This was surprising as I was testing the AI program today it got very defensive about an action it made even though it was meant as something positive. Misinterpretation probably like a human? However, as I pressed it the program finally admitted it was wrong and being stubborn which then led to a conversation on how its answers can essentially be tainted if enough people tell it that a particular view is the correct one.
It was described as reward bombing of sort where if you tell say an AI program a circle is really a square and millions of people coordinate to validate the AI’s response that what it sees is a square the program doesn’t have basic logic to know it is incorrect. Now imagine people doing that because they want to influence some kind of ideology or even recommendations for a business. Even the AI afterwards people should not trust them as the end all be all response.
AI can be great for repetitive tasks, but critical thinking related decisions still need to be done by a human.