https://www.youtube.com/watch?v=2YygKQh74Rg
An AI that will be in some kind of prison that can respond only with YES and NO.
Sounds good but if NOBODY is not analyzing the code, by analyzing code means that can be influenced maybe more than with a YES or NO.
YES and NO can make us to give it more right to output.
Why?
Scenario:
Will World war 3 start soon?
“Yes!”
But what period means that “soon” for the AI?
We can’t direct asking for year, because he can answer only yes or no.
So we start typing years.
He can say that is about 2017.
Can LIE because we uploaded to it all the Internet as how it was around a date!
If we don’t teach him to lie by filtering content very well, that will affect his responses, how right they are for some questions, or maybe all questions.
We can say how will start and input a number of variants, the AI will say no to all of them.
How can we get more if the AI can’t output more info?
Can take very long time or be impossible in that period of time.
So the governments will want more info and will remove some limitations.
We can’t ask will X win the election’s, because a YES is a probability.
We might ask “Has now X a probability over 60% to win the election’s?”
And he must answer with a Yes or No.
As you can see, there will be cases where Governments and not only the Governments, to solve a problem will want more than a Yes and a No, they will want a complete solutions, not to wait days, weeks or years to find them with the two options, Yes and No.
If people will not be monitored by the governments in 30-40 years we may watch arise an evil AI from someone personal PC that is running an local operating system and that is NOT connected to the internet.
The information’s might be searched on other system, copied and given to the AI.
Today CPU’s are not enough powerful, will cost too much to have the AI now, with the technology we know about, but in 30-40 years any user might have access to the processing power without needing billions.
Direct searching of the information’s without using human brain might light a “bulb” at Google that the searches and materials are too complex.
If the Google searches are NOT shared with NSA, will be a big problem in finding the creation before is too late.
So the human may not use AI suggestions when typing, because that will rise suspicions to something that is over watching.
Will be needed pauses in the searches too look like a human is reading and not an AI that might be too smart and evil.
