The state of AI
What I learned at Google Next ´18
AI is predicted to be the third era of computing and it is creeping into more and more of our daily routines. Now it's also becoming cloud-based which means that even the most sophisticated techniques become available to everybody. So it's about time to get started.
Can an AI algorithm draw artistic portraits? Today they can.
A few weeks ago, I attended Google Next in London — a conference with 16,000 participants, covering most of the new technology areas that are about to impact our world.
Here I had a very inspiring conversation with the owner of Dixit Algorizmi Gallery from Berlin about the future of art, while two AI-controlled robotic arms each drew their own different portrait of me. And that made me think, how far we had come in the development of machine-based intelligence.
Artificial Intelligence (AI) is a term that covers a range of different computational techniques ranging from machine learning, deep learning, neural networks to reinforcement learning, all simulating intelligent behaviour in computers.
The field of AI research dates back to the 50´s. Even when I studied to become an engineer some 30 years ago, I was taught how to make natural language processing algorithms and to code neural networks.
But a lot has happened since then. It seems like we are at the tipping point for AI — if we haven´t passed it already.
When it starts to work, we stop calling it AI
A lot of daily routines are already affected by the use of some form of AI.
Every time you turn on your new iPhone, FaceID using AI scans your face and recognizes you in order to unlock your phone.
Every plane leaving the ground each day will be partly flown by an autopilot system.
Gone are the days where we navigated our way with a map, or even a TomTom navigation system. Now it´s all on our smartphones, intelligently rerouting our way to get us to the destination in the shortest time.
And try to think back to when your mailbox was cluttered with spam? The junk mails are still being sent, I promise you. But smart algorithms from either Google or Microsoft manage to delete 99% of these, so it´s no longer a big problem for us.
Fancy a new series on Netflix? Well, thanks to pattern recognition and collaborative filtering techniques, Netflix´s recommendation algorithm not only knows exactly which series you are most likely to prefer, they also know which unique images presenting the show will make you stay the longest.
And if you happen to use Google Photos, you know that you can now search your 20,000 or more pictures for items like sea turtles or green cars or even names of friends — in your own native language. And the app will automatically show you the desired result, without you having to tag each and every photo — it´s all being handled by AI-based image recognition.
So slowly, but steadily, AI is being engraved into our daily routines, making our life smarter, simpler and more effective without us noticing it. Because when it works, we just call it navigation, searching or recommendation, without thinking about the technology underneath.
Big data needed
All the examples above are heavily dependent on access to huge data sets. Because that´s how you train your algorithm. And this tends to favour companies with the most customers.
So, no wonder that Google — with 7 application each banking more than 1 billion active monthly users (Search, Gmail, Chrome, Android, Maps, YouTube, Google Play Store) and very soon more to come (Google Drive and Google Assistant) are regarded as the leading AI company in the world.
Huge data sets are the raw material needed to get Machine Learning to work. And Google has a lot of it.
Take their Autonomous Vehicle Project Waymo as an example. Since 2009, they have self-driven more than 16 million km on physical roads. But that is dwarfed by the 11 billion km driven in simulation. All this data combined has led them to a point where they are now allowed to launch commercial driverless car services in Arizona and California. This will happen by the end of 2018.
How can small companies compete with that? Luckily, Google has an answer to this. Google already owns Kaggle— the world’s largest community of data scientists and machine learners and a database of more than 11.000 public available datasets. Now they have launched an Open data marketplace called: Google Cloud Public Datasets Program. The plan is to make big data sets available for everybody, together with the necessary sets of tools in order to start working with AI in a cloud-based decentralized way.
This is one of the reasons why we are about to take a giant leap forward. Commercial AI-based on huge data sets are way too complicated and expensive for most companies to run on-premise.
Therefore, it is crucial that these services and opportunities become available as cloud-based — pay as you go — solutions. That is exactly what Google is launching with their Cloud AI Platform. The vision behind is to democratize the use of AI and to make it accessible and easy to get started for even small companies or individuals. You will find a list of pre-trained models ready to use and all the tools and APIs ready to get started.
The other groundbreaking reason why we are about to witness a giant leap forward is the use of specifically designed hardware for AI. Google has now launched the third generation of computer chips designed and optimized for AI assignments. They are called TPUs (Tensor Process Units) and they will speed up most AI jobs significantly. They are now also made available through the same cloud solution for everybody to use.
This is not done for altruistic reasons. Of course, there is a business model behind it. It´s service based — you get the first fix for free and, after that, you pay per usage. But it is still a more democratizing model than having to build your own data centers and AI installations on site.
Centralized — and edge based, at the same time
Google is not the only company developing AI-specific hardware. Apple’s newest generation of smartphones, which uses AI for things like face detection and portrait photography, boast the A12 chip where the built-in Neural Engine can handle 5 trillion operations per second!
The world´s leading semiconductor company, ARM, whose chipsets you will find in most smart devices, sensors and IoT products across the world, has launched project Trillium to enable a new era of ultra-efficient machine learning, where most of the computation will be done on the spot, in the device, so you don´t have to send massive amounts of real-time data to the cloud but only the digested interpretation.
This means that we see a two-way development of more centralized cloud-based AI service platforms and ecosystems based on huge data centers and very advanced tools. And at the same time a movement towards the edge, which allows for local and real-time data handling. Both developments are necessary if we want to unleash the full potential of AI in a commercial setup.
What's the purpose?
All this can be quite interesting if you are into technology, but that is exactly what it is — technology. So, what are the purposes behind it? What kind of business value are companies looking to obtain by using AI?
At the Google Next conference in London, CXOs from large companies like Airbus, ING, Metro AG, HSBC, and E.ON flocked to the stage to explain how AI helped them with objectives like predictive maintenance, fraud prevention, crime investigation, sales forecasting and better customer service.
In the recently published report from Microsoft: “Artificial Intelligence in Europe”, the wide range of business purposes was divided into five categories:
- To predict
- To automate
- To generate insights
- To personalize
- To prescribe
with prediction (92%) and automation (88%) being the two most common in Denmark.
The report also showed that Danish companies (25 interviewed) are among the most advanced in Europe, with 96% stating that they are in the “piloting” AI stage or beyond.
So, if you are not already experimenting with how to gain business benefits out of using different AI techniques, it might be time to get started.
The other side of the coin
There are lots of issues to worry about.
This is an area that is still prematurely regulated. There are risks concerning data collection and data handling, not to mention the risk of information overload and false signals. Decisions made by algorithms are much less transparent than human-made decisions, which also pose a huge challenge when a company needs to explain their actions taken based on algorithmic guidance. Then there is the huge impact that the increased rollout of AI and automation will have on the job market.
And do we really want large international companies to host our data and run our intelligence?
Finally, I haven´t even touched on the fact that AI development in China and the rest of Asia is happening at lightning speed, almost under the radar. China, a country with 800 million internet users, heavy surveillance, no data regulation and a single point of decision, stated in their “Made in China 2025” plan that they aim at being the world leader in AI by 2030. And they are well underway: in 2017, 48% of total equity funding of AI start-ups globally came from China.
In other words, there is a global battle taking place as we speak. It will impact all of us and change the balance of power between countries and in industries within the next 5–10 years.
So, I strongly recommend focusing on this area and to start collecting relevant GDPR compliant customer data and to start experimenting with different AI techniques, no matter what business you are in.