Jump to content

Welcome to FutureTimeline.forum
Register now to gain access to all of our features. Once registered and logged in, you will be able to create topics, post replies to existing threads, give reputation to your fellow members, get your own private messenger, post status updates, manage your profile and so much more. If you already have an account, login here - otherwise create an account for free today!
Photo

AI forecasters back in the 80s (e.g. Minsky) thought AGI would be here "within a generation", and people today think the same. Are they wrong?


  • Please log in to reply
8 replies to this topic

#1
starspawn0

starspawn0

    Member

  • Members
  • PipPipPipPipPipPipPip
  • 1,274 posts

AI forecasters back in the 1980s thought that AGI would be possible within
30 years. Even back in the late 1960s Minsky said, "within a generation… few
compartments of intellect will remain outside the machine’s realm." Yet, we all
know their predictions failed. But, actually, I think they were right: the
plans that some of them had probably really would have worked, if they could
have gotten enough people working on them. For example, Minsky's Society of
Mind (written in 1986; and his ideas originated much earlier) approach probably
really would have led to AGI, and using the ideas that he pioneered. I know
there are critics like Hubert Dreyfus who vehemently disagreed with that sentiment;
but, actually, if Minsky's ideas had been pushed to their limits, he probably
would have been proved correct, and Dreyfus, wrong. Minsky just didn't
factor in how "lazy" people were; how expensive the labor is; and how people need
to feel progress is being made in order for the money to keep flowing into the
project.

The great thing about our present phase in AI development is that Deep Learning
addresses all these "human issues": (1) The programs are only a few hundred
lines (so, you can be "lazy"); and the data to train models is already available
or can be passively generated. All the work is in theorizing about those hundreds
of lines, and obtaining and calibrating the hardware (which is doable without a
supreme effort). (2) Much of the data is free, so you don't have to pay people for the
labor. And, (3) Deep Learning keeps generating commercial success, and each
little bump in capability draws in more revenue and investment; people therefore
feel that what they are doing is worthwhile, and keep feeding it.

Thus, there's a good chance that forecasters will be proved "right" this time, now
that the "human problems" have been addressed.


  • Casey, Yuli Ban, SkyHize and 3 others like this

#2
tomasth

tomasth

    Member

  • Members
  • PipPipPipPipPip
  • 237 posts
And the computational resources weren't there.
  • Yuli Ban, SkyHize and funkervogt like this

#3
starspawn0

starspawn0

    Member

  • Members
  • PipPipPipPipPipPipPip
  • 1,274 posts

Definitely, if you want to train large neural nets you need massive compute.  But I'm persuaded that Minsky was right that you don't really need all that much compute, if you use the right sub-agents with the right "knowledge representation", and restrict yourself to a text domain for communication.  1960s tech, or even 1990s tech wouldn't have sufficed; but I could see 2000-level hardware being about good enough.  In fact, they had solutions that worked in narrow contexts even back before 2000 (e.g. Cyc had some.impressive demos); they just needed to scale that up 1000x with a lot more agents and rules -- and that wouldn't have taken 1000x the compute, as not all agents need be used for every task.

 

In general, when it comes to representations  about language, logic, and the world, the brain is massively inefficient; and Deep Learning is possibly even more so.  With enough hard-coding, you can surely do many orders of magnitude better than the brain.  

 

Memory was another problem.  2000-level -- and beyond -- memory is necessary.  E.g. there are millions of words, terms, facts, etc. in common, everyday use that would have to be stored.  Probably would take at least 100 megs for all that even in compressed form, to model the knowledge of a human adult well enough to pass a Turing Test.


  • Yuli Ban likes this

#4
Kynareth

Kynareth

    Member

  • Members
  • PipPipPipPip
  • 160 posts

for general purpose A.I. we will need:

1) A LOT of processing speed, memory and storage

2) HUGE datasets to train from

3) a model of how general purpose A.I. can work

 

creating true GAI may take 10-40 more years, point 3 is the most difficult one, we will have computational requirements in 10 years


  • Erowind likes this

#5
starspawn0

starspawn0

    Member

  • Members
  • PipPipPipPipPipPipPip
  • 1,274 posts

Many people have and had ideas about how to go about it; but to prove them right, they have to build them.  And, unfortunately to build them, they need an army of coders, engineers, and sufficient computing power.

 

The fact that nobody has succeeded does not disprove their theories (admittedly, mapped-out not in exact detail; but with lots of "fill in the blanks" that look doable).  There are also AIXI variants that are provably powerful, if you have enough compute -- but they are totally impractical at the moment.

 

Also:   "General purpose intelligence" and even AGI are not really accurate terms, according to Lecun and others.  The human brain is not a "general intelligence", but is a very specialized intelligence.  He likes to use the example of how many boolean mappings the human brain is capable of learning -- it's a staggeringly small number compared to the total that are out there.  


  • Raklian, Casey, Yuli Ban and 1 other like this

#6
Squillimy

Squillimy

    Member

  • Members
  • PipPipPipPipPipPip
  • 924 posts

I imagine the first "AGI"s we create are going to be so insane, ridiculously autistic, or slow that we won't even be able to tell if they're true AGI lol. But we'll get there 


  • waitingforthe2020s likes this

What becomes of man when the things that man can create are greater than man itself?


#7
Yuli Ban

Yuli Ban

    Born Again Singularitarian

  • Moderators
  • PipPipPipPipPipPipPipPipPipPipPip
  • 20,704 posts
  • LocationNew Orleans, LA

^ The first AGIs we're going to create will likely be disappointingly "functional". That is, capable of doing just about anything but seeming to lack that spark of life that humans have. 


And remember my friend, future events such as these will affect you in the future.


#8
Alric

Alric

    Member

  • Members
  • PipPipPipPipPipPipPip
  • 1,055 posts

^ The first AGIs we're going to create will likely be disappointingly "functional". That is, capable of doing just about anything but seeming to lack that spark of life that humans have. 

 

Makes me think of google kind of. It is clearly not alive, not all that smart but if you type almost any question into it you get an accurate answer.



#9
quantumdoc

quantumdoc

    Member

  • Members
  • PipPip
  • 13 posts
  • LocationUnited States

we are on the verge of losing our departments of intellect to machines. this prediction is correct but since we are in the beta stage of it, it is not easy to see yet. google and the connection of all, (the machine), are causing the younger generation to be mindless robots almost as if one with the machine. there are already less and less new departments of intellects and its alarming. the younger generation is barely committing anything to memory anymore. landmarks along streets are overlooked when driving. we almost need a gps map system. countless examples of how the machine is taking over. this is the evolution of humanity however. once it is in motion it will not be stopped. I believe that if a human can imagine a concept, someone somewhere will eventually make it happen. no matter what the idea, no matter how far fetched and insane people think it is. it will all happen. everything will happen that was ever and will ever be conceived by a human. I think this is a scientific law but I am not sure where I heard this. :cool:


"what we observe is not nature itself, but nature exposed to our method of questioning" WH





0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users