Rise of AI Conference 2018 · Berlin

K.I.T.T. – I need you, buddy

Holger Weiss · CEO, German Autolabs

At Rise of AI 2018 in Berlin, Holger Weiss, CEO of German Autolabs, argues that voice assistants will have their biggest impact not at home but in the car, where drivers cannot look at a screen. He introduces Chris, a retrofit conversational co-driver that connects to a phone via voice and gesture control, built because most cars on the road are old: the average German car is ten years old, and Germany has 47 million cars while the US has 220 million, so a new-car-only assistant would take years to reach the market. Weiss cites growing distraction-related deaths, including the US 'Snapchat effect', and ADAC data on smartphone-related traffic deaths to argue for hands-free interaction. He distinguishes a co-driver from a home assistant like Alexa or Siri: it must understand context such as location, speed, connectivity gaps and cognitive load, and prioritize accordingly, for example not reading an email while a driver approaches a difficult highway exit. German Autolabs worked with the German Ministry of Transport and a Munich traffic psychology professor on modeling driver cognitive load. Chris launched as a pre-order device priced at 299 euros or dollars, in German and English.

Key takeaways

  • Holger Weiss argues in-car voice assistants will have more impact than home assistants because drivers cannot look at a screen.
  • German Autolabs built Chris as a retrofit device because most cars on the road, averaging ten years old, lack built-in AI systems.
  • Weiss cites ADAC data that smartphone use causes more traffic deaths than alcohol and drugs combined.
  • A car assistant must work offline, since vehicles regularly lose connectivity in tunnels, garages and rural areas.
  • German Autolabs worked with the Ministry of Transport and a Munich traffic psychologist to model driver cognitive overload and underload.

Detailed notes

A retrofit assistant for an aging car fleet

German Autolabs built Chris, a device that fits into any car and connects to the driver's phone, letting them answer emails and text messages by voice and browse contacts with a gesture sensor. Weiss says the company chose a retrofit approach because a system designed only for new cars would take four to five years to reach the road, while the existing fleet is old: Germany has 47 million cars and the US 220 million, with an average car age of about ten years, so most drivers cannot rely on built-in systems from a Tesla or a new BMW.

Distraction is a safety problem, not a convenience one

Weiss cites the German motoring club ADAC, which represents 19 million members, saying more people die in traffic from smartphone use than from alcohol and drugs combined. He also points to data from the US National Highway Traffic Safety Administration showing the first rise in highway deaths in a decade, which the agency calls the 'Snapchat effect'. He argues drivers will keep using phones and infotainment systems for the next 10 to 15 years because widespread autonomous driving is not coming as soon as some claim, so hands-free voice interaction is a safety necessity, not a luxury.

Why a car assistant differs from a home assistant

According to Weiss, natural language processing has crossed a threshold, with recognition accuracy rising from roughly 50-60% to about 90%, letting people speak to a system rather than memorize commands. But he argues a co-driver needs more than recognition: it must work offline, since cars regularly lose connectivity in tunnels, parking garages, rural areas or on 2.5G coverage, unlike a home or office. It also needs contextual prioritization: at a highway exit or a stressful, congested moment, the assistant should not start reading an email, because a driver's cognitive load and the situation's urgency change moment to moment in a way a home assistant never has to track.

Researching cognitive load with the German Ministry of Transport

German Autolabs received funding through the Ministry of Transport's mFund program and worked with a Munich traffic psychology professor to study the driver's 'perfect cognitive load'. Weiss describes an overload scenario, four lanes of rush-hour traffic with a full inbox and a looming meeting, where an assistant reading emails would add danger, and an underload scenario, a long empty highway stretch, where a companion that talks, tells stories or suggests a stop can keep a driver alert.

Launch and pricing

Chris was available as a consumer product for 299 euros or US dollars, with a pre-sale price of 499 during a crowdfunding-style 'Hello Chris' campaign, launching in German and English with French and Spanish planned.

Quotes

There are more people dying in traffic by smartphones than by alcohol and drugs together.

Holger Weiss

It's better if these people have their hands on the wheel and the eyes on the road.

