secret of AI from a noob

Dec 07, 2016 65 Replies

I volunteer to "experience" Aida from "Agents of S.H.I.E.L.D."

Hi,

Yep, the input can be fed in "experience" from any source. There is so much data out there already that if a perfect AI could process it all, it would surprise us with the new variation it discovers, with the #2 goal to maximize variance.

If a near infinite processing ability AI existed with an empty memory it wouldn't make sense to put it in a situation where it has to suffer physical damage or even waste time running multiple simulations to gather more data until all the existing data in the world is processed by the AI, then it would be on it's own to find and create further variation, while keeping track of new incoming data.

cheers, Jamie

Hi,

I think you are confused cause this AI has no "ego", and is unaware, it is just a machine that functions to maximize the output variance based on the input data, which I called "experience".

If the input data is complex enough, then taking it all into account, and maximizing the output variance (requiring the output to be consistent with the input data) will give what appears to be intelligent behaviour from the AI.

cheers, Jamie

Stop there.

NT

How? No experience is context free, and feeding in the whole context is a lot more information than anybody collects or passes on.

This is a hypothesis. Anybody who has collected experimental data is well aware that you have to record lots more data than you initially realised was necessary to keep track of what's going on - usually including the phase of the moon.

To find something new, you'd need to have recorded even more ostensibly unrelated data.

How does it get the new incoming data? Particularly if it's not going to be in a situation where it might suffer physical damage.

Bill Sloman, Sydney

Intelligent behavior is goal-directed. If the AI has no ego, how can it have a goal?

Bill Sloman, Sydney

whoh, keep that think away from me, I don't want it using me to maximise its variance.

This email has not been checked by half-arsed antivirus software

You can't write software that has a goal? Where does your ego come in?

Hi,

"its" variance doesn't exist, the variance is simple the input that has been fed into the AI memory of past experience. If the AI memory included interaction with you, a complex entity, ya it may attempt to interact more I guess as long as you aren't like Bill Slowman.

cheers, Jamie

Maximizing variance has the emergent property of appearing intelligent, which is why I titled the thread "secret of AI from a noob" :D

Capable of thinking on your own? In contrast with Jamie, who can't think at all? And hasn't got a clue about snipping.

Bill Sloman, Sydney

Setting the goal that the software is directed at reaching?

I wonder where Jamie read that - maybe the fortunes in gluten-free organic fortune cookies are more pretentious than the ones in fortune cookies that also contain gluten.

Really - and here I was thinking that it was more of Jamie's pretentious drivel.

Bill Sloman, Sydney

bill.sloman:

think at all?

Sloman - an expert in sniping

MK

Is there anything he's not an expert of ?

Jamie

It has the emergent property of doing stupid dangerous fatal things. That is not intelligent.

NT

Hi,

Can you give an example of a situation where a stupid or dangerous/fatal thing has higher overall variance than an alternative action that isn't stupid or dangerous/fatal?

nope that's right you can't!

cheers, Jamie

So basic logic isn't your thing either. You need to master NI before trying AI.

NT

can you give a metric for "overall variance"

This email has not been checked by half-arsed antivirus software

Hi,

That's a good question, I think a metric for differential variance is one way to do it, ie given two sequences of events, determine which one has higher variance.

This is proportional to the length of the sequences, ie a shorter sequence would be lower overall variance.

Also it is proportional to the complexity of the sequence, ie if a sequence of events is harder to describe, or alternatively compress with a compression algorithm, then that sequence has higher variance than an easy to describe sequence.

something like:

overall variance = length of sequence * complexity of sequence

cheers, Jamie

What's NI?

You sure have a lot of judgements without answering many questions :)

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required