Keynote-Lars Reger
19 Feb 2026 16:45h - 17:00h
Keynote-Lars Reger
Session at a glance
Summary
Lars Reger, Chief Technology Officer of NXP Semiconductors, delivered a presentation on the future of artificial intelligence at the edge, focusing on how AI will transform everyday devices and systems. Reger argued that while much attention is given to AI in data centers, the real transformation will come from AI-powered devices that create a world that “anticipates and automates” rather than simply responding on demand. He envisions a future where homes, manufacturing facilities, and vehicles become intelligent “cocoons” or robots that can sense, think, connect, and act autonomously while maintaining complete safety and security for users.
Drawing parallels to biological systems, Reger explained that effective AI robots should mimic the human body’s architecture, with different layers handling various functions – from real-time reflexes to higher-level cognitive tasks. He emphasized that most AI applications will require relatively small, efficient models rather than massive systems, comparing transportation robots to insects that operate effectively with only 100,000 neurons. Reger demonstrated NXP’s approach through small, powerful devices that can control drones and process large language models while consuming only 7 watts of power.
The presentation highlighted that many autonomous vehicle failures occurred not due to AI problems but because of inadequate sensing capabilities. Reger advocated for giving robots “superhero senses” through technologies like ultra-wideband communication, advanced radar systems, and enhanced audio processing that can detect sounds beyond human capability. He concluded that the democratization of AI will happen through edge devices rather than data centers, enabling AI access for everyone while solving critical energy efficiency challenges that would otherwise require unsustainable power consumption.
Keypoints
Major Discussion Points:
– Evolution from analog to digital to AI-powered world: Reger describes the progression from analog systems (1970s) to digitization (smartphones) to the current phase of AI that “anticipates and automates,” driven by megatrends like demographic changes, infrastructure upgrades, and energy constraints.
– Architecture of intelligent systems based on biological models: The speaker advocates for copying nature’s approach, comparing human biological architecture (spine for reflexes, cerebellum for basic functions, brain for complex thinking) to how AI systems should be designed with layered, distributed intelligence rather than centralized processing.
– Edge AI vs. data center AI: A central argument that 80% of AI tasks should be performed on small, efficient, tailored models at the “edge” (in end devices) rather than in large data centers, emphasizing energy efficiency and democratization of AI technology.
– Trust as the foundation of AI adoption: Reger emphasizes that functional safety and cybersecurity are essential for AI systems to be trusted and adopted, using examples of autonomous vehicles and smart home devices that could fail catastrophically without proper safeguards.
– Superhuman sensing capabilities for robots: Discussion of advanced sensing technologies (ultra-wideband, radar, car-to-car communication) that can give AI systems capabilities beyond human senses, enabling better performance in autonomous vehicles and other applications.
Overall Purpose:
The discussion aims to present NXP Semiconductors’ vision for the future of AI hardware, specifically advocating for distributed, edge-based AI systems that are energy-efficient, secure, and capable of democratizing AI technology globally rather than concentrating it in large data centers.
Overall Tone:
The tone is enthusiastic and visionary throughout, with Reger maintaining an optimistic, forward-looking perspective. He uses accessible analogies (comparing humans to “biological robots,” referencing superheroes and pop culture) to make complex technical concepts understandable. The tone remains consistently confident and promotional, positioning NXP as a key enabler of this AI-powered future while addressing practical concerns about energy consumption and trust.
Speakers
– Moderator: Role/Title: Discussion moderator; Area of expertise: Not specified
– Lars Reger: Role/Title: Executive Vice President and Chief Technology Officer, NXP Semiconductors; Area of expertise: Semiconductor design, edge AI hardware, secure and efficient AI systems for cars, medical devices, and industrial systems
Additional speakers:
– Ashwini Vishnath: Role/Title: Not specified; Area of expertise: AI (mentioned as having spoken about AI tasks at the edge in Davos)
– PM Modi: Role/Title: Prime Minister (implied); Area of expertise: Government policy, AI democratization initiatives
Full session report
Lars Reger, Chief Technology Officer of NXP Semiconductors, delivered a comprehensive presentation challenging the prevailing focus on centralised artificial intelligence development, instead advocating for a paradigm shift towards edge-based AI systems that will create a world that “anticipates and automates” rather than merely responding to on-demand requests.
Opening Context and Industry Position
Beginning with a “Namaste” greeting, Reger was introduced as the leader of NXP’s edge AI initiatives across automotive, medical devices, and industrial systems. He contextualised the current AI revolution within his personal technological journey from the analogue world of the 1970s through the heavy digitalisation of the past two decades, culminating in smartphones that transformed laptops into portable data display devices. He argued that whilst the previous era was characterised by on-demand services—ordering pizza, calling an Uber, or controlling home climate systems—we are now entering a fundamentally different phase where technology anticipates user needs and automates responses.
This transformation is being driven by persistent megatrends including demographic changes, infrastructure upgrades, supply chain constraints, renewable energy transitions, and energy limitations that have remained consistent over the past 15 years. However, Reger posed a provocative challenge to the current AI discourse: whilst there is extensive discussion about powering massive data centres, fundamental questions remain about what AI is actually accomplishing and its practical purpose in improving human lives.
Vision of an AI-Powered Future
Looking ahead 20 years, Reger painted a detailed picture of how AI will transform three critical domains. In residential settings, homes will become completely barrier-free environments that continuously monitor occupants’ health and wealth whilst providing seamless protection. These intelligent shelters will allow residents to live without touching anything, creating maximum safety and security for authorised users whilst remaining impenetrable to others.