Holger Weiss

Questions answered in this talk

Why does an in-car voice assistant need to work differently from a home assistant like Alexa?

According to Holger Weiss at Rise of AI 2018, a car assistant needs to understand context such as location, speed and connectivity gaps, and must know when not to interrupt, for example not reading an email while a driver approaches a highway exit.

How much did German Autolabs' Chris device cost?

Weiss states the Chris retrofit assistant launched at 299 euros or US dollars, with an early pre-sale price of 499 during the company's 'Hello Chris' campaign.

Why did German Autolabs build a retrofit assistant instead of a built-in car system?

Weiss explains that because Germany's average car is about ten years old and a system built only for new cars would take four to five years to reach the road, a retrofit device could reach far more of the 47 million cars in Germany and 220 million in the US.

Mentioned

natural language processing, retrofit assistant, cognitive load, voice interface, autonomous driving

Keep exploring

More from Rise of AI Conference 2018

More on Industry & Robotics

Transcript

Auto-generated transcript, not edited. · 3,118 words

Show the full transcript

Wow that's a pretty big conference I I'm quite impressed and really cut and compliments to to Fabian where he developed this from from last year so it seems that AI is really honor on rice as the name of the conferences and and will affect us in all our lives but I mean with with all respect to my fellow speakers and and to the research a lot of that is still very much in the future and some of the stuff appears sometimes very boring so therefore it's good if you actually can touch with AI something that makes a lot of sense and that is really realistic I mean this this is my idol from from the 80s right kit and the Nightrider having a car that actually turns into your body into someone that supports your in in every event and oh by the way this is my idol not this guy here I know he's extremely popular still in in in Berlin so and any American guy here in the room don't get confused the

Hoff is still a number here but that is that's actually something that we are working at that we try to build a system of they're working on a system that actually turns a car into a companion and since we have a lot of cars out there we think let's start with that I don't know how for me you are without emotive industries but if you develop a system for a car it barely will show up before four or five years from now on the road while in Germany we have 47 million cars in in the u.s. there are 220 million cars out on the road and they are old they're freaking old not everyone of us is able to afford a Tesla or the latest 7 Series BMW where you have a lot of technologies and the average kind Romany is 10 years old so we decided we go for something that fits in every car and that is Chris our hit his name Chris and Chris is a retrofit assistant that fits in every

kind you've connected with your phone and actually reach you email and you can answer your text messages just with voice it has a gesture sensor so you can easily swipe through a list of contacts you can listen to your music and and elements like that so that is something that we will bring to the market and now the question that I would like to focus a little bit today on is what makes a co-driver co-driver what makes Chris different or other systems different in a car than just a home assistant so I don't know if any of my previous speakers mentioned that but there's a reason why we see at the moment this increase of voice interfaces all over the place of course one of the reason is that technology sometimes are at the tipping point right we remember when the iPhone came in 2007 touchscreens were around for decades and and and and and mobile internet was even not part of the the first iPhone etc and so but it was at the tipping

point and then it exploded in what we know today from the smartphone so we see this at the moment also with NLP and voice interfaces definitely driven by by by platforms like Syria and Alexa and Cortana etc but also by a whole generation and growing growing into that into that use but there's also technical reason as so many things in AI with the computing power that we have today with with cloud infrastructures we are on a level that after 40 30 40 years in in voice recognition we have a breakthrough on a almost human level so before voice interfaces were able to understand 50 to 60 percent which leads actually to the fact that those navigation systems in cars that you might gonna rent or own or whatever is a gun of a frustrating experience friend of mine driving rather new German high-end car says he only uses the voice interface in his navigation system to entertain his son on long long-haul rights because it's so funny that the voice doesn't understand it so and

then if we look on that and say there's now ninety percent accuracy the question is where we will see this breakthrough and where we will see people really to adopt that those assistants and most of people say at home because this is where we see at the moment the strongest the strongest impact with all those connected speakers with Alexa with all of what we have here but we believe strongly that the environment where the voice assistant and the co-driver as we call it will make most of impact and and the fastest iteration and improvement will be the car and the reason is very simple because in the car I have to focus on the street and still for a pretty long time there are some people who want to make you believe that we all drive autonomous in 2019 I doubt about that right so for the next 10 15 years 15 20 years people will drive with a hundred forty kilometers one and a half tons and it's better if these people have their

