The intelligence of the human brain is said to be a strong factor leading to human survival. The human brain functions as an overseer for many functions the organism requires. Robots can employ artificial intelligence software, just as humans employ brains.

When it comes to the human brain, we are prone to make mistakes. However, artificial intelligence is sometimes presented to the public as perfect. Is artificial intelligence really perfect? Can AI also make mistakes?


Since the question was tagged with , I'll provide a philosophical perspective.

There is no definition of perfection. Just think about the silly example that is always brought up when we talk about autonomous cars: You are about to crash into a group of school kids and the AI's only chance to avoid them is by crashing into an old person. What should the AI do in order to react "perfectly"?

All actions that we take are based on our personal value system, so what is perfect to one person might be completely nonsense to another person. It might be possible to create an AI that always acts perfectly in terms of the value system you fitted it out with, but even in this case there are some caveats like:

  • reaction time: is the system fast enough to process all available information fast enough in order to take a fast action?

  • available information: not all information might be available with the system. Do you consider it acting imperfect, if the action of the system turns out to be wrong once you posses all information about the situation, even though some information will never be available to the system (e.g. Heisenberg's principle of uncertainty)


When it comes to the human brain, we are prone to make mistakes. However, artificial intelligence is sometimes presented to the public as perfect. Do artificial intelligent systems make mistakes?

Anything that exists outside of fiction that can be called AI and is not trivial is not perfect.


  • Alexa: creepy laugh - not trivial, but not perfect
  • Tic-Tac-Toe / Connect four: perfect algorithms exist, but trivial (you can create a game tree)

The problem is that you didn't bother to define "perfect". What does it mean to be perfect? Natural selection favors living beings that are energy efficient. Remembering everything might be considered perfect, but it is certainly not efficient.

We tend to make mistakes because some skills - especially abstract/mathematical ones or long-term decision making are not supported by natural selection.

Another big group of traits that are not supported by natural selection are health traits that are after the age of reproduction / helping children to grow up. Specifically cancer above the age of 40.


Yes it's error prone, just like us humans. But just like in chess it is just a whole lot better at dealing with it. AI does not contain all the possible knowledge in the universe out of no where without interaction with the world and such it should need to make assumptions and test those assumptions making it prone to error.

If you're asking about current techniques within reinforcement learning, image detection,etc. Those techniques are error prone, just likes us humans. During training of those algorithms you have a trade off between how good it will generalize on new data and how correct it can answer your questions. (it will "memorize" the questions and not actually learn in the latter, called overfitting)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.