The manufacturing landscape will undergo equally dramatic transformation, with most manual tasks eliminated and workers evolving into advanced equipment operators managing sophisticated autonomous systems. Reger drew a compelling parallel to aviation, noting that pilots 70 years ago were physically robust individuals who manually fought through thunderstorms, whereas today’s pilots—representing all genders, shapes, and sizes—primarily operate intelligent flying robots, with actual manual control required for only 30 seconds during takeoff on a typical flight from Germany to India.
Transportation will evolve into “rolling cocoons”—autonomous vehicles that function as mobile living rooms and office extensions. Reger observed that during the COVID pandemic in China, many people already used their cars as office spaces to escape crowded homes, demonstrating the potential for vehicles to become personalised, anticipatory environments that understand and automate occupants’ needs. He noted that on German highways, where speeds reach 250 kilometres per hour, AI-powered systems with superhuman sensing capabilities would far exceed human driving abilities.
Universal Architecture and Trust Requirements
Despite the apparent diversity of these applications, Reger identified four universal functions that all intelligent systems must possess: sense their environment, connect to cloud-based data sources, think through optimal responses using AI, and act through various mechanical or digital interfaces. Whether these “arms and legs” manifest as automotive powertrains, manufacturing equipment, or wireless connections to smart home devices, the fundamental architecture remains consistent across all 50 billion smart connected robots he predicts will exist within a decade.
Crucially, Reger emphasised that none of these capabilities matter without trust—the foundational requirement for AI adoption. He illustrated this principle through practical examples: if a refrigerator orders 500 litres of milk, users will revert to manual shopping; if a car drives erratically, drivers will resume manual control; if a thermostat sets the house to 50 degrees Celsius, killing plants and pets, homeowners will abandon automation entirely. Trust, he argued, requires two essential components: functional safety (ensuring critical systems like automotive braking never fail) and cybersecurity (preventing connected devices from being compromised or hacked).
Learning from Autonomous Vehicle Failures
To address the failures of current autonomous systems, particularly the disappointing progress in self-driving vehicles that were predicted to transport children to kindergarten without human supervision by 2020, Reger identified a critical flaw in the prevailing approach. Rather than attributing accidents to flawed AI algorithms, he determined that inadequate sensing capabilities were the primary cause—these systems were “more short-sighted than I am.”
Biological-Inspired System Architecture
Reger advocated for copying nature’s proven architectures, describing humans as “90-kilogram bags of water with a couple of bones”—essentially biological robots with sophisticated layered architectures that AI systems should emulate. The human spine provides real-time reflexive responses, immediately commanding leg adjustments when stumbling without any conscious thought or AI processing—a highly deterministic, functionally safe system. The cerebellum manages autonomic functions like heartbeat, stomach control, and stability maintenance, allowing a person to stand upright whilst the brain focuses on complex cognitive tasks like formulating speech.
This distributed architecture, with different systems handling appropriate levels of complexity, should be replicated in artificial systems rather than attempting to centralise all intelligence in a single processing unit. Reger noted that whilst humanoid robots capture public imagination, they represent only “the tiny fraction” of future intelligent systems.
The Case for Edge AI and Energy Efficiency
Reger’s most compelling argument centred on the impracticality of centralised AI approaches. Referencing Ashwini Vishnath’s observation from Davos, he noted that 80% of AI tasks will be performed on tiny, efficient, tailor-made models at the “edge”—within end devices rather than remote data centres. This approach is essential for energy sustainability: deploying 50 billion smart connected devices using current AI architectures would require three times the energy that Earth can provide.
To demonstrate the viability of edge AI, Reger showcased NXP’s hardware solutions during his presentation, including a complete drone control unit capable of autonomous flight with AI-powered navigation and target recognition, and an AI accelerator developed by Kinara in Hyderabad (subsequently acquired by NXP) that can process 10 billion parameters in a large language model whilst consuming only 7 watts of power. This energy efficiency is achieved through intermittent processing—the system only consumes power when actively analysing situations and returns to sleep mode afterwards.
Advanced Sensing Capabilities: Superhero Senses for Robots
Reger’s key insight was that robots require “superhero senses” that exceed human capabilities. He outlined several advanced sensing technologies already available: ultra-wideband technology enabling seamless device communication and access control; car-to-car communication systems operating over distances exceeding one mile with three-millisecond response times, allowing vehicles to coordinate traffic light changes for emergency vehicles; radar systems capable of detecting two people sitting adjacent to each other at distances over 300 metres, even in rain, snow, and fog; and meta standards creating a common language for smart connected devices.
These capabilities provide the foundation for truly superhuman robotic abilities: telepathic communication (like Dumbledore’s magic wand control), prescient awareness beyond line of sight (like Yoda’s telepathy), X-ray vision through adverse weather conditions (like Superman), universal language translation (referencing the Babel fish from Hitchhiker’s Guide to the Galaxy), and enhanced auditory capabilities (surpassing Daredevil or owls).
Practical Applications and Modular Design
Reger provided concrete examples of these principles in action. Using automotive microphones and AI processing, NXP has developed systems that can detect bicycle bells behind vehicles and identify vulnerable road users through audio analysis—capabilities that extend far beyond automotive applications. The company’s approach involves creating scalable “Lego brick” components that can be combined for different applications. A basic drone control unit can be enhanced with additional AI processing modules to enable sophisticated autonomous navigation through complex environments like forests.
Optimal AI Model Sizing
Addressing the fundamental question of AI complexity, Reger noted the spectrum from ants with 100,000 neurons (sufficient for effective transportation tasks) to human brains with 90 billion neurons. Most applications, he argued, require far less computational complexity than commonly assumed, enabling the energy-efficient edge deployment that makes widespread AI adoption feasible.
