By Harold Selden
Reflections on CES 2017
In many ways the nVidia keynote was a microcosm of all this year’s CES themes that impact security. Those themes include (semi)autonomous cars, AI/machine learning, smart home, and IoT. Virtual reality (VR) and augmented reality (AR) were also a big collective theme, but in game use versus alternate display technology; which has less security impact.
Gary Shapiro’s introduction of the nVidia chairman, Jen-Hsun Wong, was prefaced by a few CES history pointers. CES started as a radio-TV show, then became a PC video refuge, partly as a result of the end of Comdex. Now, the focus is on grand challenges. Not so surprisingly then, Wong talked mostly about self-driving cars, and the machine learning necessary for this.
Next year, a Tesla car will be fully automated, and nVidia’s announced partnership with Audi will produce a self driving car, Level 4, in three years. In the meantime, nVidia is partnering with Bosch to distribute the SoC described below.
nVidia has been working on deep neural networks, a step past typical machine learning. They hope to use GPUs to get software to write software, but in the meantime are using hierarchical representations to analyze patterns. The most complex game, Go, has been programmed to a point (AlphaGo) where it beat a ninth Dan Go master. Other mentions of machine accomplishments:
- Play Doom, a game of mazes and resource management
- Paint in the style of Monet or van Gogh
- Voice recognition (95% correct vs only a few years ago, where it was unusable)
- Learn a painting style
- Learn to walk, and demonstrate motor skills
- Write captions
nVidia’s interest is the use of their GPUs to accomplish the above. nVidia announced an ITX board sized supercomputer, a SoC (system-on-chip) named Xavier, optimized to do the AI computing for self-driving cars (Q4, this year). Xavier is comprised of 8 ARM64 cores, 7 B transistors, dual 8K HDR video processors, and a 512 Core Volta GPU, all built on a 16 nm+ process. Wong listed the TDP at 30 watts, and performance of 30 TOPs (DL, or Deep Learning TFLOPs)! This ground breaking capability has security implications which were mostly glossed over. Aside from the obvious questions about safety, and hacking of the SOC, what kind of personal information is left on the computer, to be downloaded by the service tech? Currently, cars connected via BT to a phone store too much information that is easily accessed.
Wong also talked about AI co-pilot cars, that display information, warnings, and even monitor driver health, and attention (gaze tracking). These assisted or semi-autonomous cars will also help the two populations most at risk in cars: teenagers and senior citizens, and are a stepping stone to fully autonomous cars.
Alexa was the 800 lb gorilla at CES: there were many entries to work in that ecosystem, and many clones that appeared to be incompatible. To compete with Alexa, nVidia also announced a new home control microphone for nVidia’s Shield, which combines Samsung’s SmartThings and Google Assistant. Shield is a streaming device. Spot is a natural language AI microphone to control anything Shield does, and with echo cancellation and beam forming to triangulate location when used with multiple Spots, can be placed throughout the home. According to Jen-Hsun Wong, Shield + Google Assistant +Spot + smart hub = JARVIS. The new Shield, supporting 4K and HDR, is expected to retail for $199., and be available in the second quarter, and Spot at ~ $50., will be announced later.
Consumer Technology Association, the Rebranded Consumer Electronics Association
A press briefing by senior CTA (Consumer Technology Association, the people who run CES) analysts pointed out the bifurcated nature of this year’s show. Half the show was composed of a few, slow growth categories: computers, TVs, and smartphones. The other half of the show categories, all high growth, was composed of health and fitness, wearables, connected systems (fastest growing)– IoT, drones and VR, and smart home. Based on current market penetration of 2.4 smartphones per household, 4G’s maturity, and Ericsson’s demonstration of 5G at over 13 Gb/user, CTA’s chief economist, Shawn DuBravac predicted that VR would be appearing soon over 5G. This means fluid motion with no jagged edges, over streaming media, essentially, remote headsets.