hands on the wheel and the ice on the road and not write and reading text messages or watching videos so that is also what we working on we working on a conversational platform as we call it for out of motive right and technically a lot of what we're doing and you see also in assistance wherever they are I mean they're home assistance I was talking about but you also see industrial systems what is very special with us is that if we have a co-driver the co-driver should work in each and every situation and what makes a car very special compared to an office and and a home in most of the countries is that the car is barely connected I mean if you're driving in the center of an environment you might gonna have a pretty good coverage but it happened to me that around the corner he had John Dom mark I couldn't open a car-sharing car from Drive now because I'm in a 2.5 g cell and then there is no internet and

with no internet I cannot open the car or if the car is in the parking or in a tunnel or in a rural area and we all know those white spots I mean that is something I cannot recommend to go to Sicily with Google Maps it might be a very stressful vacation because I mean in 75% of time you have no navigation because you you just offline so that is something that we're focusing on over focusing on this online and offline scenario and in in in our automotive environment so if we now looking to why would we need an assistant in a car and and that is the question right we sometimes have this feeling why do we all have this and and I mean talking about cars and and and and about mobility there was this phenomenal success of navigation systems 10-12 years ago right TomTom Garmin you know that and that was a use case everyone got immediately because having instead of a paper map you have a system that tells you how

to get from A to B and you're right you're your destination and then traffic information came on top that was a no-brainer but why you need an assistant in a car right or why you need at all assistant and that is something when we look on the car there is a very pressuring reason for now prospectively there are a lot of reasons to have there's assistance but today the ADAC the German automotive club 19 million members they say there are more people dying in traffic by smartphones than by alcohol and drugs together so this is a curve from from the North American Highway Transportation Safety Administration Nitsa saying that in the first the first time in in 10 years in a decade we have an increase of death casualties on you as highways and they call it the snapchat effect it's not my expression right so meaning that why we have all these old cars or relatively old cars and why we are all so addicted to things there's a new what's upcoming there is

a new Facebook thing whatever we have this high level of distraction which a set at home is no problem because you can handle five six seven different tracks at the same time but why are you driving with a hundred 40 km/h there is an impact and there's a massive impact which can lead to your own death or to the death of other people bless you so now it's the question what makes an assistant an assistant right when we talk about digital assistants and that is something that I would say 99% of people here in the room would say it's that I can talk to it or whatever right that is something that we started to as that interact with those first voice systems in navigation systems but many of you experienced this with Siri for example there was a lot of fun and I could talk it replied it's not a little bit human it felt a little bit AI it's not very much AI believe me it's a lot of ml but that is

something what what we see as one of the disciplines that we are all working out natural language processing which is something that is also part of this breakthrough of the systems that I don't have to learn anymore a special language but one of the reasons why people don't like to use the current voice systems because I have to learn those commands navigation start city/country now and then you say the city name and it doesn't understand you because whatever you have a beautiful accent or the system and is is or there's noise around or I don't know so and that's something that with their with those system that you know from from the modern and voice interfaces turns into what we call the natural language processing so you just talk to the system as you would talk to a person which makes it feel very very natural right and the most impressive presentation gave Google again two weeks ago the doublets when when a voice assistant actually called the restaurant to book a table and I

I mean I think for 48 hours I had no other those things in my Facebook and LinkedIn feet because people were just so Wow right it's a complete different discussion yesterday we had a very nice evening with the BCG digital Ventures about AI and ethics if this is something that we really want to have but that's something that's elementary to have this interaction this voice interaction with a new system but the real question in our case is what makes a co-driver co-driver right because understanding you that is one part I'm saying Alexa tell me a joke or whatever Cortana switch off the lights or Siri I need new orange juice or I don't know that's one element to understand you and having all this what we call the dialogue management the intent management out of the intent understanding we do a decision that's something what the chat bot intelligence that my my my my fellow previous speaker was explaining about this day but with a co-driver it's something different right think about you driving a

