AI Learning To Make Other AI
Post Reply
 
Thread Rating:
  • 0 Votes - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
27-03-2017, 03:22 PM
RE: AI Learning To Make Other AI
That's not apple implanting their chips. It's making blind men see. It's not an evil AI taking over.
Find all posts by this user
Like Post Quote this message in a reply
[+] 1 user Likes abaris's post
27-03-2017, 05:21 PM (This post was last modified: 27-03-2017 05:25 PM by GirlyMan.)
RE: AI Learning To Make Other AI
Elon Musk is Batman.

Elon Musk Launches Neuralink to Connect Brains With Computers

[Image: musk_zpsbnacewz0.png]

#sigh
Find all posts by this user
Like Post Quote this message in a reply
27-03-2017, 06:20 PM
RE: AI Learning To Make Other AI
(17-03-2017 02:38 AM)Mathilda Wrote:  Kurzweil is a self publicizing professional fantasist.
I totally agree. He is just as deluded in his way as any religionist ... just as theists want so badly to live forever, so does Kurzweil. He understands at least that an invisible friend won't make it happen for him, but that's about all I can say for him.

In a world where we still can't evade the Blue Screen of Death in Windows and web browsers can't keep a few dozen tabs open without consuming 200% of 16G of RAM and eventually freezing up ... no. Just, no.

I do believe there will be a tipping point but I do not believe people who want technological immortality as to when it will occur. A minority of experts are thinking more like 100 years, rather than 10 or 20, and I tend to lean that way. The shorter scenarios basically depend on self-awareness and creativity and the like being an emergent property of fairly simple constructs related in ways that we will sort of stumble upon sooner than we think. That, and/or unexpected advances such as general-purpose quantum computing.

On the other hand ... even if I think it will take 100 years that is still pretty optimistic considering that undirected evolution took millions of years to achieve the same thing.

As to the potential for the emergence of a general purpose, self-aware, strong AI being catastrophic ... it is possible, but I tend to think it's more likely to be just a failure of imagination on the order of the concerns people used to have that the telegraph or radio or television or the telephone would have dystopian outcomes. I suppose the Kafkaesque scenario where people in the same room are texting each other in preference to just conversing, might be confirming to someone looking at today through a time machine from 100 years ago, that indeed, the outcomes were dystopian. But we have had 100 years to adjust. And no one is holding a gun to my head, suggesting that I have to email my wife who is sitting right now on the other end of the couch from me. I can have a conversation with her, but sometimes, if she's busy, and I run across something I know she'll be interested in, I might well email her from the other end of the couch, so she can look at it at her leisure. [shrug].
Find all posts by this user
Like Post Quote this message in a reply
[+] 2 users Like mordant's post
27-03-2017, 06:38 PM
RE: AI Learning To Make Other AI
Here's a New Yorker article discussing the Meat Puppets (de Grey,et al.) vs. RoboCops (Kurzweil, et al.) perspectives on augmented intelligence and life extension. - SILICON VALLEY’S QUEST TO LIVE FOREVER

#sigh
Find all posts by this user
Like Post Quote this message in a reply
28-03-2017, 09:30 AM (This post was last modified: 28-03-2017 09:56 AM by GirlyMan.)
RE: AI Learning To Make Other AI
(27-03-2017 06:20 PM)mordant Wrote:  I do believe there will be a tipping point but I do not believe people who want technological immortality as to when it will occur. A minority of experts are thinking more like 100 years, rather than 10 or 20, and I tend to lean that way.

Exponential growth is hard for us to intuit and grok. Everything looks slow and gradual and then bang, where the fuck that inflection point come from? Here's a log-linear plot of human gene sequencing.
[Image: genome_zpsylybek9u.jpg]

And here's the log-linear cost curve. Who here thought 10 years ago they would be able to get their individual genome sequenced for under $100 ($200 if you want to know how you'd likely die naturally and take remedial action)?
[Image: genecost_zpsei6dhvmx.png]

#sigh
Find all posts by this user
Like Post Quote this message in a reply
28-03-2017, 05:48 PM
RE: AI Learning To Make Other AI
(28-03-2017 09:30 AM)GirlyMan Wrote:  Exponential growth is hard for us to intuit and grok. Everything looks slow and gradual and then bang, where the fuck that inflection point come from?
I both intuit and grok the concept. And that is why I am unwilling to say it's impossible that we'll see general-purpose "hard" AI in the next few years; it is exactly why domain experts are predicting anywhere from 10 to 100 years; "we see through a glass darkly".

