User Experience Design for Wearables

Apple is accelerating development of 3 new AI powered wearable devices: smart glasses, a camera equipped pendant, and enhanced AirPods. For consumers, this is exciting. For security, it’s alarming.   All three devices are expected to connect to Siri and collect surrounding data through camera and sensor systems with varying capabilities. On the surface, that sounds helpful. But when AI moves into wearables, the security boundary moves with it. Sensors are no longer devices you pull from your pocket. They are persistent, attached to the body, and continuously collecting context. That significantly expands the attack surface for both individuals and enterprises. I am all pro innovation. But innovation at this layer must be built on clear threat models and an informed understanding of risk.   Some concerns are obvious. On device AI must operate with strict permission controls. If contextual data is routinely offloaded to cloud services, exposure increases. Data flows, storage models, and third-party processing become part of the risk equation.   Other risks are less obvious. Continuous visual and audio capture can record people and spaces without clear consent. Bystanders become part of the dataset. Wearables also increase biometric signal collection, often without clearly defined retention and deletion policies. Regulators are already scrutinizing consent models, data minimization practices, and user erasure rights in AI systems.   Trust in a platform is not about brand perception. It depends on transparency around what is captured, how it is processed, who can access it, and whether users have meaningful control to delete it.   Questions worth asking now:   Will these devices limit recording to explicit actions or defined contexts? How will bystander privacy be protected? What encryption and access controls govern the full data pipeline?   If these products are intended to become mainstream tools, privacy and security architecture are the kind of conversations I want to have now. What do you think? Would you feel comfortable wearing a device that continuously analyzes your surroundings? https://lnkd.in/gw4-QhXJ

 I watched NeoSapien's AI wearable pitch on Shark Tank India last night - and it got me thinking... Even tech GIANTS are struggling with what this startup is about to face. The problem? DATA PRIVACY. When your device tracks conversations, emotions, and basically becomes your "second brain" - where does all that sensitive data go? → Apple Watch faces scrutiny over health data security → Google's Fitbit criticized for unclear data sharing practices → And now startups like Neosapien entering this complex space As someone who's helped hundreds of tech startups, I've noticed a major shift... Investors aren't just evaluating cool tech anymore. They're scrutinizing how you'll handle data privacy BEFORE writing checks. The wearable tech market is projected to hit $493 billion by 2030. But the companies that will secure funding aren't just the most innovative ones... They're the ones with rock-solid data protection strategies. 3 things I advise my fundraising clients in this space: 1️⃣ Build privacy into your product DNA, not as an afterthought 2️⃣ Create transparent data policies users can actually understand 3️⃣ Stay ahead of regulations like GDPR and HIPAA (they're constantly evolving) In my experience, a strong data privacy approach isn't just good ethics, it's becoming a DEAL-BREAKER for securing investment. What do you think? Will data privacy concerns slow innovation in wearable tech? Or push it to evolve in better ways? #StartupFunding #DataPrivacy #WearableTech #SharkTankIndia

Yesterday, we explored how multimodal AI could enhance your perception of the world. Today, we go deeper into your mind. Let's explore the concept of the Cranial Edge AI Node ("CortexPod"). We’re moving from thought to action, like a cognitive copilot at the edge. Much of this is already possible: neuromorphic chips, lightweight brain-sensing wearables, and on-device AI that adapts in real time. The CortexPod is a conceptual leap; a cranial-edge AI node that acts as a cognitive coprocessor. It understands your mental state, adapts to your thinking, and supports you from the inside out. It's a small, discreet, body-worn device, mounted behind the ear or integrated into headgear or eyewear: ⭐ Edge AI Chipset: Neuromorphic hardware handles ultra-low-latency inference, attention tracking, and pattern recognition locally. ⭐ Multimodal Sensing: EEG, skin conductance, gaze tracking, micro-movements, and ambient audio. ⭐ On-Device LLM: A fine-tuned, lightweight language model lives locally. These are some example use cases: 👨⚕️ In Healthcare or Aviation: For high-stakes professions, it detects micro-signs of fatigue or overload, and flags risks before performance is affected. 📚 In Learning: It senses when you’re focused or drifting, and dynamically adapts the pace or style of content in real time. 💬 In Daily Life: It bookmarks thoughts when you’re interrupted. It reminds you of what matters when your mind starts to wander. It helps you refocus, not reactively, but intuitively. This is some recent research... 📚 Cortical Labs – CL1: Blending living neurons with silicon to create biological-silicon hybrid computers; efficient, adaptive, and brain-like. https://corticallabs.com/ 📚 BrainyEdge AI Framework: A lightweight, context-aware architecture for edge-based AI optimized for wearable cognitive interfaces. https://bit.ly/3EsKf1N These are some startups to watch: 🚀 Cortical Labs: Biological computers using neuron-silicon hybrids for dynamic AI. https://corticallabs.com/ 🚀 Cognixion: Brain-computer interfaces that integrate with speech and AR for neuroadaptive assistance. https://www.cognixion.com/ 🚀 Idun Technologies: Developing discreet, EEG-based neuro-sensing wearables that enable real-time brain monitoring for cognitive and emotional state detection. https://lnkd.in/gz7DNaDT 🚀 Synchron: A brain-computer interface designed to enable people to use their thoughts to control a digital device. https://synchron.com/ The timeline ahead of us: 3-5 years: Wearable CortexPods for personalized cognitive feedback and load monitoring. 8-10 years: Integrated “cognitive coprocessors” paired with on-device LLMs become common in work, learning, and well-being settings. This isn’t just a wearable; it’s a thinking companion. A CortexPod doesn’t just help you stay productive; it helps you stay aligned with your energy, thoughts, and intent. Next up: Subdermal Audio Transducer + Laryngeal Micro-Node (“Silent Voice”) 

