Is artificial intelligence or machine learning exaggerated

AI and the cloud - a panacea for everything?

What are we talking about anyway? “Artificial Intelligence (AI)” or “Artificial Intelligence (KI)” in German describes the basic concept in which machines perform “smart” tasks. The subordinate term “machine learning (ML)”, on the other hand, represents the learning process with which a system itself acquires a certain behavior based on data. In general it can be said: ML is learning based on patterns, examples and experiences. However, the algorithms are still quite “simple” today. So far, programming has mainly consisted of defining rules.
Machine learning is the program that rules these
for us and continuously improved over time with the help of data.

AI is not just high tech & robotics

Artificial intelligence is often associated with robots
that will be of service to us at some point. This image is overdrawn in the media - AI often simply means that data can be used by smart algorithms to support processes. Even today, numerous AI applications from the cloud help us in our everyday life: Google Maps automatically guides us on the fastest route to our destination, households use intelligent devices to measure and independently adjust their power consumption, and the best e-mail spam Filters work thanks to AI. Machine learning can do many practical tasks in companies too: from helping with resource allocation, predicting and responding to future customer needs, to automating tedious and repetitive tasks.

Pre-trained AI systems from the cloud

Various technology providers want to make artificial intelligence more accessible and provide interfaces, so-called APIs, with which every organization can integrate pre-trained AI systems into its own processes. There are already numerous APIs that are used for image analysis. With Google AutoML, you can now create your own image recognition
model can be trained by being fed with the help of predefined models of image types, from which it learns and with relative
Various types of images, e.g. screws, can be recognized and named with little effort.

Data is an important part of AI development. And what is indispensable for an AI application is a data-driven task - i.e. a challenge based on information. However, these do not have to be available in your own company. Numerous digital data sets are used by communities interested in very different topics - from «wine reviews» to «urban sounds» to «black-
Friday consumer behavior »- in text, satellite images and
Videos made available for free.

The recognition and naming of images is already used by various companies and is always possible today
be trained more specifically. With Google AutoML, Google Cloud provides a system with which anyone can create and train their own ML model without much prior knowledge.

Two examples: Disney uses AutoML Vision to quickly assign products and product photos to specific categories in the web shop. For example, a product manager uploads photos of a T-shirt with a Spiderman motif to the website and the product is immediately labeled “Spiderman”, “T-Shirt”, “Marvel” and “Superhero”.

The Zoological Society of London (ZSL) is committed to protecting animal species around the world. In order to perform this task effectively, so-called camera traps have been installed all over the world. In this way, animals in nature can be recorded and counted. With AutoML Vision, the ZSL has trained an ML model that recognizes which animals have been photographed, which enables automated classification. So many weeks of work turned into just a few hours or even just a few minutes.


Computing power from the cloud

What does all of this have to do with the cloud? Machine learning, deep learning and artificial intelligence in general require a lot of computing power. One easy way is to rent high-performance hardware in data centers that users can access over the Internet. The latest technology that greatly improves computing power for AI systems is Tensor Processing Units (TPU). TensorFlow is an important part of numerous AI and ML systems. The TPU processor developed by Google is designed to operate the TensorFlow open source AI framework particularly quickly.

As a result, AI systems can be accelerated by 15 to 30 times. That corresponds to a leap of seven years
into the future compared to previous development cycles. In our data centers, we can install a large number of so-called pods on which the TPUs are attached. One pod alone has an output of 11.5 petaflops. Hardly any company can provide such hardware locally. Out
for this reason, cloud computing plays a key role in the development of AI systems.

The three points for a successfully applied AI are therefore:

  1. It takes high quality data sets that ML systems use to learn patterns. Ideally, you train ML systems with data that is representative of the real world.
  2. Good tools and frameworks are essential. Although the basic ML algorithms can be described in a few minutes, they are quite complicated to implement. Therefore, a number of services are required that do not require any ML or programming knowledge.
  3. A particularly large computing power is necessary. The cloud offers this powerful hardware that companies and developers can use for their own ML model.

With these simple steps you can get started with the application of AI in the cloud.