Artificial Intelligence: An Exchange of Views | Topics
Current applications and the vision for future developments in AI
In this interview, industry experts Samuel Thomas Stähle, CEO of PowerBrain.Shop®, and Ronald Sieber, CEO of SYS TEC electronic AG, shed light on the latest trends and developments in the field of artificial intelligence.
In your view, how have the development, architecture and capabilities of artificial intelligence software – commonly referred to as ‘artificial intelligence’ – evolved over the last few decades?
Sieber: In recent years, the industry – and our company included – has made significant progress in the development of ultra-compact computer architectures and embedded systems. This has enabled us to increase their robustness and steadily reduce their prices. This is largely due to the ever-increasing integration density of modern components. Ultimately, this has meant that today’s industrial edge devices have many times the computing power that was available to the Apollo spacecraft some 50 years ago, which carried people from Earth to the Moon and back again safely. Today’s edge controllers are now so powerful that we can run modern AI software directly on them.
Stähle: From a computer science perspective, the algorithms and data structures underpinning AI have been the subject of research and development for around 70 years – and this process is far from complete. It will probably never be completed, but will continue to evolve – with a steady trend towards standardisation and simplification in terms of usability.
I programmed the first AI algorithms myself more than 20 years ago, back in the 1990s. We used them for optical character recognition to assess the quality of printed text on curved surfaces in Germany. We used expensive and large specialised computers, very expensive digital line-scan cameras and optics, as well as software we had written ourselves. At that time, the idea of purchasing modular AI as a product or service was unthinkable.
Today, however, there are much cheaper and miniaturised computers, complete with specialised chipsets for AI data processing. AI software is now available on the open market in industrial-grade quality, either as a cloud service or for use directly at the edge, such as our Edge AI PowerBrain™.
This has, of course, revolutionised the use of AI and ultimately follows the logic of a historical market observation: a new quality is superseded by a new quantity, which in turn – iteratively – will be superseded by a new quality, and this in turn by a new quantity. Over the last quarter of a century, it has become apparent that this cycle, as a development paradigm, also encompasses AI.
What is your perception of the market – what is the state of customer acceptance and project applications of AI in 2021?
Sieber: For years, our company has been steadily producing more and more AI-enabled embedded systems on behalf of our customers, for example using i.MX 7 processors or NVIDIA® chipsets and large memory configurations. We would therefore describe the demand as steadily growing – primarily from the Industry 4.0, mobility and infrastructure management market segments.
In the field of Industry 4.0, users are increasingly interested in who they entrust their data to and what happens to their data. Concepts such as data sovereignty and data minimisation are becoming ever more important. Here, our edge controllers offer the major advantage that the devices already possess sufficient computing power to enable AI projects to be implemented locally on-site. In most cases, only status information or error messages are then transmitted to the cloud, whilst the often sensitive sensor readings no longer leave the local edge controller itself.
A similar scenario also applies in the mobility and infrastructure management sectors. In addition to protecting sensitive data – which can often be directly linked to specific individuals – edge computing is specifically utilised here to ensure independence from fluctuating internet connections. The Edge Controller-based AI system thus operates largely autonomously and can, where necessary, fall back on a cloud-based backend for actions where timing is not critical.
Our case study: Edge AI thanks to NVIDIA® Jetson Nano™
In your opinion, how has project work in the field of AI changed between the start of your career and today?
Stähle: Over the past quarter of a century, technology has – naturally – advanced considerably.
On the hardware side, significantly greater computing capacities are now available at lower prices. Modern, industrial-grade hardware features powerful processing architectures and high data storage capacities in both end devices and sensor systems.
On the software side, numerous programming languages and software development tools have come onto the market, which make life significantly easier for AI developers and boost their efficiency. Numerous tools are now also available to data scientists to improve efficiency when analysing and simulating training data.
Our PowerBrain.Shop® team has also been able to contribute to innovation in AI development. For example, we have integrated AI into products and software services for standard industrial use cases; these can be purchased to an industrial standard and deployed within minutes as AI PowerBrains™ in project and solution-based business. Implementation is usually carried out by integrators or manufacturers such as SYS TEC electronic AG. This has represented a major step forward in terms of ‘ease of use’, thereby facilitating market entry for many small and medium-sized integrators and manufacturers.
One serious ‘megatrend’ that we have seen emerge and slowly fade away over the last two decades is the narrative of cloud AI, driven by data centres, manufacturers, operators and government bodies. The assumption was that AI always required huge computer farms (aka the cloud).
Nowadays, this perspective is shifting noticeably, as computing power and sensor data processing and storage capacities in edge devices are steadily increasing. The historical dependence on data centres is thus slowly dissolving. Much like the evolution of the ‘personal computer’, it is being replaced by ‘personal AI’ in local embedded systems, offering higher quality and greater independence. This eliminates the costs of data aggregation to the data centre or the cloud, cloud transaction costs, and attack vectors originating from the internet.
Which success factors play a particularly important role in the development of edge AI projects?
Stähle: We believe a holistic view of the specific use case is crucial. In the field of condition monitoring or predictive maintenance, for example, it is important that sensor data from the critical points on the machine or plant is available in sufficient quality and in a timely manner. After all, machine learning models can only learn from the data that is provided during training and is available during operational use. The resulting performance and quality of the Edge AI in operation reflect the context(s) in which the training datasets were collected, or the focus the project team had when selecting the sensor data and the locations where it was collected.