A wearable lie detector? – well, somebody had to build it! Our paper, titled ‘Wireless, skin-interfaced multimodal sensing system for continuous psychophysiological monitoring—A wearable polygraph device,’ published in Science Advances (https://lnkd.in/gUa2UVCp), and highlighted in this press release from Northwestern University (https://lnkd.in/gj5inaVW), introduces a device capable of precisely measuring the full suite of parameters (and beyond!) captured by state-of-the-art polygraph systems, but in a soft, wireless wearable form that mounts on the chest. Our emphasis is not on lies specifically, but instead on stress in general, and particularly for its relevance to medical care, as demonstrated in various representative cases – from pain experienced by infants, to sleep disruptions in babies with Down syndrome, to confusion in adults with hearing impairments, to strains experienced by medical students in training for emergency room care. The devices leverage the concepts of hybrid soft electronics for continuous measurements of body sounds, body movements, skin impedance, temperature and thermal transport. The results yield heart rate, heart rate variability, respiration rate, variability and depth, sweat gland activation, cardiac amplitude (as a rough surrogate for blood pressure), near-surface blood flow and skin temperature. Machine learning algorithms rely on these data streams to determine stress. The paper presents not only a broad range of uses, as mentioned above, but also validations – from correlations to commercial polygraph instruments, to tools for pupillometry, to nurse scoring sheets, to polysomnography systems. Fun project, with many additional applications in sports, worker safety, etc – complementing previously published work from our group and impressive papers from other leading teams that consider stress biomarkers in sweat, but without the cumbersome process of collecting and chemically analyzing this class of biofluid. In fact, the platform introduced here involves only biophysical sensors, bypassing the need for disposable components and avoiding the various confounds (hysteresis, drift, lack of specificity, inadequate sensitivity and others) associated with biochemical measurements. Thanks to former postdoctoral fellows in the group – Prof. Sun Hong Kim (University of Seoul), Prof. Jae-Young Yoo (SKKU), Prof. Tianyu Yang (ASU), Prof. Seonggwang Yoo (Inje University) and Dr. Seunghee Cho (Samsung) – for their leadership and to many current group members – Dr. Tae Wan Park and others -- for their contributions. Also grateful to our collaborators across the medical school, specifically Prof. Debra Weese-Mayer, Dr. Khaytin Ilya and Dr. Jana Jaffe, and those in the Department of Communication Sciences and Disorders. Finally, thanks to Amanda Morris for the nice writeup and press package for the Northwestern release. 

That small LED on the Ray-Ban Meta glasses? Most people selling them can't explain what it means. A joint investigation by Swedish journalists just revealed that footage captured by these glasses — when users invoke the AI assistant — is reviewed by human workers at a subcontractor in Nairobi, Kenya. Workers described seeing bathroom footage, people undressing, bank cards, intimate moments. All captured by someone who thought they were just asking their glasses a question. The person wearing the glasses consented to a privacy policy. Everyone else in the frame never got the chance to. This isn't just a Meta story. It's a preview of what happens when ambient AI recording meets inadequate consent frameworks at scale. Four things every security professional needs to understand about this: → Wearable AI devices are ambient surveillance by design. A single LED is not a sufficient consent mechanism. → "Processed locally" is a marketing phrase. Packet analysis on these glasses found constant communication with Meta servers — contradicting what store employees were telling customers. → Third-party data annotation is the invisible layer of every major AI product. The people reviewing your footage are not Meta employees, may operate under weaker data protections, and are bound by NDAs. → Bystander consent in wearable tech is still an unresolved legal gap. NOYB and EPIC have both filed actions. Watch this space. We are putting cameras on faces before we've built the legal and ethical infrastructure to govern them.