Democratisation and Global Implications
Reger concluded by directly addressing Prime Minister Modi’s vision of bringing AI to everyone. The solution, he argued, lies not in expanding data centre capacity but in democratising AI through edge devices. This approach enables AI access in developing regions without requiring massive infrastructure investments or unsustainable energy consumption.
The semiconductor industry’s role involves continuing its 50-year tradition of scaling and optimisation: developing safe and secure architectures with ultra-low power consumption, pushing the boundaries of physics to create sensing capabilities that exceed human abilities, and building modular components that can be combined to create intelligent systems across diverse applications.
Reger’s presentation fundamentally challenges the dominant narrative of AI development, advocating for distributed intelligence that brings AI capabilities directly to end users whilst solving critical energy and accessibility challenges. His vision suggests that the future of AI lies in democratised, efficient edge devices that can provide AI capabilities globally whilst maintaining the trust, safety, and security essential for widespread adoption.
Session transcript
And now I would like to invite Mr. Lars Reger, Executive Vice President and Chief Technology Officer, NXP Semiconductors. As we all know, artificial intelligence runs on chips, and Lars Reger is at the frontier of designing the semiconductors that will power the next generation of edge AI. In cars, in medical devices, in industrial systems. NXP’s work on secure, efficient, real -world AI hardware is essential to everything on the stage. Ladies and gentlemen, please welcome the Chief Technology Officer of NXP Semiconductors, Mr. Lars Reger.
Namaste. Hello everyone and thanks for having me here. When we are talking about AI, at the moment there is a lot of talk about how do we pump AI in big data centers, how are we energizing these big data centers, but very honestly, there is a lot of questions. What is this AI for? What is this AI at all doing? And if I’m looking at my own lifespan, I’m coming from an analog world, was born in the 1970s. Then there was some heavy digitization in there over the last 20 years, when someone stuffed a laptop into… into a mobile phone and they called it smartphone. So we had a data display device. We could run topics that were on demand.
So on demand, I need a pizza, I need an Uber, I need to switch on the climate control in my house. And now my Marcom people would say, Lars, we are entering a phase of the world that anticipates and automates. And this little world that anticipates and automates is driving megatrends around us. And these megatrends are unchanged over the last 15 years. We have demographics changes. We have infrastructure upgrades. We have supply chain constraints. We have renewable and we have energy constraints. So out of all of these drivers, what is this modern world that anticipates and automates able to do for us? Well, jumping forward maybe 20 years, how is the cocoon that I’m living in going to look like?
I will have a shelter. I will have my house, and that house is totally barrier -free. That house will check about my health, my wealth, will protect me. I can enter and I can live. I can live. without touching anything. No one else can do the same and my property is protected very seamlessly. No barriers for me, but maximum safety and security. How will be my manufacturing landscape look like? Well, most of the manual tasks are gone. I need better education and I may be the most advanced equipment operator in the world. Look at airplane pilots 70 years ago. They were guys my size. These type of muscles and arms were flying in thunderstorms, real heroes, mechanical pilots.
Today we have more pilots, but they are all genders, shapes and sizes because they are operating flying intelligent robots. So when I come from Germany here to India, a pilot, mechanically, I’m not a pilot. has to work for 30 seconds at the end of the runway, pull up the plane, and all the rest is happening already today autonomously. And that’s going to get better in the industrial world. And of course, also in the transportation world. How are cars going to look like in 20 years? Well, they are rolling cocoons, rolling robots, and these cars are rolling living rooms. You have seen this during the COVID pandemics in China, for example. A lot of people use their cars as their office extensions.
Too many people in the house at home, the kids were too noisy. You go to these type of places, so you have a rolling cocoon again that is anticipating and automating what you want to do, what you want to achieve. And what does this all have in common? I mean, most of the people are asking me now, okay, Lars, nice. You are predicting that there is 50 billion of these smart connected robots out there in 10 years from now. But they have so different form factors. What does that mean? Well, simple answer. They have all the same ingredients. Each of these little robots has to sense its environment. So what’s happening around me? Has to connect to the cloud to get the data.
Last ones to drive from here to Mumbai, how is the traffic situation? Getting the information from the web. And then you start thinking of a smart advice. This is where AI comes into the play. At that moment, you have to think of what is my best advice to the arms and legs to my robot. And whether these arms and legs are an automotive powertrain and a steering wheel, is a manufacturing arm, or is the wireless connection to my climate control from my smart thermostat. I don’t care. Sense, think, connect, act are the ingredients for every of these 50 billion robots. Now, the only thing is, that all is nothing if you cannot trust. Because if your fridge starts ordering 500 liters of milk for the next weekend, you go shopping alone if your car does erratic driving you start driving manually again and if your thermostat sets your house on 50 degrees centigrade and your flowers are dried out and your cat is dead you go organizing it all manually again so trust is the essence and how does a nerd like me define trust that’s very simple this is functional safety like in automotive make sure that your braking system never ever fails and make sure that your connected device your car or whatsoever is never ever being hacked and then you can trust your device you can be sure that it doesn’t turn against you so these underlying levels make sure that you are energy efficient because otherwise you cannot be battery powered make sure that you are trustworthy so safe and secure you and then make sure that you can sense, think, connect, and act.
