Artificial Intelligence (AI) in Engineering
Martin Eigner, EIGNER Engineering Consult, Baden-Baden 2024
.
The current optimism about technology and the simultaneous concern about possible effects on employees and labor markets are based on the idea of leading economists that artificial intelligence (AI) - like digitalization - is a ‘universal technology’. Such technologies have overarching effects on the entire economy [1]. In this post, the focus will be limited to the area of engineering. Although these technologies often take years and decades to realize their potential, they can trigger a sustained productivity boost and strong economic growth.
The term AI (artificial intelligence) was first mentioned in 1955 [2]. Nevertheless, there is no generally accepted definition, only a multitude of interpretations shaped by different perspectives [3]. This may also be because there is no generally accepted definition even of intelligence.
Opinions differ widely in the scientific discussion about the current state and relevant research goals of AI. Sam Altman, CEO of OpenAI, and Mark Zuckerberg, CEO of Meta, believe that the Large Language Model (LLM) represents a significant step forward on the road to artificial general intelligence (AGI). The CEOs of the other AI providers (Antropic, xAI) assume that AI at a human level can be developed between 2025 and 2029 [4]. Yann LeCun, Chief Scientist for AI at Meta, takes the opposite view. He is skeptical of the assumption that Large Language Models (LLMs) will lead to the creation of an artificial general intelligence. In his opinion, these models have only a very limited understanding of logic and are not capable of thinking and planning like humans. LeCun believes that, while possible in principle, AI is not yet technically feasible. It will take some time before a breakthrough is made [5]. This view is also shared by Bernhard Sch?lkopf, a leading German researcher in the field of machine learning and causality. Both researchers are making an important contribution to the further development of AI with their realistic and innovative approaches. The view that it will take a long time before AGI is achieved is shared by 1,712 leading researchers. They estimate the probability of a machine being developed by 2028 that can perform tasks better than humans at 10 per cent. Even these researchers believe that the probability of this occurring by 2047 is only 50% [6]. ?
AI models with up to eight levels are presented, which extend very far into the future [7]. Daniel Frei has put together a very interesting and readable five-level classification of AI [8]. I use the rather pragmatic three-level model from IBM and extend it to a four-stage model [9]. However, the three levels of IBM should be preceded by a clear distinction to highly intelligent systems, which are designed and realized by humans - mostly through hard-coded rules. I want to introduce this level because the term AI is often used incorrectly. I call this level "Human Intelligence (HI)". The main difference between these human-programmed systems and actual AI lies in the way they provide learning, processing and decision-making capabilities:
1. Human programmed systems (Human Intelligence - HI):
These systems follow fixed rules and algorithms developed by humans and implemented in software. They are designed to perform certain tasks efficiently and accurately, but they are not capable of learning and therefore cannot expand their application independently. However, they can perform self-contained tasks that humans can no longer perform without their help. This level currently plays an important role in engineering. It should be emphasized here because the term AI is often used for it. Figure 1 shows some examples:
2. Artificial Narrow Intelligence (ANI):
Also known as weak AI. ANI is widely used in engineering to perform specific tasks with high accuracy and efficiency. Here are some examples [10]:
领英推荐
These applications show how ANI can improve the efficiency and accuracy of various technical tasks. ANI is highly specialized in the tasks it is assigned, but it does not have general intelligence or the ability to perform tasks outside of its specific domain. It also lacks general adaptability and the ability to transfer learned knowledge to other contexts. Without human intervention, ANI cannot independently expand its range of functions, make autonomous decisions or develop new contexts. It is certainly debatable whether the creative element is really missing at this level. Although ANI is based entirely on data learned using ML, it is also possible to combine learned 'knowledge' in ways that have never been experienced before. In summary, this level is characterized by solutions that enable a high degree of automation and thus increased performance, but which can still be implemented (more slowly) by humans.
3. Artificial General Intelligence (AGI)
Also known as strong AI. AGI is a theoretical form of AI that aims to replicate human intelligence so that machines can understand, learn and apply knowledge to a variety of tasks. While AGI has yet to be realized, there are some hypothetical examples of what AGI could achieve [12, 13, 14]:
These examples illustrate the potential of AGI to transform many aspects of our lives. However, it is important to note that AGI is still a theoretical concept and that significant advances in research are still needed to make it a reality. It is crucial that the quality of outcomes and the range of tasks reach the human level at all levels (e.g. including emotional). Figure 3 illustrates what AI needs to do to reach the AGI level.
4. Artificial Superintelligence (ASI):
This is a hypothetical AI that exceeds human intelligence in all aspects, including creativity, problem solving and emotional intelligence. ASI is still a theoretical concept and is projected far into the future [1].?
Literature
?
Professor no Programa de Pós-Gradua??o em Engenharia de Produ??o da UFABC
3 个月Highly recommended
Digital Business/IT Strategist | IT Director | Program Management | Enterprise Architectures | CRM-ERP-PLM-SCM Consultancy |
3 个月Martin, At this moment Alphabet is writing more than 25% of its software code with AI. https://fortune.com/2024/10/30/googles-code-ai-sundar-pichai/ Alphabet’s Deepmind initiative is capable of creating models in many different areas. DeepMind has specialized in applying machine learning to develop functional algorithms for many uses. It focuses on using deep learning on a convolutional neural network, with a model-free reinforcement learning method called Q-learning. For more information read: https://deepmind.google
PLM Consultant, Information Analyst, Engineering Services
3 个月Effective use of AI currently drills down to asking the program the right questions in the right order. Meaning that you as human need to understand your problem first and better so that AI can produce an answer solving your problem.How many young people learn to analyze deeply their problems and ask then the right questions? Youth today are informed by their own bubbles shaped by the magnificent 7.
Team Leader MBSE & Principle Consultant
3 个月Martin Eigner: I'm looking forward to our next discussions regarding AI in Engineering and our next publications together ??!
Mechanical design 4.0
3 个月After the amazingly rapid adoption of 3D CAD and PLM in the engineering industry after 2000, we are eagerly awaiting the next productivity boost in a now globalized business world. The exuberant marketing jokes at every engineering conference are a sign of this: PMI, MBD, MBSE, E-BOM, Digital Twin, and now AI... But where is the industrial implementation of these buzzwords today or tomorrow? Globalization is a game changer in mechanical engineering, as we are still mostly talking about self-manufactured components. Catalog parts - as popular as they are in electrical engineering - are rare in mechanical engineering. It is obvious that the link between a function-based geometric specification (formerly "technical drawing") based on GD&T or ISO GPS and the essential manufacturing specifications (verification, finished part, tool control) must be successful.