AI and the Cloud Go Hand in Hand

Artificial intelligence and cloud computing dominate both the IT industry itself as well as business in general. Between the omnipresence of mobile devices and the accompanying apps that rely on cloud services, both topics have become permanent features of our daily private and business lives. Outside technical circles, terms like AI, machine learning (ML), deep learning and cloud are often thrown around interchangeably and incorrectly. The following article will define them, clarify the differences and use a public research project to illustrate how they fit together.

What exactly is AI?

There are numerous definitions of artificial intelligence. In computer science, AI research is defined as the study of "intelligent agents," that is any device that perceives its environment and takes actions that maximize its chance of successfully achieving its goals. A more elaborate definition describes artificial intelligence as "a system's ability to correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation.”

The three core pieces of AI are:

  1. Goals
  2. Data
  3. Algorithms

Intelligent behavior is produced by a set of instructions or algorithms, whose aim is to reach a specific goal. That may be something simple like “Win a game of Go” or more complex such as “Perform actions which are similar to ones which succeeded in the past.”

Part of the real world is represented as machine-processable data and can be automatically analyzed using mathematical methods and logic. Approaches such as this date all the way back to Aristotle.

Let’s get a little more technical and look at some of the different methods.

State Space Search

State space search is a process in which successive configurations, also known as states, of an instance are evaluated in order to find the goal state. This has historically been the most important method in AI and has its origin in the mathematician Leonhard Euler (1707-1783), who introduced a kind of graph theory, which still serves as the basis of many systems today. Rule-based systems are usually used to solve problems described in this way.

Behavior-based AI

Behavior based AI tries to mathematically analyze, model and predict the behavior of artificial, social and biological systems. Cybernetics for example, is a well-known subfield and this method is often successfully used in robotics (Cambrian intelligence).

Artificial Neural Networks (ANN)

ANNs, or just neural networks, have been studied since 1961. However, they were largely forgotten as the state of computer technology and computing power did not allow for any meaningful use or experimentation at the time. In neural networks, knowledge is represented by patterns that represent themselves and objects in the real world. Since patterns can be represented as mathematical vectors, similarities between different patterns can be calculated. These neural networks are nowadays often seen as synonymous with AI, but they are only one part of it.

To summarize: Artificial intelligence is much more than machine learning or deep learning. There are a variety of methods, each of which is best suited to different use cases. However, as Harald Huber's blog post explains, for certain applications it can be cheaper, more useful and overall more pragmatic not to train intelligent behavior by machine learning, but to achieve it via other methods.

As of today, most good solutions need one or more humans to create the AI algorithm or model and someone who contributes the technical expertise, i.e. either contributes the knowledge to be represented or evaluates the self-learned results in order to improve the learning algorithm (supervised learning). We will come back to both points again later.

Advantages of the Cloud / SaaS

Cloud computing refers to on-demand computer resources such as storage and computing power without direct management or hardware access by the user. Google’s suite of online services provides a simple example from everyday life. The physical hardware and mail software for Gmail are hosted on a distributed series of servers worldwide and available on-demand in any browser or mobile device. Users do not need to build, configure and maintain their own physical mail server nor run local mail software such as Thunderbird for example, in order to send and receive email.

The idea behind this is to make IT infrastructure available via a computer network so no local installation, setup or management is required. Cloud computing comes in three main types:

  1. Infrastructure as a Service (IaaS)
  2. Platform as a Service (PaaS)
  3. Software as a Service (SaaS).

IaaS primarily involves the provision of technical infrastructure such as computing power, networks or storage space. The scope of these services can be adjusted on the fly time to meet changing needs. For example, an online provider of medical masks may need to quickly scale up the number of CPUs and RAM their server has to cope with the flood of orders generated by COVID-19. This can be done in a few clicks today, instead of having to physically purchase hardware, take the server offline, upgrade and test, and then relaunch it.

PaaS offers users a platform for developing and offering their own software applications. For this purpose, programming and development environments are made available.

SaaS is the most widely known and understood, also known as Software on Demand, and represents the top level of cloud computing. In the past, users had to purchase physical copies of software on disk or CD and install them locally. This developed further into downloading the application from the internet, installing it locally and activating it via an unique license key.

SaaS takes this to its logical conclusion, with the software existing solely in the cloud and not locally while users pay a recurring fee to use it.

There are many advantages of these approaches, both economic and technical. Financially, it enables both software providers as well as consumers to better manage costs as they are spread out over time and thus smaller than one large initial investment. For providers, it helps manage cash flow and ensures a steady stream of predictable revenue.

For companies, IT departments and even everyday users, cloud computing offers significant time and cost savings over managing one’s own infrastructure, not to mention the flexibility of spinning up new resources on demand. This may be a developer who needs a new system for testing for a few days or a company looking to permanently expand its resources.

Moreover, software as a service ensures the end user or organization receives regular updates and upgrades by the provider, both for hardware and software, as well as do not need to concern themselves much with the implementation of legal requirements for data protection or IT security.