And you can build every robot in the world. And that is, of course, interesting for a semiconductor maker because for us, volume matters in these semiconductor chips. Now, you will ask me, but Lars, we have so long already these discussions on autonomous vehicles. In 2018, the entire press community thought, in 2020, my kids are going to the kindergarten without a steering wheel and without me, autonomously. That did not happen. Why? Because we have designed these robots wrong. And how do you design the robots right? Well, try to copy from nature. That normally works. And here on stage is a 90 -kilo bag of water with a couple of bones, or in other words, a biological robot.
And that robot has a certain architecture. That robot has different layers. That robot has a real -time system, highly functional safety, and that is my spine. And that is my spine. and if I stumble, the reflexes in my spine tell me already straighten your leg. In real time, highly undisturbed, very, very fast. No AI, not big AI, very deterministic system. Then I have in green my cerebellum that is working also in a highly functional, safe environment for heartbeat, stomach control, stability control. I can stand here and stand in a stable way because only the blue part is trying to find out what is the next sentence that I’m firing towards you. And green and orange are working to manage the infrastructure in a functional, safe way that is standing here in front of you.
So why don’t we copy these approaches into vehicles, into cars, into houses, into planes again? Well, there are simple architectural constraints and building mechanisms. There are building blocks that we need and we need to scale. But how big does the AI really have to be? So that AI in these systems. can be comparatively tiny. If you’re talking about transportation robots and how these transportation robots should look like, well, look at intelligent transportation robots, insects, for example. These insects have 100 ,000 neurons and an ant is already a very, very nice, very sexy transportation device. It’s not as intelligent as a human being with 90 billion neurons, but for most of the tasks, it is also not needed in this way.
And Ashwini Vishnath said it very nicely in Davos. 80 % of the AI tasks around us will be on very tiny, efficient, and very, very tailor -made models at the famous edge, so in the end devices. And this is what we are designing for. So in other words, NXP is trying to build all these Lego blocks where you can start scaling small, medium, and large devices. You have these devices here. This is, for example, sorry, very small devices. This is a complete drone control unit. And this is a complete drone control unit. that also flies with AI, artificial intelligence, and reaches targets, not only remote control, but is operating the entire drone and is finding via the camera its way.
What I have here is an India -made AI accelerator from Kinara and Hyderabad that NXP has acquired. This is carrying 10 billion parameters in a large language model. So it is not as big as JetGPT. But the combination of those two systems carries a large language model and operates an intelligent system at the edge for a power consumption of 7 watts. So in other words, you can build these type of plug -in combinations and you have a system, for example, at a computer tomograph that is taking my entire X -ray pictures and is writing the doctoral report that is operating at my fridge and tells me how many bottles of milk are missing. you do not have to have it always on and always operational so these seven watts are only consumed the moment where the fridge tries to find out what is missing and then you can go to sleep again that is the answer for this global quest of how many nuclear power plants do we need when we send one question to chat gpt and that is what the edge is going to solve for all of us but beyond all of that we are always talking about ai and the brain structure of these robots most of the cars that have created fatalities in the last 10 years these autonomous vehicles didn’t create these fatalities on the roads because they had a bug in the brain structure they created these issues because they were more short -sighted than i so wouldn’t it be great to have these robots with superhero senses wouldn’t it be great if these robots out there would have telepathic capabilities?
I do not need to touch anything, but the stuff around me is arranging, like Dumbledore. One move of the magic wand, everything is arranged. Wouldn’t it be great to know what is ahead of your line of sight, like Yoda, telepathy? Wouldn’t it be great to have X -ray vision like Superman? You look in rain, in snow, and in fog what is around you. Wouldn’t it be great for the very old ones amongst us to be like in Hitchhiker’s Guide through Galaxy? You have one bubble fish that you plug into your ear and you understand the entire universe, every language that is spoken. A German can understand Hindi without a big barrier in between. And wouldn’t it be great if our robots would have better ears than Daredevil or an owl in real life and would be able to hear what is being spoken out there in the outer ranks?
If we would have that. then the driving robot that replaces Lars is way better than Lars the driver himself. But I am the entry ticket for driving 250 kilometers an hour on the left lane of a German highway with my car. Now, you think we cannot have that for our robots next to this little bit of AI that we need? Well, let me tell you, we have it already. We have ultra wideband technology that is opening gates and car keys from my watch to everything around me. I have car to car communication over more than one mile of distance in three milliseconds. I can immediately tell the device there is an ambulance rushing into the crossroads, switch the traffic lights to green for that ambulance and to red for me.
Telepathy. We have radar systems over 300 meters that see two persons sitting like you next to each other. And we can detect. I’m in rain and snow and in focus. We have meta standards. So the English for smart connected devices. There is a common language in place and all devices are talking to each other. The home gateway is talking to the window blinders, is talking to the solar cell management. This is the entry tickets for this democratization of AI functionality and for the entrance of these tiny devices here with a little bit of AI, a lot of functional, safe and secure architectures to building the right devices. And what we have done with a little bit of AI and a couple of microphones in cars, we can take the in -car microphones, the sound in a way that we hear a bicycle bell behind the cars.
And we can easily detect whether there is vulnerable road users, for example, behind the cars. We can do this in any other settings as well. But automotive is there a very nice one. So in other words, where are we at? At the moment, when I’m talking to my fellow nerds and the and the semiconductor researchers. it is not about ai alone it is how you can build systems that you absolutely can trust how can you go low power and then the key question how big does the brain has to be and the answer is somewhere between a hundred thousand and a hundred billion neurons beyond all of that there is very very interesting questions that we have to solve and where india is deeply with the europeans in research and in the activities how do we make the wiring harnesses how do we battery operate all of that how we are sensing in the right way how do we think so all of these separate topics and to not make it too nerdy and too complex all of these silicons here are driving then these form factors a lot of people are only talking about humanoids and sorry to say humanoids are the tiny fraction of robots because why should a robot look like a human being I mean, that only makes sense in a very human environment, climbing stairs or whatsoever.
