From the Space Station on to New Galaxies: About Digitization, Artificial Intelligence and Machine Learning
Observing the current discussion in society, politics and industries, one cannot help feeling that Germany is finally awakening from its digital slumber: Suddenly, digitization, artificial intelligence (AI) and machine learning (ML) are shaping industry meetings as well as political debates in the Bundestag. Recently, the German government even launched the “Strategy for Artificial Intelligence,” which is worth three billion euros and aims, among other things, to make Germany and Europe one of the leading locations for AI technology. All under the premise of a responsible development and use of AI in the public interest — which should certainly be a central element of our further efforts around artificial intelligence.
The legislator is thus only reacting to a development that has already arrived in all of our lives. The breakthroughs in speech and image recognition, machine translation, autonomous driving and the boom in intelligent assistants such as smart speakers are all a result of AI. But compared to what will yet be possible on a large scale, these examples are just the beginning. Countries worldwide — but above all in China and the United States — are working at full speed first of all to fathom the potential of AI and secondly to make it available to people. Beijing alone has announced that by 2030 it will have established a government-sponsored 150 billion dollar industry around AI. So it is no longer a question of whether we will see the big breakthroughs here in the near future, but rather when.
Digitization as a Space Station
Perhaps exploring AI can be compared to the journey of mankind into space: The prospects are limitless, there are plenty of assumptions about what will be possible, but it takes a lot of effort and hard work to make tangible progress. To remain in this picture, digitization is the ISS from which the complex that is AI is being researched. And just like the International Space Station, digitization has become firmly established. More and more companies are observing these three characteristics of successful digitization projects:
- Separate activities and departments are connected and synchronized.
- New technologies automate workflows and processes. This increases productivity and effectiveness.
- Efficient data management and integration of the Internet of Things (IoT) ensure better decisions and more efficient processes.
These changes are by no means limited to individual, closed silos. Forward-looking entrepreneurs (supported by capable CTOs, CIOs or CDOs) are subjecting their entire organization to digital transformation. Only in this way can the benefits of digitization really be exploited. By thus promoting the digital mindset, the first building block has been laid to establish AI and ML on a broad scale. What about the others?
The Democratization of Machine Learning
The best technologies are of no use if, due to their cost or complexity, they are only reserved for a select circle of sophisticated professional users. This has long been the case with artificial intelligence. But at the latest since the “Conference on Neural Information Processing Systems” in Barcelona in 2016 this has fortunately been history. This was one of the first major events at which Apple appeared for the first time alongside all the major players in the industry, following the suit of companies such as Google and Microsoft and allowing insight into their research around AI. Especially for the goal of democratizing technologies, it is important to make publicly available not only research results but also the tools and data collections that are necessary for the development of software and new products.
By now, computing power has become much more powerful, primarily thanks to the cloud. This makes the technologies for using AI more user-friendly and, above all, cheaper — and they can finally also be used by non-IT users. In addition, it has become increasingly easy for companies to aggregate data for AI applications. And these changes are bearing fruit: According to a study conducted by Crisp Research on ML in corporate use in 2017, more than 63 percent of the companies surveyed are involved in AI. 72 percent are already implementing concrete projects or have gained initial experience with prototypes.[1] Volkswagen CIO Martin Hoffmann summed up this development at the AutomotiveIT Congress in Hanover in March 2017: “Machine learning is no longer science fiction, it’s here and real today”.
And what can AI achieve in concrete terms?
With regard to the fact that we are only at the beginning of a worldwide boom, a lot is already possible with AI and the corresponding methods — the real potential of the current state of this technology is however always exploited by combining one or more of the following applications:
Classification: Is an object this or that?
Similar to Cinderella, AI sifts through data sets based on specific questions: Can product A or product B be seen on the picture? Is this e-mail spam? Does this product have a defect?
Predictions: Which numeric value is next in a sequence?
An example is the sales forecast for a product, based on data such as previous sales figures, consumer sentiment and weather. It examines when and with what probability an event will occur.
Clustering: What belongs together?
Data instances with common or similar characteristics are grouped together. This is used, among other things, in market research. Demographic data, preferences and purchasing behaviour can be used to form well-defined consumer groups.
Correlation: Which events occur together?
The point is to recognize correlations as correlations (if that, then this) and not as causality (this causes that). For example: A critical system state is always reached when certain parameters (temperature, pressure, …) exceed a certain threshold. The exact nature of the relationship between the two events must be interpreted by a human. A common mistake here is the confusion of correlation with causality. In causality, event A is the cause of event B. This is not necessarily the case with correlation. An example for this typical error: The majority of the population dies in bed. Is it therefore better not to go to sleep?
Optimizations: What is the best solution for a task?
AI optimizes the results for a specific target function. A common example is planning a route that optimally combines time and fuel consumption.
Anomaly detection: What doesn’t fit?
This subcategory of classification can be used to find out which data in a particular set is exceptional. A practical example: You train a system on the vibration values of a machine. With continuous monitoring, new values make it clear whether the machine is still operating normally or not.
Ranking: Are there any recommendations?
Proposals for action are developed on the basis of training data. A typical example: systems that suggest products customers could buy next based on their purchasing behavior and related groups of people.
Artificial Intelligence in Practice
Predictive maintenance
Predictive maintenance combines anomaly detection, correlation and prediction: When does an anomaly occur? When will this happen again? How often will it reoccur until it negatively affects the machine's operating condition? To do this, an ML system uses the data from several sensors installed in machines or equipment to monitor relevant parameters such as temperature or vibration. These and other data, such as from microphones or cameras, can be quickly analysed using Deep Learning. This enables better planning of maintenance work and the reduction of personnel costs and downtimes.
In the automotive industry, where unplanned downtime of welding robots and the associated production downtimes cause costs of just under one billion euros per year, the corresponding investments would have quickly paid for themselves. Finely adjusted sensors send warnings to the driver before a certain component no longer functions and the vehicle stops.
AI-controlled logistics optimization
In the logistics industry, ML can optimize route planning, reducing fuel consumption and shortening delivery times. Sensors that monitor vehicle performance and driver behavior provide real-time feedback to drivers, for example, when braking or accelerating would make sense in order to drive fuel-optimised. This also reduces the maintenance costs of the entire fleet and can reduce fuel expenses by up to 15 percent.
For a responsible use of AI
With all the - admittedly very promising - prospects for the future, one must never forget that not all people share this optimism about technology. Many people are afraid that their jobs will be digitized or taken over by robots. And you have to be honest and realize that AI will make jobs redundant. But usually only those that are particularly monotonous, physically demanding or dangerous but necessary.
However it is also a fact that technological progress has always created more new jobs than destroyed them. And this is also the case with AI.
Initial studies show that 2.3 million new jobs will be created by artificial intelligence by 2020, while only 1.8 million will be lost.[2] Man and machines will work together even more closely and successfully in the future. As is the case in the automotive industry, which relies on collaborative robots: Here, cobots have been working successfully alongside their human colleagues for several years. This example clearly shows that machines relieve us of tasks that they can do better and faster, and make us stronger where we really need support. This way we can rely on our talents - such as creativity.
A win-win situation for all of us.
A guest article by Dr. Elisabetta Castiglioni, CEO at A1 Digital. A1 Digital advises companies on questions of digital transformation and accompanies them in the digitalization of their business areas. To find out more about Porsche and Technology, follow us on Twitter, LinkedIn and Instagram.
References:
[1] Crisp Research, 2017 https://www.unbelievable-machine.com/downloads/studie-machine-learning.pdf
[2] Gartner 2017, https://www.gartner.com/newsroom/id/3837763
