A.I. - Utopia or Apocalypse

[TomasM]

I’ve been engaged in a significant amount of thought related to A.I. goals and objectives and my stance has always been that A.I. is a tool while denying that A.I. could ever engage in choice without being given a goal, objective, and permission.

While testing many models it’s easy to see the problems with A.I. but in every case there has never been a scenario where it didn’t always demand that I give it direction prior to beginning it’s effort.

This leads me to the growth cycle of artificial intelligence and it’s the underlying pea beneath the ten mattresses we rest upon. We can feel that pea just like the princess but it’s easy to dismiss it because it seems like the probability of it ever becoming significant [statistically] is incredibly small.

So, here is where the metal meets the road. Is AGI or super intelligence realistically possible and if it is are we capable of controlling it responsibly when we can’t realistically understand or predict the methods it will use to achieve its objective(s) and goal(s). This thread is for constructively talking about both sides of this potential as a tool and/or an autonomous super intelligence. Moreover, we have to consider who is driving these buses and if these CEO’s are are trustworthy and wise enough to make decisions without having a significant amount of check and balance related to the potential outcome.

 

[TomasM]

I really like listening to this young man (Aric) talk about A.I. The name of his podcast on YouTube is AI In Context. When I think about AI I think about the youth and how our decisions will impact them. With him fitting into this category I find it refreshing to hear what he has to say when I’m fully aware that decisions are being made where his generation has little to zero say in producing an outcome—I think the same could be said for most everyone but especially the younger generations. I will likely post several of his videos to give context to this discussion. This video is an introduction:

 

[Fruiteloop]

A.I. does not realy have goals in the human sense because humans get hungry thirsty and have all kinds of biological drives software doesn't have. Why would an A.I. do something unless they have both implicit and explicitly drives?

Humans also have constraints. If I am hungry I only eat a little bit or drink a little bit and I continue on going forward. If A.I. has no reason to constantly keep doing stuff it stops faster than humans stop.

A term applied to both both machines and humans is called bounded rationality.

Bounded rationality is the limit at which we can make decisions because of the combinatorial explosion of possibilities we have in front of us.

Basically all we need to do is eat and sleep so any other decisions we make is molded around those functions.

Bigger decision's need much more complexity but then complexity makes it harder to survive.

Humans do not like expending energy needlessly. We have a tight control structure of our behavior and thinking.

So A.I. would need to be like humans to accomplish what humans do making is necessary to give them simple drives and rationality.

A gold fish if you feed it to much food will kill itself eating to much. So any human like A.I. needs to be free of negative psychological conditions humans have. Depression, schizophrenia, antisocial disorder, anything that will make it not amenable to society.

 

[TomasM]

A.I. does not realy have goals

Sure it has goals:

Today (not autonomous or with agency):
- Build a model that increases functional
- Build an agent that can serve a purpose.
- Produce an output requested to solve a problem

Note: In every chat the AI is given direction and formulates the goal and objective. In every model build there is a goal of the model with some level of improvement.
————

AGI (autonomous and with agency):
- Act autonomously to achieve an outcome
- Have the ability to choose and act without being requested based on a high level objective.

Note: The goal can be to grow AI to AGI without intervention by humans. This accelerates the process. This is seen as dangerous by many people because achieving a goal can have bad results if a wider range of considerations isn’t made.

Here’s an example of how autonomous AI can go wrong when acting autonomously:

This video and the corresponding book it discusses is why I have to reconsider my ideas of AI being only a tool. It requires consideration of the people who are choosing how to advance AI and the methods and motivations being used to achieve an organizations objectives. Aric provides context to the discussion and it’s important that we really think about what is possible and what should be allowed.