Otherwise, you have robots that are looking like ultrasonic devices, that are looking like infant monitor devices on neonatology stations in hospitals. There is no need to look like me. But all of this, we are equipping already with silicon in the hundred thousands today, and the ingredients are always the same. And just to get this pitch here down on the runway, to say it in the drone language, what do you need to do? What do we need to work on? What does the industry do at the moment? Well, in a very simple way, we are working on safe and secure architectures that are ultra low power, ultra energy efficient. Again, otherwise, this dream of 50 billion smartphones.
Connected devices will not work. because these 50 billion smart connected devices need three times the energy that Mother Earth can provide. So that is the absolute must for these markets to come into play. Then what we need to work on is we have to push the boundaries, the envelope of physics, and we are doing. We are sensing better than human beings in the meantime. And then what we need to do is just a simple game that semiconductors have done since 50 years now. We need to scale in the right way. So we need to build these little Lego bricks and say, okay, here is a complete drone control unit that you can fly autonomously. You want to fly with large language models and very, very smart AI slalom between the trees.
Plug this little dongle in, and you have everything on board that you can do. And the same you can do for building control systems with manuals. You can do this for any form factor that you like. And that is what we are doing at the moment. while the AI models are getting much, much more efficient, smaller, and we carry them here. So my pitch is, when PM Modi says he wants to bring AI to everyone, this is the answer. The answer is not data centers. They will exist. But the democratization of AI and equipping everyone in Togo, as we heard earlier, or in India, or in Germany, with the right levels of AI that create the world that anticipate and automates, the answer lies at the famous edge in the end device.
Thank you.
Lars Reger
Speech speed
159 words per minute
Speech length
2776 words
Speech time
1042 seconds
AI as enabler of barrier‑free, automated daily life
Explanation
Reger envisions a future where AI removes physical and usability barriers in homes, creating environments that are fully accessible and safe for all occupants.
Evidence
“I will have my house, and that house is totally barrier -free.” [1]. “No barriers for me, but maximum safety and security.” [2].
Major discussion point
Vision of an AI‑driven, anticipatory world
Topics
Artificial intelligence | Social and economic development | Closing all digital divides
Projection of 50 billion smart robots with sense‑think‑connect‑act functions
Explanation
Reger predicts that within a decade there will be around 50 billion interconnected robots that continuously sense, think, connect and act, forming the backbone of an anticipatory world.
Evidence
“Sense, think, connect, act are the ingredients for every of these 50 billion robots.” [16]. “You are predicting that there is 50 billion of these smart connected robots out there in 10 years from now.” [17].
Major discussion point
Vision of an AI‑driven, anticipatory world
Topics
Artificial intelligence | Environmental impacts | Social and economic development
Functional safety and cybersecurity must guarantee devices never fail or get hacked
Explanation
Reger stresses that trustworthy AI systems require deterministic functional safety and robust cybersecurity so that devices never malfunction or become vulnerable to attacks.
Evidence
“this is functional safety like in automotive make sure that your braking system never ever fails and make sure that your connected device your car or whatsoever is never ever being hacked and then you can trust your device you can be sure that it doesn’t turn against you” [38].
Major discussion point
Trust, safety, and security as foundations for AI adoption
Topics
Building confidence and security in the use of ICTs | Artificial intelligence
Biological analogy: spine and cerebellum layers ensure trustworthy behavior
Explanation
Reger uses a nervous‑system metaphor, describing a robot’s real‑time safety layer as the “spine” and its adaptive control layer as the “cerebellum”, to illustrate deterministic, layered safety architectures.
Evidence
“That robot has a real -time system, highly functional safety, and that is my spine.” [27]. “And I have in green my cerebellum that is working also in a highly functional, safe environment for heartbeat, stomach control, stability control.” [48].
Major discussion point
Trust, safety, and security as foundations for AI adoption
Topics
Building confidence and security in the use of ICTs | Artificial intelligence
Mimicking natural layered architectures yields reliable, low‑latency robots
Explanation
Reger argues that copying nature’s hierarchical designs—layers, spine, cerebellum—produces robots that are both reliable and capable of ultra‑low latency operation.
Evidence
“That robot has different layers.” [25]. “Well, try to copy from nature.” [54].
Major discussion point
Designing AI hardware inspired by nature and edge efficiency
Topics
Artificial intelligence | Capacity development | Environmental impacts
80 % of AI tasks will run on tiny, efficient, tailor‑made edge models
Explanation
Reger predicts that the majority of AI workloads will shift from large cloud models to highly optimized, small models running directly on edge devices.
Evidence
“80 % of the AI tasks around us will be on very tiny, efficient, and very, very tailor -made models at the famous edge, so in the end devices.” [58].
Major discussion point
Designing AI hardware inspired by nature and edge efficiency
Topics
Artificial intelligence | Environmental impacts | Capacity development
NXP’s “Lego‑brick” modular AI blocks (e.g., drone unit, Kinara accelerator) at ~7 W
Explanation
Reger describes NXP’s strategy of offering plug‑in AI modules that can be combined to build complete solutions, such as a drone control unit running a large language model on only seven watts.
Evidence
“So in other words, NXP is trying to build all these Lego blocks where you can start scaling small, medium, and large devices.” [66]. “But the combination of those two systems carries a large language model and operates an intelligent system at the edge for a power consumption of 7 watts.” [68].
