The AI Year 2018
With the end of the year coming up, a lot of articles start to appear providing predictions and forecasts for Artificial Intelligence for 2019. And I get why this is exciting: what will happen next, what are the technology trends in AI in 2019? What is the AI industry forecast? In our fast-paced world though, I think it sometimes makes sense to stop for a moment and take a look back. It helps to reflect on things, see what worked and what didn’t. Especially in the field of Artificial Intelligence, often bold predictions can be observed. When will my car get me home although I’m drunk? When are conscious robots going to enslave us all?
Help! The uprising has started! Tell the peop..gargllghagagglll
With the weekly briefing The AI Weekender that I send out every Friday, I collect a lot of news about Artificial Intelligence during the year. Let’s take a look what happened so far and what 3 main things made headlines. Let me know whether I missed anything way more important in the comments!
Autonomous driving
No doubt, autonomous driving has been huge in 2018. It might be that The AI Weekender is a little biased by including articles about self-driving cars a little more often than others. But the trend is real and this year, some remarkable things happened.
Following the classic hype-cycle though, autonomous driving has been accompanied by accelerating news over the last years, coupled with exaggerated expectations. 2018, we witnessed the technology hype enter the downward slope, where the public can finally see it‘s actual capabilities. And indeed, everybody has been very busy backpadeling on all those grand claims (*cough* Tesla *cough*). Volvo wanted to put 100 families in self-Driving SUVs by 2017, Tesla’s Full Self-Driving Autopilot has been removed from its order page and even Waymo admitted that it it will take a long time for wide-spread availability of autonomous cars.
Instead, we are currently seeing the (slow) rollout of fleets in shared mobility business models. That means a company is not selling a car equipped with a certain autonomous functionality to end-consumers (like an OEM), but rather providing shuttle or on-demand taxi services. This allows a way more controllable approach in comparison to handing this technology out to consumers. The company Voyage for example has struck a deal with Florida’s biggest retirement community to provide their autonomous services there.
At the forefront of technology. (Photo by Groomee/iStock / Getty Images)
It’s the same approach that Waymo follows. Barely one issue of the AI Weekender goes by without an article about the Alphabet company. And with good reason: Over the year it became clear that Waymo is about to get more public about their product. Though it remains to be seen over the 2019, how good their solution works and how autonomous it actually is (because it currently seems to be not-so-much), it is still very impressive. Waymo will now launch a service in Phoenix, Arizona called Waymo One, that leads chosen customers call for an autonomous Taxi from pre-defined pickup and drop off locations.
Why we can’t just simply plug the technology into a car yet, has become painfully clear, when in late March, an Uber autonomous prototype hit and killed a pedestrian crossing the street. The story unveiled a gross disrespect for safety considerations at Uber, halting the program for months. In retrospect, it makes an interview with Uber CEO Dara Khosrowshahi in January look even more sinister. Uber seems to have pushed the technology way outside its capabilities, disabling safety measures and building up an environment where it was impossible to voice concerns. But crashs from Tesla and GM demonstrate that this technology is definitely not where we all wish it to be.
This might be the reason that the automotive OEMs seem to be late to the party, but busy with announcements: Daimler with Bosch, Volkswagen with Apple and Aurora, GM with lots of money for Cruise, Volvo with NVIDIA, etc. What will come out of that are interesting questions for the next year and the one after, given the traditionally long development cycles in the automotive industry.
Also, one very important question was (somewhat) answered in a huge study conducted by the MIT Media Lab: Who should a self-driving car kill in an unavoidable accident? They asked millions of people whether to prioritize humans over pets, passengers over pedestrians, young over old, etc. The answer? Depends on where you are from! But it shows to what extent the general public and non-tech sciences are thinking about Autonomous Driving - it has truly arrived in mainstream discussion.
China
As might become clear from this article, I have moved to China this year. China has made headlines this year mainly due to the trade war with the U.S. But in the world of Artificial Intelligence, China also plays a huge role. Mainly driven by China’s ‘Made in China 2025’ strategy, AI research (although lacking behind Europe) is exploding as well as practical implementation, leading to the prediction that China will at 2030 become the world’s leader in AI technology.