The skin feels not only pressure but also interprets temperature, movement, intensity, threat, comfort, and contact as part of the same experience. This is why the article “Microfluidic-enabled stretchable thermoelectric device array for multimodal haptic interfaces”, published in Microsystems & Nanoengineering, is particularly interesting. The authors address a key limitation of thermal haptic interfaces. Stretchable thermoelectric devices can modulate skin temperature, but cooling capacity and heat dissipation often limit their performance. At the same time, simple thermal stimuli are not enough to reproduce the richness of real tactile perception. Their solution is a stretchable thermoelectric device array with embedded microfluidic architectures, designed to improve thermal flux and enable rapid, precise, closed-loop regulation of skin temperature. By coordinating multiple devices and varying temperature range, frequency, spatial distribution, and stimulation patterns, the system can elicit more complex sensations, including pressure, pain, and sliding. This work suggests that future haptic interfaces will not rely only on vibration or force feedback. They will need to speak a richer sensory language, closer to the one used by the body. For prosthetics, rehabilitation, virtual reality, and wearable technologies, this means moving toward interfaces that do not simply notify the user that something happened, but recreate a more embodied experience of interaction. 👇 Link in the first comment #haptics #wearabletechnology #bioelectronics #thermoelectrics #tactilefeedback #sensoryfeedback

Privacy, as we once understood it, is dead. Every time we tap "Accept All Cookies" or scroll past a terms-of-service agreement to download a fitness app, we hand over intimate details about our bodies, our habits, our vulnerabilities. We present a compelling case for transparency mandates around health data transactions. The uncomfortable starting point is one the paper dances around: the traditional framing of privacy as something we can protect through consent and de-identification is largely a fiction. Our health records, wearable data, and genomic information are already circulating through a commercial ecosystem most of us never agreed to and barely understand. The real question isn't how to lock the barn door; it's who took the horse, where did they ride it, and who got paid along the way. What we need is a disclosure framework built on that honest foundation: Who is selling our data? What are they doing with it? Who is profiting? And who is being harmed? That kind of radical transparency won't restore privacy in any nostalgic sense, but it can restore something arguably more important: accountability. And accountability, specifically, relational accountability, unlike privacy, is something we can still fight for. https://lnkd.in/erPcbYKn

Google just released a foundation model that learns directly from incomplete wearable sensor data without any imputation. Wearable health data is notoriously fragmented. Existing AI models typically rely on imputation or discard incomplete samples altogether. But that’s not scalable to real-world, day-long multimodal sensor streams. 𝗟𝗦𝗠-𝟮 𝘄𝗶𝘁𝗵 𝗔𝗜𝗠 𝗶𝘀 𝗮 𝗳𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻 𝗺𝗼𝗱𝗲𝗹 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻 𝗿𝗼𝗯𝘂𝘀𝘁 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 𝗳𝗿𝗼𝗺 𝗶𝗻𝗰𝗼𝗺𝗽𝗹𝗲𝘁𝗲 𝘄𝗲𝗮𝗿𝗮𝗯𝗹𝗲 𝗱𝗮𝘁𝗮 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗶𝗺𝗽𝘂𝘁𝗮𝘁𝗶𝗼𝗻.  1. Pretrained on 40M hours of day-long multimodal sensor data from 60,000+ people; none of which had 100% data completeness.  2. Introduced a dual-masking approach (Adaptive + Inherited) to model real-world missingness using learnable tokens instead of filling gaps.  3. Outperformed prior foundation models across 10 downstream tasks spanning classification (hypertension, anxiety), regression (age, BMI), and generative imputation.  4. Preserved physiological signal importance (e.g., removing nighttime signals dropped hypertension F1 by 5%), while removing daytime signals had near-zero effect.  5. Maintained performance even when key sensors were removed, showing 73% smaller degradation and +15% higher accuracy under missingness compared to LSM-1. It's cool that the authors position AIM as a 𝗿𝗲𝗽𝗿𝗲𝘀𝗲𝗻𝘁𝗮𝘁𝗶𝗼𝗻 𝗹𝗲𝗮𝗿𝗻𝗲𝗿 𝗳𝗼𝗿 𝗵𝗲𝗮𝘃𝗶𝗹𝘆‐𝗺𝗶𝘀𝘀𝗶𝗻𝗴 𝘄𝗲𝗮𝗿𝗮𝗯𝗹𝗲 𝘀𝘁𝗿𝗲𝗮𝗺𝘀, contrasting it with prior work that either (i) treats missingness in simpler tabular data or (ii) tackles irregularly‑sampled EHR events. Also, cool to see the 𝗮𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲 𝗰𝗵𝗼𝗶𝗰𝗲𝘀 𝗲𝗺𝗽𝗵𝗮𝘀𝗶𝘀𝗲 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝗹𝗲𝗻𝗴𝘁𝗵 rather than sheer parameter count:  1. A 25 M‑param ViT‑1D encoder (384‑d hidden, 12 × Encoder / 4 × Decoder) ingests an entire 24‑hour day (3 744 tokens) with a 2‑D positional scheme (time × sensor).  2. Union masking lets AIM keep sequence length high while pruning only the artificial‑mask tokens—avoiding the hard 𝐷‑=‑const assumption of classic MAE. Here's the awesome paper: https://lnkd.in/gcgmRYZp Congrats to Maxwell Xu, Girish Narayanswamy, Kumar Ayush, Xin L., Daniel McDuff, and co! Free access to the best LLMs and ask any medical questions here: labela.ai/healtholymp I post my takes on the latest developments in health AI – 𝗰𝗼𝗻𝗻𝗲𝗰𝘁 𝘄𝗶𝘁𝗵 𝗺𝗲 𝘁𝗼 𝘀𝘁𝗮𝘆 𝘂𝗽𝗱𝗮𝘁𝗲𝗱! Also, check out my health AI blog here: https://lnkd.in/g3nrQFxW

