Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Where will artificial general intelligence come from?

docs.google.com

Post reply on HN
Print view View original

211–220 of 252 posts

Re: Where will artificial general intelligence come from?

#211

Post by gumby »

stephengillie wrote: post #73
eb3c90 wrote: post #53

Earlier quoted context omitted.

This might be a better link. It explains how we might get to IA. TL;DR It is a mix of machine learning with different parameters and inputs/outputs and language translation into programs with the economy acting as the force to guide this evolving set of programs. https://improvingautonomy.wordpress.com/2017/08/22/a-possibl...

Interesting idea. Instead of thinking of computing as an authoritarian schema, it could be a community schema, using a kanban or currency system to communicate resource needs between units. It's also similar to negotiating memory over commitment in virtualization. VMWare's driver on the VM "inflates a memory bubble" to communicate host memory constraints to all client VMs. This is often done when VMs have allocated 1…

Consider Minsky's Society of Mind.

Re: Where will artificial general intelligence come from?

#212

Post by hatsunearu »

mythrwy wrote: post #67
rdlecler1 wrote: post #29

Background: I did AI and Philosophy of Mind in undergrad, an MSc focused on ALife, then a PhD at Yale under a Macarthur Fellow who developed the theoretical framework for the 'evolution of Evolvability' where I worked on computational evolutionary biology. I can say we're not going to blindly brute force our way forward, but instead we'll need to reverse engineer nature's core algorithms to generate hard AI. Every ti…

Background: Some guy on the Internet with an unrelated Bachelors degree he didn't study that hard for and has picked up a little Python and Javascript somewhere along the way. I know of no biological forms that have evolved wheels. But wheels have turned out to be a hell of a lot faster than fins or legs. I see no reason "intelligence" has to follow the pathways or limitations of biology or neurology at all. Although…

If you consider rotary motion as wheels, there are some bacteria that have flagella that rotate to propel itself, complete with molecular bearings.

That said, biological evolution tends to get stuck on local maxima very easily (see convergent evolution of eyeballs), and wheels are kind of hard to evolve because of the difficulty of making large bearings biologically.

Not to mention wheels kind of suck unless you're on pavement (look at tanks, etc)

Re: Where will artificial general intelligence come from?

#213

Post by hatsunearu »

skummetmaelk wrote: post #91
jandrewrogers wrote: post #64

Earlier quoted context omitted.

The pessimism over simulating a human brain is two-fold. First, the human brain is built on a computational substrate that is completely and utterly unlike silicon. It is extremely inefficient to effect computation by simulating a computing model on silicon that is almost pathological for silicon to express. The abstract computational model of the human brain necessarily has an equivalent direct expression in computi…

Silicon is actually closer to the brain than you might think. Neurons transfer charge by diffusion, so do transistors operating in sub-threshold. The problem is that we almost exclusively use transistors operating above threshold because it is required for digital logic. Analog CMOS circuits can approach the energy efficiency of real neurons.

Uh oh, not this again. There is so much woo about using subthreshold FETs to simulate neurons when we don't even know how neurons work. I've seen the work of J. Hasler in school and she seemed to be fond about simulating a type of neuron (winner-take-all) that is hard to train with backpropagation (vanishing gradients just by inspection) and has limited grounds on physical simulation of neurons.

Do you have any other resources about serious attempts at using subthreshold FETs to simulate neurons?

Re: Where will artificial general intelligence come from?

#214

Post by p1esk »

hatsunearu wrote: post #213
skummetmaelk wrote: post #91

Earlier quoted context omitted.

Silicon is actually closer to the brain than you might think. Neurons transfer charge by diffusion, so do transistors operating in sub-threshold. The problem is that we almost exclusively use transistors operating above threshold because it is required for digital logic. Analog CMOS circuits can approach the energy efficiency of real neurons.

Uh oh, not this again. There is so much woo about using subthreshold FETs to simulate neurons when we don't even know how neurons work. I've seen the work of J. Hasler in school and she seemed to be fond about simulating a type of neuron (winner-take-all) that is hard to train with backpropagation (vanishing gradients just by inspection) and has limited grounds on physical simulation of neurons. Do you have any other…

I think what people (including Hassler) care about is not so much simulating neurons, but the energy efficiency. Once we figure out AGI algorithms, we will want to build them in hardware. This justifies the continuing research in subthreshold FETs, because they could allow very efficient computation (not necessarily biologically realistic).

Re: Where will artificial general intelligence come from?

#215

Post by eli_gottlieb »

return0 wrote: post #141
eli_gottlieb wrote: post #108

Earlier quoted context omitted.

Bayesian brain theories of perception. The neuroscientists and cognitive scientists are still waiting for the AI theorists to stop obsessing over that one paper with the cat's visual cortex and get off the deep learning train of sacrificing correctness for cheapness to compute.

this is not a case that happened

As a matter of fact, the mainstream AI community hasn't adopted probabilistic programming, but it does exist.

Re: Where will artificial general intelligence come from?

#216

Post by mannykannot »

CuriouslyC wrote: post #182
mannigfaltig wrote: post #181

Earlier quoted context omitted.

The main objective function in nature is very simple though: maximize the number of copies of your genes. Such an objective enforced in a resource-constrained multi-agent world (as suggested in the slides) could really lead to quite complex sub-objectives which may as well lead to general intelligence. For example, if each agent can process information and perform work, it follows that individuals that have better co…

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 CuriouslyC's point.

Re: Where will artificial general intelligence come from?

#219

Post by skummetmaelk »

hatsunearu wrote: post #213
skummetmaelk wrote: post #91

Earlier quoted context omitted.

Silicon is actually closer to the brain than you might think. Neurons transfer charge by diffusion, so do transistors operating in sub-threshold. The problem is that we almost exclusively use transistors operating above threshold because it is required for digital logic. Analog CMOS circuits can approach the energy efficiency of real neurons.

Uh oh, not this again. There is so much woo about using subthreshold FETs to simulate neurons when we don't even know how neurons work. I've seen the work of J. Hasler in school and she seemed to be fond about simulating a type of neuron (winner-take-all) that is hard to train with backpropagation (vanishing gradients just by inspection) and has limited grounds on physical simulation of neurons. Do you have any other…

Well that's kind of a non-starter attitude isn't it? We don't know how neurons work so we shouldn't try to figure it out by emulating their behaviour with electronics?

Analog neuromorphic approaches to not attempt to simulate neurons, they emulate them in silicon. Partly because of a belief that research in this area is required to produce ultra low power computational devices and partly to explore the real time dynamics of spiking neural networks.

There are very few research groups working on this, but you can look up Kwabena Boahen's group at Stanford. They do large scale real time emulation and are currently building a sub threshold neuron accelerator for the neuro engineering framework from Chris Eliasmith at Waterloo which is famously used to create SPAUN. There is also the Karlheinz Meier group at Heidelberg university which does wafer-scale networks of neurons in accelerated time. Giacomo Indiveri at ETH Zurich has silicon neurons with on-chip learning circuits that the others are missing.

Re: Where will artificial general intelligence come from?

#220

Post by LukeB42 »

malmsteen wrote: post #204
LukeB42 wrote: post #203

Earlier quoted context omitted.

The objective function for humans seems to be "Perfect the Human Soul" which can have conflicting optima in different contexts. Obviously no other mammal optimizes for it, either. See any slice of human culture or human activity ever for reference. Thanks for putting this slideshow together.

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.
Post reply on HN