The reason for this bold prediction is the combined effort of Chinese research institutes, universities, private companies and the government. It is estimated that China spent a total of $12 billion USD on artificial intelligence systems in 2017 - and the forecast says this might increase this to at least $70 billion USD. 14 Chinese AI companies are now valued at $1 billion USD or more, while Chinese startups have raked in VC deals in the range of $27.7 billion USD in 2017 alone. All of this is leading to results: Alibaba, the Chinese tech giant, for example will introduce it’s own neural network chip (as Google has already done in 2016, while having unveiled the 3rd version of there TPU this year). This is remarkable, since technology like this has been a stronghold of the US industry. The White House meanwhile, stays silent, although the west could learn a lot from the approach.
Who would’ve thought that a country might come up with a plan on how to lift itself above simple production lines?
China is also installing impressive amounts of facial recognition technology to catch criminals - it took 7 minutes to capture a BBC reporter in a test run. The system is now live and able to identify a single fugitive within a crowd of 50.000 people or publicly displaying people jaywalking.
Another huge opportunity for AI technology is the Chinese automotive market. The self-driving car industry in China is huge and the big internet players - Baidu, Tencent, Alibaba, - are all heavily invested in it. This, together with the government funded support, caused an explosion of startups and smaller companies in the field.
The AI hype is fueled by data and China has a lot more than other countries. Mobile payment data is maybe 50 times bigger than for the US, Kai-Fu Lee says. Additionally, the availability of cheap labor enables the generation of massive amounts of labeled training data with the Chinese adaptation rates of digital services like WeChat providing the foundational data. This lead to the surprising revelation, that Alibaba already employs a digital voice assistant way more capable than what Google unveiled earlier this year - and in daily operation.
The creation of content with AI and DeepFakes
In 2014, Ian Goodfellow and Others invented the Generative Adversarial Network. This technology can generate life-like pictures by pitching two neural networks against each other. Together with some more tech from the deep learning space, lots of interesting and - arguably - frightening things are happening.
At the end of 2017 and beginning of 2018, the internet - being as helpful as ever - used this technology to generate porn featuring celebrities or basically anybody you have enough photos of. The problem became so serious, that Reddit, tumblr and all the big porn sites (not going to put links here) had to explicitly ban DeepFake porn.
The story gets worse though: with the technology, you can even let politicians say things they never did. This led to a debate over a whole new type of Fake News: how can we detect very realistic looking pictures of events or even videos of people doing certain things that are fake? What does it mean for our society in general, for election processes?
So far though, this debate seems to have cooled off and what stayed are the useful applications of the underlying technology. I have used it myself to generate and sell T-shirt’s online and in fact, the creation of digital media seems to be of quite some interest. NVIDIA has managed to generate absolutely realistic images of people that don’t exist and then used that to make a tool you can use to repair seriously damaged photos. A portrait generated with a GAN recently even sold for 432.500 USD at a Christie’s auction. (TODO Link).
Still images are one thing, moving ones another: while we are (supposedly) far away from actually fully-automated production of movies with fake characters, GANs were used to generate Flintstone episodes. They seem to be also quite useful in creating special effects, an otherwise very expensive technology. Fans even used it to digitally insert a young Harrison Ford into Solo: A Star Wars Story. And in China, a virtual news anchor is generated to read the news online and on television 24 hours a day, replacing the need for expensive human anchors.
In the end, it apparently doesn’t stop with images and movies: with the help of GANs so-called “Master fingerprints” have been generated, enabling researchers to unlock many smartphones or biometric security systems. Also, a whole Heavy Metal album has been produced. I doubt this will be the last time we see the misuse of this technology.
Roundup
I know that there have been so many more things to talk about in 2018. And that is to be expected, with new AI technologies entering our devices, websites, cars and other daily products more and more. Besides all the noise around though, most real progress is being made behind the scenes, in the algorithms dominating our daily life’s already. The three things I’ve picked out are merely a collection of very high-level fields in which lots of different trends and forces are working in different speeds. It will be interesting to see, where we are at the end of next year! Did I miss an important area? Let me know in the comments!
