onto the world. So, thank you so [20:40] much for joining us today. Thanks to all [20:41] of you. Please give door a round of [20:43] applause. [20:44] >> Thanks for having me, for joining us. [20:46] Thank you.