For software providers, the cloud-based approach to software provision also offers advantages in terms of maintenance. Only one system needs to be maintained for all customers. Salesforce for example, can boast that all its customers are running the same version of its software versus say Microsoft users worldwide running a wide range of Windows versions which presents problems.

Special attention is also paid to so-called platforms or PaaS. The approach is designed to provide a centrally available development platform, where users supplement their own functional units using existing modules or basic functions and combine them into a complete system.

This approach may have far-reaching consequences, because entire software ecosystems may emerge in which the added value no longer comes from a solution from one manufacturer, but from the contribution of several so-called market participants. The most prominent and first major example of this kind is Apple's Appstore. The added value no longer comes from the mobile device and proprietary applications alone (as was previously the case with e.g. Nokia), but from the open (but controlled) platform and ecosystem of applications provided by third parties. Amazon’s pivot from seller to marketplace with its decision to allow 3rd party sellers on the site is similar.

The disadvantages, however, should not be overlooked. Depending on the design and extent of IT outsourcing to the cloud, a so-called lock-in effects can occur, i.e. the user is de facto bound to the cloud provider because switching would be too time-consuming and costly. This is the case for many companies locked into legacy enterprise products such as Microsoft Office despite the fact that newer competitors may offer better features and pricing.

In addition, large, centralized platforms are more attractive to attacks. It is true that higher technical security measures are generally implemented. Yet so much data is stored there (and thus so much value), that the incentive is very high to attack it. The long list of major data breaches over the past several years is evidence of this.

AI in the cloud: 1+1=3?

Without cloud computing, the current machine learning (ML) hype would not be possible. Processing and learning from massive amounts of data can be an extremely computationally intensive undertaking. Only easy access to cloud-based infrastructure (IaaS) or AI services such as machine learning (SaaS) and the computing power required for this can enable the development of new "intelligent" products, services and business models.

This is especially critical for small and medium-sized companies. Developing high-performance and scalable AI systems can quickly become an expensive undertaking. After all, training algorithms requires huge amounts of computing power in the learning phase, which are then no longer needed later to operate the corresponding analytics systems. Scalable cloud resources are a clear and major advantage here.

Furthermore, the development of AI applications or the intelligent extension of existing applications requires easy access to computing power, data, connectivity, but above all additive platform services (e.g. certain AI or ML methods). This also facilitates cross-departmental and cross-company collaboration between AI and ML experts. The central approach of the cloud also supports a predominant paradigm in AI development: centralized learning and decentralized execution. It is advantageous for example to combine data from multiple production systems centrally, to learn from the larger pool of data and thus to develop better algorithms.

The cloud is therefore a clear advantage for AI, although not a necessity. For example, USU is developing such an AI cloud as a platform approach.

AI and Cloud in practice: The Service Meister project

Service-Meister is a project funded by the German Federal Ministry of Economics and Energy (BMWi) to research and develop an AI-based service ecosystem for technical services in Industry 4.0.

Service-Meister Logo

German industry is undergoing a fundamental change in value creation from products to services. Novel business models are in demand, which require German SMEs to use and market their in-house "service knowledge". However, the service knowledge required for industrial plants exceeds the knowledge of individual service technicians and in some cases even of entire companies. Due to the lack of skilled workers, German SMEs will face an enormous challenge in the coming years to secure their lead in the provision of services.

In order to support German medium-sized businesses in this process, Service-Meister will develop an AI-based service platform that spans systems, departments and companies. An important sub-goal is to enable less educated specialists to provide complex services with the help of digital advisors, e.g. AI-based service bots and smart services. The second sub-goal is to enable cross-company scalability of services by providing the digitalized service knowledge on a platform. This will create a service ecosystem that will counteract the shortage of skilled workers in Germany and make German SMEs competitive in the long term.

The project therefore combines both of the above approaches. On the one hand, AI-based services are developed to support business users and then provided centrally on one platform. The entire range of different AI methods will be used. Perhaps the best example is the intelligent service bot. As discussed in other blog posts, in many use cases it makes perfect sense not to implement bots via machine learning, but to map essential components such as intent recognition and dialogue strategy.

On the other hand, bots can benefit immensely from topics such as machine reading, i.e. special ML methods for indexing and "understanding" large amounts of text. The goal of Service-Meister to combine both approaches, i.e. to automatically process texts with machine reading and to make this knowledge available to bots as a knowledge base. The bots communicate via other, more controllable methods and thus do not run the risk of painting (or learning) themselves into a "corner" so to say in an uncontrolled manner.

The second aspect that is special for users is the central provision of such AI-based services on a single platform. All-encompassing, omniscient AI does not exist. Such AI-based solutions are currently and for the foreseeable future selective "idiots of expertise", i.e. tailored to specific functional or thematically limited areas. Only an accumulation of intelligent support systems along the entire service chain, also supplemented by partners or the customer himself, will allow a value-added system or ecosystem to emerge.

Share article:

More interesting articles