The CTA panel predicted that the next big operating systems will be digital assistance devices: Amazon Alexa and Google Home. 4.5 M voice digital assistants were sold this past year, and this number will increase at least 52%. The fastest growing category is VR, at 79%, and the most revenue is generated by smartphones (55.6 B $, up 2%), but smart homes are expected to generate 3.5B$ in sales, at a growth rate of 63%. Are these devices listening even when we turn them off? Do we know? The next series of observations are based on the panel’s contributions.
Because of the “friction” in integrating devices, music and lights are usually included in the most commonly purchased initial smart home packages. A hub to integrate other peripherals is key, but how would that inter-manufacturer communication be secured? The next wave of smart home devices will include smarter devices that make integration seamless through more sophisticated software, as evidenced by Symantec’s new home router.
Home Automation Underpinnings
Voice recognition has improved considerably. Forty-eight months ago, the error rate was 25%. In the last 30 months, that error rate has improved substantially, enough to make home automation devices realistic. At the first CES in 1962, we thought the future was “Jetson’s magic.” Devices would spit out pancakes, do laundry, fold clothes, and they all looked like a giant computer. In the 80’s and 90’s, we saw structural automation. Building temperature would be static mostly, but could vary on certain days. Today we see devices reactive to the environment, that are nuanced, that improve quality of life in subtle ways.
1981 – mass market the original CD
1998 – the first digital assistant
2001 – iPod debuts
2003 – the first smartphone
The potential for abuse of home automation is staggering. From running up someone’s utility bill, to creating a soto voce notification to a thief when someone goes on vacation, the possible attacks seem limited only by one’s imagination.
New vehicle technology is forecast to be up 12%, with 17B $ in revenue. Tier 1 suppliers have made major upgrades. Devices have digitized and sensorized the environment to customize and adapt it to the user. Mercedes even customized the scent to the user. Fifty percent of the car value is the installed tech. More importantly, most safety features that are practical are included in modern cars; self-driving features are the last way to reduce traffic fatalities.
Driving functions must improve. Multiple AIs now overlap, and integration is a challenge. AIs now monitor a driver’s eyes, position of other cars, and driver posture. This data is combined holistically to obtain the full picture of each instant.
Just like server utilization and the push towards virtualization, the average car is used 4%. By sharing hardware, automobile utilization could improve to 95 -96%.
Hacking smart cars has been covered by every press outlet. Protecting data gathered by a car’s sensors will become even more important as the amount of data increases logarithmically. Will that data maintain it integrity as it is being collected, transferred, used, and stored? Tier 1 providers are at least paying lip service to this problem. Not surprisingly for a CES address, Jen-Hsun Wong made no mention of information security, but did mention how AI services could improve road safety. What if the AI lip read something about a crime to take place?
Health and Fitness
Sports tech was a big driver for the improvements we see; devices are very good at counting steps. One panelist dubbed this “the sensorization of everything.”A basketball can now measure the intensity of a dunk, or calculate the angle of the shot and velocity. One problem is the lack of uniformity between different brands of fitness devices. Another panelist is on a committee to write the standards for sleep and step counting.
The 2017 market for fitness wearables is predicted to be 47M. Though Underarmour rolled out some very pricey PJs designed to promote sleep, improving sleep will really come about by putting diet, exercise, and sleep data together to provide a better recommendation for improving sleep and fitness. Towards these goals, Fitbit announced new software and apps that provide more personal recommendations based on analyzing combined data from different Fitbit functions. One hundred years ago, no temperature or weight data was gathered in a doctor’s office. Fitness devices gather much more data than what is obtained in a typical visit. This is the democratization of healthcare.
Unfortunately, all this tracking makes you more vulnerable to inadvertent disclosure of personal information. Manufacturers are only slowly realizing that they have to protect this information. Fitbit, for example, engaged a security firm to vet the security around their cloud storage for the new software features. One can only hope that other manufacturers will do the same, and that older insecure apps and devices will fade away.