Major discussion point
Semiconductor strategy: modular, low‑power edge AI solutions
Topics
Artificial intelligence | The enabling environment for digital development | Environmental impacts
Ultra‑low‑power ultra‑wideband communication and common standards enable scaling to billions of devices
Explanation
Reger highlights ultra‑wideband, ultra‑low‑power communication and interoperable standards as key to deploying billions of AI‑enabled devices within Earth’s energy constraints.
Evidence
“We have ultra wideband technology that is opening gates and car keys from my watch to everything around me.” [46]. “because these 50 billion smart connected devices need three times the energy that Mother Earth can provide.” [18].
Major discussion point
Semiconductor strategy: modular, low‑power edge AI solutions
Topics
Environmental impacts | Artificial intelligence | The enabling environment for digital development
Edge AI as the practical path to “AI for everyone” (PM Modi’s vision)
Explanation
Reger positions edge AI as the answer to political calls for universal AI access, enabling affordable, localized intelligence in places like India, Togo, and Germany.
Evidence
“So my pitch is, when PM Modi says he wants to bring AI to everyone, this is the answer.” [15]. “But the democratization of AI and equipping everyone in Togo, as we heard earlier, or in India, or in Germany, with the right levels of AI that create the world that anticipate and automates, the answer lies at the famous edge in the end device.” [14].
Major discussion point
Democratization of AI across regions and societies
Topics
Artificial intelligence | Closing all digital divides | Social and economic development
Moderator
Speech speed
69 words per minute
Speech length
87 words
Speech time
75 seconds
AI hardware relevance to the broader agenda
Explanation
The moderator emphasizes that secure, efficient real‑world AI hardware underpins all the topics discussed on stage, linking semiconductor advances to the overall AI ecosystem.
Evidence
“NXP’s work on secure, efficient, real -world AI hardware is essential to everything on the stage.” [13].
Major discussion point
Vision of an AI‑driven, anticipatory world
Topics
Artificial intelligence | The enabling environment for digital development
Framing the role of semiconductors in the AI ecosystem
Explanation
The moderator frames the session by noting that artificial intelligence runs on chips and that Reger’s work at the semiconductor frontier is critical for next‑generation edge AI.
Evidence
“As we all know, artificial intelligence runs on chips, and Lars Reger is at the frontier of designing the semiconductors that will power the next generation of edge AI.” [62].
Major discussion point
Democratization of AI across regions and societies
Topics
Artificial intelligence | The enabling environment for digital development
Agreements
Agreement points
Edge AI is more practical and efficient than centralized data center approaches
Speakers
– Lars Reger
Arguments
Current AI focus on data centers raises questions about practical applications and purpose (Lars Reger)
80% of AI tasks will be performed on tiny, efficient, tailor-made models at the edge (Lars Reger)
AI democratization requires edge devices rather than centralized data centers (Lars Reger)
Summary
There is strong consensus that edge AI processing is superior to centralized data center approaches for most practical applications, offering better efficiency, accessibility, and real-world utility
Topics
Artificial intelligence | Environmental impacts | Closing all digital divides
Energy efficiency is critical for sustainable AI deployment
Speakers
– Lars Reger
Arguments
Energy efficiency is crucial as 50 billion devices would require three times Earth’s available energy (Lars Reger)
Edge AI solutions can operate large language models with only 7 watts of power consumption (Lars Reger)
Summary
There is clear agreement that energy efficiency is not optional but essential for AI scalability, with edge solutions providing the necessary power efficiency
Topics
Environmental impacts | Artificial intelligence
Trust through safety and security is fundamental for AI adoption
Speakers
– Lars Reger
Arguments
Trust is essential for AI adoption, requiring functional safety and cybersecurity (Lars Reger)
Real-time reflexive systems should handle immediate responses while AI handles complex decisions (Lars Reger)
Summary
Agreement that trust, built through functional safety and cybersecurity, is the foundation for successful AI deployment and user acceptance
Topics
Building confidence and security in the use of ICTs | Artificial intelligence
AI systems should be modular and scalable across different applications
Speakers
– Lars Reger
Arguments
All smart robots share common ingredients: sense, think, connect, and act capabilities (Lars Reger)
Scalable semiconductor building blocks enable AI functionality across diverse form factors (Lars Reger)
Summary
Strong consensus on the need for modular, scalable AI architectures that can be adapted across different form factors and applications while maintaining common core functionalities
Topics
The enabling environment for digital development | Artificial intelligence | Information and communication technologies for development
Similar viewpoints
Both speakers emphasize the critical role of semiconductor hardware in enabling practical AI applications across various sectors
Speakers
– Lars Reger
– Moderator
Arguments
NXP’s role in AI advancement through secure, efficient, real-world AI hardware development (Moderator)
Scalable semiconductor building blocks enable AI functionality across diverse form factors (Lars Reger)
Topics
The enabling environment for digital development | Artificial intelligence
Consistent emphasis that advanced sensing capabilities are more important than AI intelligence for safety and functionality
Speakers
– Lars Reger
Arguments
Enhanced sensing capabilities are more critical than AI intelligence for autonomous vehicle safety (Lars Reger)
Robots need superhero-like sensing abilities including telepathic communication and X-ray vision (Lars Reger)
Current technology already provides ultra-wideband, car-to-car communication, and advanced radar systems (Lars Reger)
Topics
Building confidence and security in the use of ICTs | Artificial intelligence
Unexpected consensus
AI intelligence requirements are much smaller than commonly assumed
Speakers
– Lars Reger
Arguments
Most AI tasks require much smaller neural networks than human-level intelligence (Lars Reger)
80% of AI tasks will be performed on tiny, efficient, tailor-made models at the edge (Lars Reger)
Explanation
Unexpected consensus that most AI applications don’t need human-level intelligence, challenging the common assumption that more powerful AI is always better. This has significant implications for resource allocation and development priorities
Topics
Artificial intelligence | Environmental impacts
Sensing capabilities are more critical than AI intelligence for safety
Speakers
– Lars Reger
Arguments
Enhanced sensing capabilities are more critical than AI intelligence for autonomous vehicle safety (Lars Reger)
Current autonomous vehicle fatalities weren’t caused by AI brain structure bugs but by inadequate sensing
Explanation
Surprising consensus that the focus should be on improving sensors rather than AI algorithms for safety-critical applications, which challenges the common narrative that AI intelligence is the primary bottleneck
Topics
Building confidence and security in the use of ICTs | Artificial intelligence
Overall assessment
Summary
Strong consensus around edge AI being superior to centralized approaches, energy efficiency being critical for sustainability, trust being fundamental for adoption, and the need for modular, scalable architectures. Unexpected agreement that AI intelligence requirements are smaller than assumed and that sensing capabilities are more critical than AI intelligence for safety applications.