car and someone is sitting next to you knowing you very well right your wife your husband a good friend someone who really knows you and while you're driving you might gonna be stressed or whatsoever so it needs a fundamental amount of information about yourself and like your driving patterns cetera to do the very right thing at the very right time why does it matter because while being mobile we have two more dimensions we have the relocation that counts that doesn't count while I'm at home because it's just home and we have speed acceleration right so if you're approaching a highway exit and there is one of these classic double exits right first exit next exit a couple of 200 meters behind it's fundamentally super important for you to listen to the navigation command in the right moment and that means that in this very right moment your co-driver should not start to read an email for you which actually happens with common systems because there is no prioritization so or thinking about that that is

the a what run rate here in in Berlin it's a picture from 8:30 7:30 in the morning the very same picture at 11:00 no cars and you all know these situations in your in your environments and the cities the 101 San Francisco down to the valley all over the world right so we have contextual influence that a co-driver actually needs to know which at home again is not too important or weather conditions historical conditions that you know okay this is a dangerous curve that I only know because I drove it before and it's something where I can warn you proactively like the Israeli driver that that that we saw before oh that's something behavior right as I sat at the moment you're approaching a highway exit and you're focusing on what is the right thing to do now do I have to take the turn or the next turn something like that if you're talking in a situation like that to an assistant that would actually work or react like a general assistant at home

why you would stop talking the assistant would decide okay dialogue over I switch off this activity now while in a car because it's something where you cannot just say okay I finish one activity and then I stop the other just by the fact how how I just described it if you just turn around and looking you will not talk in this situation people will not do that right women maybe man definitely not I mean they turn off the radio to ask for the destination I mean that's that's something what natural human beings do right it's it's a cognitive law and that is something that someone who knows you and who learned about you definitely would have to consider in the process of decision-making so as a we said a co-driver assistant that is a co-driver needs about the contacts it needs to know about the context it is in and in this situation it's and and about your behavior and the behavior is not a it's it's not a single situation it's something that you

have to learn so you have to spend time with the person to learn out of that and in our case you also have to adopt this behavior to to other people right all like that we working here on a project with the German Ministry of Transport we we got this M font wallet it's a it's a it's a subsidiary that subsidies that actually help you to develop certain R&D project that's the the question in which situation a driver has the perfect cognitive load so what does that mean so this both contrast Myers working on this for 30 years it's a it's a professor for traffic cycle psychology from from Munich so we all can imagine this situation that I had before on the a100 that it's a stressful thing you have four lanes it's seven o'clock in the morning you're 40 months in the Inbox the first meeting is at 8:00 you're late already you have to take the Lane whatever right something like that completely stressful in a situation like that you definitely don't

want to listen to an email that an assistant actually could read to you that would be the overload but in the same way there is an under load if you're driving three hours on a on a long-haul trip and nothing is happening and you're just driving driving driving you under loaded cognitively write something we're in a real life someone will start to play a game with you telling you a story because you're about to fall asleep or actually someone would say hey come on we have two more hours it's midnight why don't we just do a stop and there's a hotel around the corner and we continue the right at four o'clock in the morning so that's something that also a co-driver should know so and that is what we are working out in addition to to this just technical capabilities to train an assistant to speak that's something how you train a children a kid like a baby every baby is coming with the capability to talk but it takes quite a while before

it makes sense what it says and then it takes even much more before you want to send it to school etc that's something that we do and the other element is said is that what I just described is this cognitive understanding the cognitive prioritization of of a code so in a nutshell what we're doing is building this this this infrastructure this platform that initially will will now bring as a consumer product you can buy it already and and which which fits in every car prospectively it's about exactly these intelligence I always say but they're the soul and the brain of Chris the soul and the brain of a co-driver and that's it thank you very much and I have two and a half more minutes for questions no questions the the consumer product 299 euros or a u.s.

dollars in there at the moment you can audit 499 at our website because we have a pre-sale free third campaign hello Chris not a I but it will be $2.99 and would come in German and English and that's the point I mean you have to teach a kid to speak and at the moment we teach it in German and English but it will also speak French and Spanish and all these fancy great languages get no more questions and then thank you very much have a great day and thanks for listening

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.