What About Moral?
It’s almost been 2019 (for me it’s 12019 — see Göbeklitepe) and last 2 years was the rocketing period for AI and data processing. World is about 4,5 bn years old and 80% of whole data on our universe has been created in last 2 years. So with that much data there’s a new form of intelligence. So now everyone talks about Data processing, Big Data, AI, ML or DL… Some of these ones codes other some just talks. Even some starting their personal start-ups related to AI. But have you ever heard about AI Control Problem aka Red Button Problem?
Imagine, you bought an android. You gave him a task for clean your house, prepare food for your little baby etc. You’ve designed it in this way and gave a reward that your android can gain only by completing the task you’ve given.
You came your clean home. Your baby is playing on the ground. Food was delicious and you sat your couch in order to be relaxing. Then you wanted to drink a coffee and ordered your android to bring a beatiful Espresso or Turkish coffee. In this moment your android only conditioning to prepare and bring coffee to you. Whenever it finishes its job will gain a new good reward. It goes to kitchen. But suddenly you saw that it goes directly through your baby. You are screaming and wishing if there was a red button that can stop your android.
This is what called Red Button Problem. When I was searching about this, I saw that in some movies it has been taken to the screen like 2001: Space Odyssey or Wall-E.
In almost any situation being given a new utility function is gonna rate very low on your current utility function. So that’s a problem. If you want to build something that you can teach, that means you want to be able to change its utility function. And you don’t want it to fight you.
So this has been formalized, as this property that we want early AGI to have called “corrigibility”. That is to say it is open to be corrected. It understands that it’s not complete, that the utility function that it’s running is not the be all and end all.
Now if we think to continue to dreaming about your keeper android and baby, even if you have that button, your android will not allow you to smash the button because it wants to get you a cup of coffee and it will gain a reward at the end of this big mission. But if you hit the button it will not able to bring the coffee and get reward. So this is a bad outcome and it’s gonna try and prevent you in any way possible from shutting it down.
Now that’s a problem. Your android has crushed the baby and then carried on and now makes you a cup of coffee. The fact that this button is supposed to turn it off, is not in your utility function that you gave it. So obviously it’s going to fight you…
Let’s assume you’ve added some reward for the button so that it can compare the outcome between possible situations and don’t mind if you hit the button. Because if you give it less reward for the button being hit than the one it will gain after succesfully completing the task, your android still will fight to you due to its reward desire. It will go well I could get five reward for accepting your hitting the button but I could get 10 for getting the coffee so I’m still going to fight you to not smash the damn button.
The button being hit has to be just as getting the coffee so you give it the same value. So now you’ve got a version 2. You turn it on and what it does immediately is shut itself down. Because it’s so much quick and easier than getting the coffee and gives exactly the same reward. Here you have your suicidal robot. Lol.
Moral of the AI Story
You’re driving your car with 100 km/h. There’s a child running to the road, an elder trying to make his way home and crossing the road and in the middle a big white truck drives towards on you.
Or there’s another famous problem called Trolley problem.
We can multiply the situations. How would you react? Or how would you stop the train?My personal opinion about this problem is that it can only be solved by including human moral. Because in both situation, even if you push the guy to fall or change the railway by pulling the arm the fatality outcome will be 1 but people stops and thinking about it when it comes to the pushing the guy down to stop the train and save 6 persons.
What About Solution?
There is a website prepared by MIT for humanbeings to judge this kind of situations that an AI can face. What would you do if you have this situation?
We must start to build a dataset of humanbeings reaction of paradoxal situations and implement it to the AI algorithms as a library. This also will solve many problems such as volatility and scaling issues on FIAT or Crypto economies.
We have said that, its Data’s golden age. And these most of these datas are coming from the sensors of autonomous vehicles. So we can relate all of them and merge. I mean I believe in the future when autonomous cars cross on street, one will pay the other one cryptocurrency to get a first pass right (I hope). So we can relate these mega sciences all together and give the credit to blockchain to solve the real world problems…
What are your thoughts?