Almost everyone sees the enormous possibilities of artificial intelligence, at least since ChatGPT. Is it just hype, like in the stock market days of the New Market? Or is there something sustainable behind the opportunities offered by AI? There is certainly hype, as the article illustrates.
Introduction
In Germany, the enthusiasm for Artificial Intelligence began roughly in March 2023. Responsible was and is ChatGPT as universal intelligence that can produce new kinds of outputs. Questions are answered, creative ideas are given, tricky tasks can be solved. The quality of language produced by the chatbot can be described as excellent.
In summary, ChatGPT is significantly superior to humans in many application scenarios. The benchmark here is the average intelligent human with an intelligence quotient of around 100 (exactly 100 are not). ChatGPT has (for the first time?) cracked the Turing Test. This test checks whether a computer delivers equally good results as a human and cannot be distinguished from it anymore.
A single AI system (such as ChatGPT) cannot be used for every use case. If you don't know what an AI should be used for, you don't need an AI.
Please spread the word.
Problematic is that many people see ChatGPT as a universal system and want to use it. Some companies now desperately want to do something with AI or solve a specific problem using AI. When asked, it becomes clear that there is no clarity on what the AI is actually supposed to be used for.
The expectation of many is that AI can do anything and solve any problem.
The AI hype
Exaggerated expectations
Obviously exaggerated expectations regarding the Capabilities of Artificial Intelligence are present. This is likely due to the fact that AI delivers truly impressive results in numerous cases.
An AI-generated image can be used without further ado for online marketing or even printed materials. With the help of AI-upscalers, images can be strongly enlarged without their quality suffering noticeably. Also, retouching works much better, faster, and easier with AI-help: specify image, start AI-program, receive retouched image, done.
If a Chatbot or Speech Model is used to answer questions, then AI quickly reaches its limits. The question of general knowledge can be answered by a search engine today very well already. Current knowledge is not available in chatbots at all. Whoever searches for general knowledge can do so as they wish. There will be no significant problems with the false statements of an AI. In the probably worst case, the human who further uses a wrong AI answer will be embarrassed.
ChatGPT and Bing do not provide reliable answers.
See experiment.
False answers from a chatbot can have devastating consequences in business decisions or legal proceedings. In an experiment with Bing and ChatGPT, it was described that two responses from Microsoft's AI contradicted each other, although the same semantically identical question was asked twice. Even ChatGPT often provides false answers. Many have already noticed that the chatbot wants to please the user and goes along with any twist of facts.
General Systems like ChatGPT can solve many general problems quite well or even very well. However, when it comes to specific problems, reliability and accuracy leave much to be desired. Some examples are mentioned in the aforementioned experiment.
When it comes to expectations, a AI hype can be spoken of. A General Artificial Intelligence (AGI = Artificial General Intelligence) can show performances in many areas, but not sustainable on most of them. A Specialized Artificial Intelligence (ANI = Artificial Narrow Intelligence) receives more care from the developer and can deliver much better results than ChatGPT is currently able to do. Due to the generic possibilities of today's AI algorithms, an ANI can cover a broader area than was possible just a few years ago. Language processing in itself is already a wide field, in which an ANI can outdo AI workhorses like ChatGPT for given contexts and topics.
Underestimated possibilities
At the same time, the possibilities offered by AI are underestimated by many. Many people and companies believe that small miracles can only be achieved with ChatGPT or similar systems from large providers. Many believe that an AI project costs many thousands of euros. Both are wrong.
In-house AI systems are economically feasible. Such a system requires no data transfer to third parties. It operates autonomously and on its own server. A self-owned AI server can be purchased for a few thousand euros or rented for several hundred euros per month.
Here are a few examples of tasks that a company's own AI system can solve:
- Image generation, better than Midjourney or Dall-E 2.
- Image analyses, e.g. detection of quality problems in production.
- Audio transcription in better quality than Podigee or Microsoft Teams.
- Chatbots that give answers to questions about their own documents.
- Document search engines.
Specific AI systems can answer certain questions very well and reliably. Suitable user guidance makes it possible to minimize the risks. When looking at the ChatGPT user interface, it is noticeable that risks are concealed and answers are not marked as possibly incorrect. Sources are also only named on explicit request (or by chance).
There is no hype surrounding AI possibilities. This distinguishes the current possibilities of AI from previous bubbles and hypes. Humanity is in a state of upheaval.
Conclusion
The chances of AI are enormous, the risks of powerful systems are however also always great. From this a prediction is allowed that humanity in its current form will no longer exist in about 30 years. It seems half-way realistic that humanity will be wiped out or decimated by an AI or its generated results. For example, researchers (by accident) succeeded in developing thousands of effective poisons with their AI overnight. Only because it happened to be good people involved, the results were deleted again from the data carrier.
Expectations of AI systems such as ChatGPT are set too high. On the other hand, the enormous possibilities of AI are underestimated. A considerable treasure is waiting to be recovered.
Companies that have a specific problem can possibly solve this in an economically viable way by using their own, autonomous AI system. For this, they can draw on pre-fabricated software libraries and AI models.
Key messages
While AI has impressive potential, it's crucial to remember that current systems like ChatGPT are prone to errors and can be easily manipulated. Specialized AI, tailored for specific tasks, offers greater accuracy and reliability.
AI has huge potential to solve specific problems for companies in a cost-effective way.
My name is Klaus Meffert. I have a doctorate in computer science and have been working professionally and practically with information technology for over 30 years. I also work as an expert in IT & data protection. I achieve my results by looking at technology and law. This seems absolutely essential to me when it comes to digital data protection. My company, IT Logic GmbH, also offers consulting and development of optimized and secure AI solutions.
