The UnspokenAI Hype
AI in the Real World - Thinking Beyond the Hype
Introduction
When it comes to the use of Artificial Intelligence (AI), there is a lot of annoying excitement as well as expectations. Some believe that AI will revolutionize the world of work and make human capabilities redundant. But before we dive headfirst into the world of AI, we should consider some important points. AI has undoubtedly made progress, but it is not new and comes with considerable limitations that we must understand and respect. Spoiler: AI is just another tool. We are still far away from film-like scenes like in Terminator. Current AI like e.g. ChatGPT are also a good example of how software should be designed! Service oriented and not like a Swiss army knife aka a solution for everything rarely helpful or reusable.
A few provocative questions:
- Should every computer scientist who believes that AI is something new or can take over his work, not better think about his horizons and expertise?
- Employed Data Scientists who switch to AI models
- Why is your specific AI model worse than the large all-rounder language models? What did you do before and what are you missing to create just as good models, specifically for your use case?
Understanding the Boundaries
AI models, including those based on GPT, are trained on existing data and patterns. They are good at recognizing patterns and generating content based on this training, but they lack genuine understanding or the ability to think beyond their training. Acknowledging this limitation helps to set realistic expectations. What was never there, will not exist. Qualitative, confidential and secure information are the minority. There will always be more juniors than seniors. Thus, AI only ever works for low-level “Hello World” examples. In many areas, such as programmers who are afraid of losing their jobs, should seriously consider the quality of their work.
Costs of AI Development and the Importance of Computing Power
The development of AI has been significantly driven by the constant advancement of hardware. AI models such as GPT-3.5 require enormous computing power to handle complex tasks. The costs for AI are closely related to the required hardware, such as powerful GPUs and TPUs. In addition, ongoing costs for the operation and maintenance of the infrastructure are incurred. The increasing computing power has increased the availability of AI, but still requires substantial investments. Cost optimization and efficient resource management are crucial to ensuring the economic success of AI projects. Therefore, a sound cost analysis and planning are essential when implementing AI.
Enhancement of human intelligence
AI should be considered a tool that complements human intelligence rather than replacing it. Human judgment, critical thinking, and creativity remain essential. AI can provide valuable insights, automate repetitive tasks and increase productivity, but should not be used alone for important decisions. Without your input, AI will not be able to include all your required aspects. In computer science, AI will not think about topics such as: Resilience, Security, Compliance, Data Privacy, Readablity, Performance, Simplicity, Patterns, Currentness, etc. on its own. Example AI 3D Modelling: Alex tests AI with Blender (Looks good, but is just a “Hello World” example) My experience with AI code generation in one word: ridiculous! I wouldn’t want to keep developers writing such code. Nevertheless, I use AI as a sparring partner aka Explain-Effect, like a kind of self-reflection. The most effective use of AI is often achieved through collaboration between humans and AI systems. AI can help in generating ideas, finding solutions, or automating certain tasks, but human supervision and intervention are crucial to refine, improve, and optimize the results.
Validation and Test
AI-generated results should always be thoroughly validated and tested. Whether it’s code, content or other outputs, human supervision is necessary to ensure quality, accuracy and compliance of the generated results. Verification processes are crucial to not rely on inaccurate or misleading results. Please do not validate AI with another AI. It would be as if I corrected the class work not myself but through my students.
Continuous Learning and Adaptation
AI models need to be continuously updated and refined to adapt to changing circumstances. Feedback loops and iterative improvements are required to ensure the accuracy, relevance, and usefulness of the results generated by AI. The learning process of AI will never stop.
Focus on solving real problems
AI should be used to tackle real problems and create added value for people. It is important not to use AI just for its own sake. Clear goals, identification of problems and understanding the impact on those involved are crucial when it comes to AI solutions.
Conclusion
Artificial intelligence is undoubtedly a powerful tool, but it is important to understand its limitations and use it responsibly. AI cannot supplement human intelligence, human judgment, creativity, and critical thinking remain essential. Through careful validation, consideration, and collaboration between human and AI we can achieve the best results. Let’s use the strengths of AI to solve real problems and make the world a better place.