Limits of AIs like ChatGPT: They can’t do that (yet)

by Pramith

ChatGPT and other artificial intelligences have been the talk of the town lately. They write texts, manipulate images and collect ideas. But the systems are not yet perfect. We show you what AIs can’t do yet.

This is what AIs can’t do: an overview of the limitations of the systems

Artificial intelligences can write convincing texts. But they also still have numerous weaknesses.

  • Comprehension: Artificial intelligences like Chat GPT that perform text genesis do not understand the content of their statements. They merely determine, based on a probability function, the next word or phoneme that is the most plausible continuation.
  • Correctness: Most artificial intelligences that are not limited to only one specific task work with probability functions. An artificial intelligence cannot test its statements logically or by semantic knowledge of the world, nor can it say for sure whether they are correct. For example, Chat GPT fails simple arithmetic tasks such as multiplying two numbers if the result is not in the input data.
  • Creativity: Current AIs only replicate and combine ideas and structures that were also contained in your input data. They cannot produce their own ideas, which are based on real considerations and which have not yet existed in this modified form.
  • Actuality: The knowledge of an artificial intelligence is limited to your input data. For example, Chat GPT does not yet know that Olaf Scholz is the Chancellor of Germany because the database used is from September 2021.
  • Universality: Artificial intelligences only work reliably within the task areas for which they have been trained. Artificial intelligences have so far relied on external feedback to improve. A strong AI is one that is indistinguishable from the human mind.
  • Structural design: Some experts believe artificial intelligences cannot get smarter by sheer increase in the amount of training data. Much more thought needs to be given to structuring the data and working on logical understanding to achieve human-like intelligence, they say.

Related Articles

Leave a Comment