Your browser is not supported. Please update it.

10 January 2019

AI at CES 2019 – approach with extreme caution

Injecting a piece of new programmable software into a product alongside some generic algorithms and calling it artificially intelligent is not a new phenomenon, yet CES more than any other trade show is the place where you can bet your hat on quasi-AI being most rampant. Nevertheless, we have hand-picked a few TV-related AI announcements from the show which we feel are worth investigating – some of which are more obviously genuine innovations than others.

First off, the US arm of Chinese electronics heavyweight, Hisense USA, used CES to launch a trio of new TV models all powered by so-called AI chips to improve picture quality based on the content being viewed. Hisense says its AI algorithms can improve motion rate when watching sports, for example.

“It will make it smoother, versus a movie experience where the TV will detect the movie at 24 frames-per-second and then will make it a more surround-sound feeling. The colors are slightly more cinematic,” said David Gold, VP of Consumer Electronics at Hisense USA.

The new Quantum Dot ULED Hisense U9F TV, set to launch in June 2019 costing $3,500, incorporates its own Hi-View chipset and AI-based picture quality algorithms to automatically recognize how to improve color, contrast, motion and brightness to improve the viewing experience, which is about as far as the explanation goes. We are naturally skeptical about the AI element here and are almost certain most people in the industry have witnessed demonstrations of similar automatic picture-adjusting techniques in the past. After all, automation does not equal AI.

Another Asia Pacific manufacturer was particularly busy at the show. LG added Amazon Alexa to its ThinQ AI system this week to accompany Google Assistant on select TV sets, and will soon release a remote control with separate buttons for the two competing digital assistants. Covering all bases regarding voice assistants emerged as a trend last year and this year the hardware makers will push hard to become all-encompassing. That said, LG will soon follow Samsung’s lead (see separate story in this issue) by adding iTunes TV and Movies content as well as AirPlay support.

Soon, it seems, smart TV sets will offer significantly more diverse viewing experiences than they were originally intended, for which to stream a variety of entertainment and with the rise of Android TV operator tier, the wealth of apps is staggering. As smartphone sales stagnate, the walled gardens of iOS and Android experiences could soon hinder TV viewing experiences and instead drive viewers back to the big screen as prices fall. Although, mobile devices could eventually take a more open approach, as we have seen with some niche Android launches offering both Alexa and Google Assistant.

Not to be left out, after its Apple integration announcement Samsung said its line of Tizen smart TVs will offer Alexa and Google Assistant as well as Bixby.

From TVs to social media now, as Intel and Facebook used CES to ink an AI-based silicon collaboration. Again, automation is the premise of the deal with trivial features like automatically tagging friends in Facebook photos, something the company has been attempting to do for years to disastrous effect. The Nervana Neural Network Processor for Interface (NNP-I) is the first real fruit to arrive from Intel’s acquisition of Nervana for $350 million back in 2016.

But this wasn’t all Chipzilla got up to in Vegas, also teaming up with Chinese giant Alibaba in a project for the Tokyo 2020 Olympic Games. The aim is to apply deep learning algorithms to 3D tracking of athletes, using multiple cameras to build a “3D mesh” model. By combining techniques such as advanced pose modelling with computer vision, running on Alibaba Cloud infrastructure, the announcement says broadcasters can use biomechanical data for analyzing highlights during instant replays, plus it could also be rather handy for coaches.

In the gaming sector, meanwhile, Nvidia – under mounting pressure from Intel – unveiled its latest RTX GPU architecture with AI capabilities. A deep learning technique called Deep Learning Super Sampling can render input images into high-resolution outputs, with the model trained with rendered frames and intermediate buffers, it says.

Finally, on Friday Xiaomi is set to launch a set top at CES apparently powered by some sort of AI, according to TechGenYZ. Disappointingly for a Chinese company, judging by the pre-launch details the device sounds like a laughable attempt at jumping on the bandwagon with zero explanation of what exactly sets it apart from your bog standard set top apart from the product name. The Xiaomi Box 4 SE set version will be priced at just 189 yuan ($27.85) and comes with voice control functionality, running a Cortex-A7 quad-core processor with Mali-400 GPU.

As expected, every man and his dog were showcasing an AI product in some guise or another this week, but while some were genuine jaw-droppers, others fell short and unfortunately clogged up email inboxes and social feeds the world over with nonsense. It wasn’t easy, but we hope our manual filtering system did a decent job of covering the most significant video-related announcements in AI from Las Vegas so far. Next up, Mobile World Congress in Barcelona, where last year we witnessed an AI frenzy.