_ Dr. Boris Hollas, professor at the HTW Dresden, faculty of Computer Science and Mathematics, member of the German study commission “Artificial Intelligence” Bundestag. Munich, 6 May 2021.
In recent years there have been great advances in image recognition and machine translation. These advances were made possible by numerous step-by-step improvements to already known processes and, last but not least, the availability of considerably more powerful hardware. This made it possible to use models for which the computing capacity was insufficient in previous years.
The industry in Germany has been using artificial intelligence processes for many years. This includes the use of robotics in production, image recognition in quality control, language processing to process customer inquiries, pattern recognition to detect suspicious activity and fraud. The aforementioned advances in image processing can further improve existing processes.
Craft businesses can only benefit little from artificial intelligence because customer requirements have to be treated individually, and most activities cannot be automated or can only be automated with unreasonable effort. Artificial intelligence methods are only suitable for uniform problems.
With the current methods, autonomous driving is only possible with considerable restrictions. Fundamentally new approaches of the AI with which these limitations could be overcome are not discernible.
Language processing methods are still unable to understand language and can therefore only support people, not replace them.
One difficulty when using data-driven methods, especially the deep neural networks successfully used in image processing, is the high, personnel-intensive effort required to obtain specific training data in sufficient quantity and quality. Therefore, depending on the problem to be solved, it can make sense, especially for SMEs, to use other methods.
The knowledge transfer between science and companies is guaranteed by the IT graduates, the transfer centres and cooperation between universities and companies.
The number of computer science graduates at the Technical University of Munich increased by 259 percent between 2013 and 2020. A similar development can be observed nationwide. At the same time this goes hand in hand with major deficits among the first-year students, especially in understanding abstract and mathematical relationships. The MINT subjects need to be strengthened, the requirements in mathematics lessons in high schools need to be raised. Alternatives to studying computer science should be promoted. There are also extensive tasks in artificial intelligence that do not require a degree.
Social, ethical and environmental impact
Since the improvements that have been achieved in some areas of artificial intelligence are a further development of an existing technology, effects on the world of work can be expected that will remain within the scope of other technical developments. Deep neural networks require a great deal of computing power and thus electricity for training, but not for their application. Since large data centres in countries with low electricity prices can be operated and connected via the Internet, the use of AI is not expected to have any effects on electricity consumption in Germany that exceed the level associated with increasing technology.
Regulation and political agenda
Advances in image and speech processing have been driven by the large US-based technology groups, which benefit most from them. German companies only offer these processes an advantage for some of their products or production processes. There is political resistance in Germany to military applications that are driving the development of intelligent robotics and pattern recognition in the USA. Due to the very high tax and levy burden in Germany and higher salaries for AI specialists in Switzerland or the USA, Germany has a locational disadvantage.
The need for data protection is less pronounced internationally than in Germany. In addition, foreign end customers are more prepared to accept restrictions in terms of data protection if the product can be used better as a result.
The GDPR requires numerous documentation and information obligations that can complicate the collection and processing of data, as they are necessary for training learning processes. The additional effort can be uneconomical, especially for SMEs. Automated decisions are subject to severe restrictions, so that there is little incentive to develop them in Germany. Uncertainty in the interpretation of the regulations can prevent companies without their own legal department from developing AI products that process personal data.
The establishment of new professorships for artificial intelligence as part of the high-tech agenda harbors the risk of damage to other subjects in the short term and to AI in the long term. Since AI professorships that are now being filled will not be filled again in the next 20 years, this career path is blocked for young scientists, which is why they can turn away from AI. Science needs reliable funding across all areas that is secured in the long term.
The state should create the framework conditions that are necessary for the development of the economic activity of companies, but not try to exert a direct influence on their investments through AI funding programs. Companies are better able to decide for themselves how to make the most of their financial resources. The effort involved in preparing and reviewing funding applications deprives the economy of productive working time.