Sieber: Indeed – we have found this to be the case as well. It is absolutely crucial to install vibration sensors at ‘hotspots’ for vibrations on a machine or plant in order to capture the actual mix of oscillations and vibrations as early as possible and with maximum precision. In particular, to protect high-value machine components such as bearings and shafts, as well as gearboxes and drives, the correct choice of vibration sensors and their mounting locations is crucial for optimising their maintenance.
What future trends would you expect to see in artificial intelligence?
Stähle: Presumably – much like with computing systems and computers – there will be a two-pronged development.
On the one hand, gigantic ‘data octopuses’ in clouds and data centres will process increasingly complex data structures, AI models and algorithms to process the vast amounts of data collected, thereby developing, for example, highly specialised AI solutions such as DeepMind’s AlphaFold 2 to predict protein folding and advance scientific understanding.
On the other hand, the ever-increasing availability of computing and storage capacity in end devices will drive the ‘democratisation’ of AI: every company will be able to train and deploy its own on-premises edge AI on machines and plant, completely independently of the internet and the cloud. An ever-increasing number of signal processing and deep learning algorithms will be utilised in this context.
The next major qualitative step would likely be the kind of potentially revolutionary research being carried out, for example, by Prof. Christoph von der Malsburg, aimed at finding even more intelligent and optimised AI data structures and algorithms. These would subsequently also be deployed in edge AI and data centres.
Furthermore, we also expect many optimisations in the field of AI training visualisation, monitoring and quality assurance, which will enable us, for example, bias – that is, the bias inherent in AI models – to be detected, and the mental models of the trained AI to be visualised – so that opportunities for structural improvement can be identified here too, in the interests of continuous quality assurance. All of this without the need for a university degree in data science or AI.
Sieber: We have already successfully completed our first projects using PowerBrain™ on our sysWORXX CTR-700 edge controller, and we are amazed by the impressive results delivered by edge AI. The computing power required for this can certainly be described as moderate. Overall, the sysWORXX CTR-700 is powerful enough to carry out signal pre-processing and the actual machine control simultaneously alongside the edge AI. I think this points to the trend for the future: innovative edge controllers such as the sysWORXX CTR-700 will take on increasingly complex control tasks and, in doing so, will also increasingly process multidimensional signal types such as structure-borne noise, which – thanks to integrated AI solutions such as PowerBrain™ – are reduced locally to information vectors that are easier to analyse.
Stähle: Speaking of vectors: reducing the number of potential attack vectors to zero represents another fundamental paradigm shift. The edge AI software operates autonomously on-site within the embedded system and requires no backchannel to the internet or to hosted server farms. This makes it impossible to establish contact with the artificial intelligence from the outside and manipulate it. Given the annual rise in cybercrime and automated attack attempts, this maximisation of cyber security is a major driver for our end users. Furthermore, we are seeing significant customer interest in data sovereignty, data security and data protection.
Following the countless leaks and sales of highly personal and confidential data on the dark web and similar platforms in recent years, users today place greater value on security. They no longer wish to transfer their machine, operational or plant data halfway around the world unless absolutely necessary.
Even training data for artificial intelligence – which could provide information on plant operating modes, their status or, for example, production capacities – is hardly ever stored and processed by users on cloud servers in arbitrary locations around the world without question. Rather, the paradigm is shifting here too, driven by the availability of increasingly affordable and high-capacity local storage and memory cards within embedded systems. This increases the degree of self-sufficiency for users and their machine fleets, whilst simultaneously reducing costs and dependencies.
Essentially, Edge AI therefore represents a revolutionary change in the implementation and operation of artificial intelligence. I see this development as similar to the revolution that once accompanied the invention and market launch of affordable personal computers (PCs) for individual households and businesses in the 1970s. The days of oversized data centres were numbered. Who still remembers the quote attributed to Thomas J. Watson, the former chairman of IBM: ‘I think there is a world market for maybe five computers’?
What has emerged from these centralised mainframe systems are systems that are now far more flexible, reliable, self-sufficient and cost-effective, with significantly greater stability. This is made possible by decentralisation and ‘democratisation’, through a much greater proliferation of information technology.
We will all witness the continuation of this development in the field of artificial intelligence as well – and enjoy much more secure edge AI in both our professional and private lives. Moreover, it will be much more affordable in future. Even today, for example, customers of our AI PowerBrain.Shop® can create and use their own Edge AI PowerBrains™ for various end devices, embedded systems and hardware platforms within minutes. They benefit conveniently and cost-effectively from this natural trend.
Would you also like to benefit from the opportunities offered by AI in edge devices? Get advice now!
About our partner: POWERBRAINSHOP Holding Corporation
PowerBrain.Shop® is a growing AI software provider with a global presence, which develops an AI product development and training platform for the next generation of artificial intelligence, as well as powerful AI software ‘brains’ – known as AI PowerBrains™. The impossible is made possible by enabling and significantly simplifying the next disruptive step in development – the widespread adoption of internet-independent edge AI – including the development, training, validation and quality assurance, deployment, monitoring and auditing of artificial intelligence for users and use cases across various industries, business sectors and organisations. As a driving force in the artificial intelligence sector, we partner with all those who wish to make our world more efficient through intelligence.