I just don't think it's likely.

We have made striking progress in some respects (for example, there's an open-source library written in Python that effectively simulates the human prefrontal cortex; we have largely figured out the basics of how pattern-matching works in the brain). But putting the various pieces together to achieve that kind of AI is a much harder problem.

One of the other things that argues against exponential growth is that there is an upper limit to how fast society and the marketplace and individuals can absorb and adjust to technological change. Moore's Law, to the extent it has held up, is another exponential growth phenomenon that has been largely irrelevant because computers have been Fast Enough for quite awhile now. How much power do we need to bring to bear on Word documents?

I suppose that hard AI is something that could use that extra capacity. So is virtual reality and augmented reality. But we are still struggling to make those things work well and to resolve certain blocking issues (such as, when it comes to VR, most people don't want to wear a heavy breadbox on their face for any length of time for some reason) and to get everyone to agree on standards and best practices. How much more of a train wreck is hard AI apt to be?

I don't have much more time on this planet and I frankly HOPE there are huge breakthroughs before I'm gone because I'd enjoy seeing them happen. I'm not a luddite or nay-sayer. I just am not feeling it at this point.
Find all posts by this user
Like Post Quote this message in a reply
[+] 2 users Like mordant's post
28-03-2017, 06:42 PM
RE: AI Learning To Make Other AI
(28-03-2017 05:48 PM)mordant Wrote:  One of the other things that argues against exponential growth is that there is an upper limit to how fast society and the marketplace and individuals can absorb and adjust to technological change.

This is a good point. The adoption rate always lags but it too grows exponentially. Cellphones are ubiquitous and I remember only 15-20 years ago only the big shots had them. And Batmusk don't look like he lets such petty matters get in his way.

What I think is the game changer is that now our hardware is sufficiently powerful that deep learning is now possible with multilayered perceptrons. Problem before was the training time. But what used to take 100s of years of training just a decade ago now can be done in weeks. One group has made an ANN which learned a 160 billion parameter function. If we stipulate that much of human behavior can be modeled in terms of mathematical functions then the Universal Approximation Theory guarantees that the vast majority of human physical behavior can be reproduced artificially. The question is whether the brain can be modeled mathematically and I see no intrinsic reason it can't.

#sigh
Find all posts by this user
Like Post Quote this message in a reply
[+] 1 user Likes GirlyMan's post
28-03-2017, 08:09 PM
RE: AI Learning To Make Other AI
(28-03-2017 06:42 PM)GirlyMan Wrote:  The adoption rate always lags but it too grows exponentially.

For some things but not for others. I see the Trump presidency as a backlash against changes coming to rapidly for people to adapt. How will they feel when heavy duty automation kicks in?

Undecided
Find all posts by this user
Like Post Quote this message in a reply
28-03-2017, 08:18 PM
RE: AI Learning To Make Other AI
(28-03-2017 08:09 PM)Jay Vogelsong Wrote:  
(28-03-2017 06:42 PM)GirlyMan Wrote:  The adoption rate always lags but it too grows exponentially.

For some things but not for others. I see the Trump presidency as a backlash against changes coming to rapidly for people to adapt. How will they feel when heavy duty automation kicks in?

Undecided

Only thing Trump and Congress can control (apart from imposing regulations which they don't seem keen to do) is federal funding for science. In many of the less sexy and more obscure fields federal funding means survival or not, not so in tech. Federal investment in basic research is something the tech giants encourage so they can leverage it but they don't rely on it and direct federal investment is something they neither want nor need. Trump can't stop the tech train. Nobody can. It's advancing on its own.

#sigh
Find all posts by this user
Like Post Quote this message in a reply
[+] 1 user Likes GirlyMan's post
28-03-2017, 08:21 PM
RE: AI Learning To Make Other AI
(28-03-2017 08:18 PM)GirlyMan Wrote:  Trump can't stop the tech train. Nobody can. It's advancing on its own.

Then a lot of people are going to get run over by the tech train.

Sad
Find all posts by this user
Like Post Quote this message in a reply
Post Reply
Forum Jump: