Where will artificial general intelligence come from? • Hacker News Board

Re: Where will artificial general intelligence come from?

#241

Post by visarga » Tue, Sep 05, 2017, 4:08 AM UTC

mirimir wrote: post #186

CuriouslyC wrote: post #182

Earlier quoted context omitted.

I don't know that "maximize descendants of self" is necessarily right though. It seems like a better statement might be "maximize the development of complexity in the universe". Just as an example, lots of people choose to forgo having children to focus on contributing to the universe in other ways (myself included). This isn't just a self-centered drive for fame/wealth/etc either, as many people pursue their quests…

Well, for biological organisms, it's all about reproductive success. I mean, what exits today reflects what managed to reproduce, and how well. Overall, that has created lots of complexity. But that's just because there are so many niches and ways to be successful in them. What you say about people reflects cooperative behavior that drives reproductive success for shared gene complexes.

> for biological organisms, it's all about reproductive success

Another way of putting it - the source of meaning is life, or death (prolonging life, avoiding death as much as possible). Reproduction is just the start of life. From this game of life and death come reward signals that teach us how to act in the world (our values).

Re: Where will artificial general intelligence come from?

#242

Post by visarga » Tue, Sep 05, 2017, 4:11 AM UTC

TheOtherHobbes wrote: post #207

mirimir wrote: post #186

Earlier quoted context omitted.

Well, for biological organisms, it's all about reproductive success. I mean, what exits today reflects what managed to reproduce, and how well. Overall, that has created lots of complexity. But that's just because there are so many niches and ways to be successful in them. What you say about people reflects cooperative behavior that drives reproductive success for shared gene complexes.

It took nature four billion years to invent humans, who are actually - if we're honest - pretty terrible as an example of workable AGI. In fact what nature invented was a persistent colony organism with external memory. Wild solo humans are only a little smarter than wolves individually, but being able to share and externalise invention and learning created a massive advantage. Humans are successful because although…

Transfer learning is a thing (one NN learns from another or multiple NNs), also, large ontologies representing billions of facts.

Re: Where will artificial general intelligence come from?

#243

Post by visarga » Tue, Sep 05, 2017, 4:17 AM UTC

mannykannot wrote: post #216

CuriouslyC wrote: post #182

Earlier quoted context omitted.

I don't know that "maximize descendants of self" is necessarily right though. It seems like a better statement might be "maximize the development of complexity in the universe". Just as an example, lots of people choose to forgo having children to focus on contributing to the universe in other ways (myself included). This isn't just a self-centered drive for fame/wealth/etc either, as many people pursue their quests…

There is no guarantee that pursuing complexity as a goal in itself will lead to intelligence. The one thing going for mannigfaltig's proposal is that it has been known to work, though very inefficiently, and we don't have enough examples to estimate the yield. One might suggest that having the right definition of complexity would produce the desired result, but coming up with that definition takes us right back to Cu…

I give you a simple definition: maximize diversity (differentiation) and integration. They are opposites, to a degree - so there is a tradeoff and maximum at the middle in both diversity and integration. This idea comes from the Integrated Information Theory of Consciousness (Giulio Tononi, Christoph Koch).

In a neural net we do just that - maximize diversity by splitting the signal over many neurons, each with different weights, computing different things. Integration is maximized by the mixing together of signals from other neurons and training them together with a common loss function.

Even the internet as a medium requires diversity and integration to be successful. For example, net neutrality is related to diversity. Integration is related to national firewalls, copyright barriers, filter bubbling effect (where one sees only content from parts of the internet they agree with), walled gardens (like the app stores), and other things that cut the connection between people.

You can apply diversity and integration to other fields as well, for example, in politics/governance. We can compare a federal system (more diversity) with a centrally planned system (less diversity) and see the effects. With integration - we can compare free trade with regulated trade. The same principles apply to free speech - where diversity and integration are basically promised by the constitution.

Re: Where will artificial general intelligence come from?

#244

Post by visarga » Tue, Sep 05, 2017, 4:26 AM UTC

bad_alloc wrote: post #161

CuriouslyC wrote: post #151

Earlier quoted context omitted.

Plot twist: figuring out that objective function may prove as intractable as the original problem! We'll need an objective function writing objective function, and then it's turtles all the way down.

Unless it were possible for each objective function to converge to a next one that slightly extends the current one. However, what would the starting point be? Is this mathematically definable? What does the (im)possibility to define such a function tell us about AI?

> However, what would the starting point be?

The starting place is simulation. We use games (and other kinds of sims) to learn to act intelligently in a virtual environment. In such a place we can define many tasks and a learning curriculum.

Re: Where will artificial general intelligence come from?

#245

Post by visarga » Tue, Sep 05, 2017, 4:29 AM UTC

LukaEn wrote: post #217

akvadrako wrote: post #205

Earlier quoted context omitted.

What, oh wise one, is the goal?

Nothing. Evoultion came about randomly, not by intent, so it has no goal.

> Nothing. Evolution came about randomly, not by intent, so it has no goal.

Evolution created a goal from nothing, which is self-replication. It works on many levels - self replicating DNA, self replicating cells, self replicating ideas (memes), self replicating ecosystem, even the economy has become a self replicator.

Re: Where will artificial general intelligence come from?

#246

Post by visarga » Tue, Sep 05, 2017, 4:31 AM UTC

akvadrako wrote: post #223

LukaEn wrote: post #217

Earlier quoted context omitted.

Nothing. Evoultion came about randomly, not by intent, so it has no goal.

That's pretty meaningless - in that case the word goal doesn't have a meaning because everything is just as random as evolution, which isn't very random but whatever.

Evolution is not random. It requires self replication in order to transmit and evolve genes. Everything else is much more random.

Re: Where will artificial general intelligence come from?

#247

Post by visarga » Tue, Sep 05, 2017, 4:34 AM UTC

p1esk wrote: post #148

karpathy wrote: post #139

I gave this talk a while ago to a small group of attendees. It was not recorded (I saw some ask below). It's based on a document I wrote a while ago called "You suck at writing AI" (never published). The basic argument was that people are comically inadequate at writing complex code. You can't write the code to detect a cat in an image and the correct thing to do is to give up, write down an objective that measures t…

It's unfortunate that AI researchers pay so little attention to the brain (a single slide in your talk). We need more ML people evaluating ideas from neuroscience or from such initiatives as Numenta or Spaun.

Demis Hassabis (Deep Mind) published recently a manifesto about the importance of neuroscience for AI.

Neuroscience-Inspired Artificial Intelligence http://www.cell.com/neuron/abstract/S0896-6273(17)30509-3

Re: Where will artificial general intelligence come from?

#248

Post by visarga » Tue, Sep 05, 2017, 4:36 AM UTC

LukeB42 wrote: post #220

malmsteen wrote: post #204

Earlier quoted context omitted.

Nah it's just "survive and replicate". like every other species. Except that it's harder for us than other species, in particular, because a small human child is super fragile (more than like a giraffe baby) so you need a lot of care, and a lot of intelligence to do that care.

Nah I maintain "Survive and replicate" is a sub-problem in the larger "Perfect ones' Soul" objective function.

What is a "soul"? I can define "survive and replicate" for you, on the other hand.

This exemplifies the issue I have with current day philosophy. It's too blissfully unaware of the discoveries in AI. While they redefine consciousness the 1000th time, the AI researchers make "Reinforcement Learning Agents" that play Go, drive cars, paint, draw and can take a pizza order from you. Philosophers, get more concrete.

Re: Where will artificial general intelligence come from?

#249

Post by mannykannot » Tue, Sep 05, 2017, 12:06 PM UTC

visarga wrote: post #243

mannykannot wrote: post #216

Earlier quoted context omitted.

There is no guarantee that pursuing complexity as a goal in itself will lead to intelligence. The one thing going for mannigfaltig's proposal is that it has been known to work, though very inefficiently, and we don't have enough examples to estimate the yield. One might suggest that having the right definition of complexity would produce the desired result, but coming up with that definition takes us right back to Cu…

I give you a simple definition: maximize diversity (differentiation) and integration. They are opposites, to a degree - so there is a tradeoff and maximum at the middle in both diversity and integration. This idea comes from the Integrated Information Theory of Consciousness (Giulio Tononi, Christoph Koch). In a neural net we do just that - maximize diversity by splitting the signal over many neurons, each with diffe…

Well, maybe, but it looks very speculative to me. I think anything deserving the label 'definition' would have to be much more definite than that.

Re: Where will artificial general intelligence come from?

#250

Post by visarga » Tue, Sep 05, 2017, 12:14 PM UTC

rdlecler1 wrote: post #190

trott wrote: post #178

Earlier quoted context omitted.

> Actually convnets were inspired by Fukushima's Neocognitron, which was itself inspired by visual cortex. That doesn't contradict what I wrote. ConvNets require the synchronization of weights between neurons, which is not considered to be biologically plausible. Some aspects of the architecture (the receptive fields, in this case) may well be plausible, with the complete architecture still implausible.

You're focusing too much on the implementation details and assuming that because they're different that it's not equivalent. The secret sauce here is that network topology (and threshold rules), not the implementation details, are largely what determine the functional properties of that network. Show an electrical engineer the circuit diagram of a 4 bit adder and they'll know it's function immediately. Artificial Neu…

> Artificial Gene Regulatory Networks

I've been following AI for years but this is the first time I found such a concept. So basically a cell is like a small neural net with as many neurons as genes, each gene having (chemical) input and outputs signals. That means a cell's DNA is much more dynamic than I previously imagined. It's a self-replicating m.f. computer, that's what it is. We can only dream of similar accomplishments.

Previously I was aware of the huge workload carried out by DNA - for every protein in the body, DNA replicates the blueprints - an amazing amount of fine detailed work. It's not just sitting there waiting for reproduction. Seeing it not just as a factory, but also as a neural net is another level.