Consensus level
High level of consensus with significant implications for AI development priorities, suggesting a shift from centralized, high-power AI systems to distributed, efficient, sensor-rich edge solutions. This consensus could reshape industry investment and research directions toward more practical, sustainable AI implementations.
Differences
Different viewpoints
Unexpected differences
Overall assessment
Summary
No disagreements identified in the transcript
Disagreement level
This transcript contains only one substantive speaker (Lars Reger) presenting his vision for AI development, with the moderator providing only an introductory statement. There are no opposing viewpoints, counterarguments, or alternative perspectives presented. The absence of disagreement is significant as it represents a one-sided presentation of edge AI solutions without critical examination of potential challenges, limitations, or alternative approaches to AI development and deployment.
Partial agreements
Partial agreements
Similar viewpoints
Both speakers emphasize the critical role of semiconductor hardware in enabling practical AI applications across various sectors
Speakers
– Lars Reger
– Moderator
Arguments
NXP’s role in AI advancement through secure, efficient, real-world AI hardware development (Moderator)
Scalable semiconductor building blocks enable AI functionality across diverse form factors (Lars Reger)
Topics
The enabling environment for digital development | Artificial intelligence
Consistent emphasis that advanced sensing capabilities are more important than AI intelligence for safety and functionality
Speakers
– Lars Reger
Arguments
Enhanced sensing capabilities are more critical than AI intelligence for autonomous vehicle safety (Lars Reger)
Robots need superhero-like sensing abilities including telepathic communication and X-ray vision (Lars Reger)
Current technology already provides ultra-wideband, car-to-car communication, and advanced radar systems (Lars Reger)
Topics
Building confidence and security in the use of ICTs | Artificial intelligence
Takeaways
Key takeaways
AI’s future lies in edge computing rather than centralized data centers, with 80% of AI tasks being performed on small, efficient models at the device level
The next technological paradigm shift is moving from on-demand digital services to a world that ‘anticipates and automates’ user needs
All smart connected devices share four universal functions: sense, think, connect, and act – regardless of their form factor
Trust is the fundamental requirement for AI adoption, requiring both functional safety (like automotive braking systems) and cybersecurity protection
Biological architecture should inspire robot design, with layered systems where real-time reflexes handle immediate responses while AI manages complex decisions
Energy efficiency is critical for scaling AI – 50 billion smart devices would require three times Earth’s available energy without efficient edge computing
Advanced sensing capabilities (ultra-wideband, radar, car-to-car communication) are more crucial than AI intelligence for autonomous system safety
AI democratization requires scalable semiconductor building blocks that can be combined for different applications, from drone control to smart home systems
Most AI applications don’t require human-level intelligence – simple systems with 100,000 neurons can perform many transportation and automation tasks effectively
Resolutions and action items
Industry must focus on developing safe and secure architectures that are ultra-low power and energy efficient
Need to push the boundaries of physics in sensing capabilities to exceed human sensing abilities
Semiconductor companies should build scalable ‘Lego brick’ components that can be combined for different AI applications
Continue development of edge AI solutions that can operate large language models with minimal power consumption (7 watts demonstrated)
Unresolved issues
How to solve the massive energy requirements if 50 billion smart connected devices are deployed globally
Technical challenges in creating the wiring harnesses and battery systems for widespread AI device deployment
Specific implementation details for achieving superhero-like sensing capabilities across all device types
How to ensure interoperability and common language standards across diverse AI-enabled devices
Scaling challenges for moving from current hundreds of thousands of devices to projected 50 billion devices
Suggested compromises
Use AI models sized appropriately for tasks rather than pursuing maximum intelligence – most applications need far less than human-level AI
Focus on intermittent AI processing (7 watts only when needed, then sleep mode) rather than always-on systems
Prioritize enhanced sensing capabilities over complex AI processing for safety-critical applications like autonomous vehicles
Combine small edge AI processors with cloud connectivity rather than relying solely on either approach
Thought provoking comments
When we are talking about AI, at the moment there is a lot of talk about how do we pump AI in big data centers, how are we energizing these big data centers, but very honestly, there is a lot of questions. What is this AI for? What is this AI at all doing?
Speaker
Lars Reger
Reason
This opening challenge immediately reframes the entire AI discussion from technical implementation to fundamental purpose. Rather than accepting the prevailing focus on large-scale AI infrastructure, Reger questions the very utility and direction of current AI development, setting up a contrarian perspective that will drive his entire presentation.
Impact
This comment establishes the foundational tension for the entire discussion – shifting focus from ‘how to build bigger AI’ to ‘what AI should actually accomplish for humanity.’ It creates the intellectual framework for everything that follows.
We are entering a phase of the world that anticipates and automates… jumping forward maybe 20 years, how is the cocoon that I’m living in going to look like?
Speaker
Lars Reger
Reason
This concept of ‘anticipates and automates’ provides a compelling vision that goes beyond current on-demand technology. The ‘cocoon’ metaphor is particularly powerful as it suggests AI creating protective, personalized environments rather than just tools. This reframes AI from reactive services to proactive life enhancement.
Impact
This vision shifts the discussion from current AI limitations to future possibilities, establishing a concrete framework for understanding how AI should integrate into daily life. It moves the conversation from abstract technology to tangible human benefits.
Most of the cars that have created fatalities in the last 10 years… didn’t create these fatalities because they had a bug in the brain structure they created these issues because they were more short-sighted than I
Speaker
Lars Reger
Reason
This insight challenges the common assumption that AI failures are primarily software problems. By identifying sensory limitations rather than processing power as the core issue, Reger fundamentally reframes how we should approach AI safety and development priorities.
Impact
This comment pivots the entire technical discussion away from computational power toward sensory capabilities, leading directly into his ‘superhero senses’ concept and reshaping how the audience thinks about AI development priorities.
Try to copy from nature. That normally works… here on stage is a 90-kilo bag of water with a couple of bones, or in other words, a biological robot
Speaker
Lars Reger
Reason
This biomimetic approach provides a profound architectural insight by comparing human neural architecture (spine for reflexes, cerebellum for autonomic functions, brain for cognition) to how AI systems should be designed. The self-deprecating humor makes complex technical concepts accessible while delivering serious architectural principles.
Impact
This biological analogy provides a concrete framework for understanding distributed AI architecture, moving the discussion from abstract AI concepts to practical system design principles that the audience can visualize and understand.
80% of the AI tasks around us will be on very tiny, efficient, and very, very tailor-made models at the famous edge… when PM Modi says he wants to bring AI to everyone, this is the answer. The answer is not data centers.
Speaker
Lars Reger
Reason
This directly challenges the dominant narrative of centralized AI development and offers a democratization thesis. By connecting edge computing to political goals of AI accessibility, Reger makes a powerful argument for distributed rather than centralized AI development, with significant implications for global AI equity.
Impact
This comment provides the climactic argument that ties together all previous points, offering a concrete solution to AI democratization that challenges big tech’s centralized approach. It positions edge AI as not just technically superior but socially necessary.
These 50 billion smart connected devices need three times the energy that Mother Earth can provide. So that is the absolute must for these markets to come into play.
Speaker
Lars Reger
Reason
This stark energy constraint revelation adds urgent environmental and practical dimensions to the AI discussion. By quantifying the impossibility of scaling current AI approaches, Reger makes energy efficiency not just desirable but existentially necessary for AI’s future.
Impact
This energy constraint argument provides compelling justification for the entire edge AI thesis, transforming it from a technical preference to an environmental imperative that gives moral weight to his technological arguments.
Overall assessment
These key comments fundamentally restructured what could have been a standard technology presentation into a comprehensive challenge to prevailing AI development paradigms. Reger systematically dismantled assumptions about centralized AI, computational focus over sensory capabilities, and energy-intensive approaches. His progression from questioning AI’s purpose to providing biological architectural models to demonstrating energy constraints created a compelling narrative arc that positioned edge AI not as an alternative approach, but as the only viable path forward. The discussion evolved from technical specifications to philosophical questions about AI’s role in society, ultimately presenting edge computing as both technically superior and socially necessary for true AI democratization. His use of accessible metaphors and self-deprecating humor made complex technical concepts digestible while maintaining intellectual rigor, creating a presentation that challenged expert assumptions while remaining accessible to broader audiences.
Follow-up questions
How do we make the wiring harnesses for these smart connected devices?
Speaker
Lars Reger
Explanation
This is a technical challenge that needs to be solved for the implementation of billions of smart connected devices, and Reger mentions it as an area where India and Europeans are collaborating in research
How do we battery operate all of these smart connected devices?
Speaker
Lars Reger
Explanation
Power management is critical since 50 billion smart connected devices would need three times the energy that Mother Earth can provide, making ultra-low power solutions essential
How are we sensing in the right way?
Speaker
Lars Reger
Explanation
Improving sensing capabilities beyond human levels is crucial for creating robots with ‘superhero senses’ that can operate safely and effectively
How do we think – referring to AI processing architectures?
Speaker
Lars Reger
Explanation
Understanding optimal AI processing methods for edge devices is fundamental to creating efficient, trustworthy autonomous systems
How big does the brain have to be for these AI systems?
Speaker
Lars Reger
Explanation
Determining the optimal size of AI models (between 100,000 and 100 billion neurons) is crucial for balancing functionality with energy efficiency in edge devices
How do we push the boundaries and envelope of physics for better sensing?
Speaker
Lars Reger
Explanation
Advancing sensing technology beyond current human capabilities is necessary to create the ‘superhero senses’ required for truly autonomous and safe robotic systems
Disclaimer: This is not an official session record. DiploAI generates these resources from audiovisual recordings, and they are presented as-is, including potential errors. Due to logistical challenges, such as discrepancies in audio/video or transcripts, names may be misspelled. We strive for accuracy to the best of our ability.
Related event