Last week, New Yorkers staged "interventions" with $1 million worth of subway ads for an AI wearable device called "Friend." Unlike with the circa 2012 ads for the Mad Men tv show, where people got creative with the empty space; this time, they were angry. (Something the creative campaign may have invited, by leaving so much white room in their ads.) "Friend" costs $130 and looks like a pendant, which you wear as a necklace. It listens to your conversations and texts messages to your phone, ranging from supportive to snarky (reportedly styled after the founder's personality). NYC's response was swift: Friend will "steal your data" "steal your identity" Friend (definition): "mass surveillance" So, what does Friend actually do with user data? ⚠️ AI TRAINING Looking at its Terms and Privacy Policy, both make clear that the company may use conversations "to train and improve our artificial intelligence models and machine learning systems . . . " ⚠️ BIOMETRIC INFORMATION COLLECTION Friend's Privacy Policy states it is collecting biometric data, such as voice recognition, from users. The Privacy Policy does not explain the purpose of the collection, or for how long this biometric data is stored. And it does not explain what Friend will do with the biometric information that is captured from OTHER people (not wearing the device) in the user's surroundings. ⚠️ LIABILITY ON THE USER Friend's Terms state that it is up to the individual wearing the device to comply with state privacy laws. We see this a lot in tech consumer products. (Do any non-privacy lawyers know their state's privacy laws?) Here, that means that the user would have to: ➡️ inform people that they are recording the conversation (if they are in an all-party consent state like California), ➡️ get their consent for the recording, and ➡️ also get their consent for the collection of biometric data (if they are in a state with a biometric information privacy law), when Friend has not even explained how that biometric data may be used and for how long it may be stored. PRIVACY AS FORETHOUGHT VS. NOT EVEN AN ADD-ON? Far too often, companies are shifting the burden of privacy-compliance downstream, onto their users. And products, instead of using privacy-by-design from the beginning, are lacking basic privacy features (as add-ons), all while engaging in massive data collection. It seems, without even having read Friend's Terms and Privacy Policy, people are fed up with it. All friendships take work, but this seems like a lot.

🛏️ Revolutionizing Sleep Monitoring: From Wristbands to Real Biomarkers Traditional wearables fall short when it comes to sleep diagnostics—they miss direct respiratory signals, which are vital for identifying sleep stages and disorders like sleep apnea. This new research introduces a low-power, skin-integrated mechanoacoustic (LMA) sensor that changes the game. It doesn't just guess your sleep from motion—it listens to your breath and heartbeat. https://lnkd.in/g_yZW69p 🔬 Key innovations: - Multimodal sensor captures respiratory rate, heart rate variability (HRV), respiration rate variability (RRV), body movement, and more. -Paired with LMA-SleepNet, an interpretable machine learning model that detects sleep stages and apnea events with clinical-grade accuracy. - Uses physiology-based features like baroreflex and muscle tone—giving deeper insights than motion-based trackers. Outperforms other wearables in real-world accuracy. 📊 Why this matters: - Directly measures respiration—a core but missing biomarker in most wearables. - Enables continuous, personalized, and explainable sleep tracking in the home or clinic. - Opens doors for smart OSA detection, snoring tracking, and even future on-body therapeutic interventions. 💡 Bonus: Real-time UTC synchronization allows scalable multi-sensor studies across environments and populations. 🔁 This is more than a device—it’s a complete hardware-software platform for next-gen sleep and health monitoring, with huge potential for precision healthcare, chronic disease management, and behavioral science. 👀 Sleep isn’t just rest—it’s integrated physiological data. And now, we can measure it better than ever. #SleepScience #WearableTech #DigitalHealth #MachineLearning #SleepApnea #RespiratoryHealth #PrecisionMedicine #Bioengineering #HealthTech #HRV #RRV