h a l f b a k e r ynon-lame halfbakery tagline
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
|
|
|
The suggestion is to measure the maturity of artificial
intelligence using a human child model, from birth to 25
years old, 25 years being the time that a human being is
fully developed mentally and psychologically.
At the point where AI would be deemed at the 25 year
mark, or "fully grown"
it would have to be impossible to
tell the difference between a synthoid or synthetic being
and a human in normal conversation.
Markers between infancy and maturity would be
delineated and I'd venture that it's about at the 3 year
old
toddler mark currently.
organoids
https://www.technol...06/brain-organoids/
The cells divide, take on the characteristics of, say, the cerebellum, cluster together in layers, and start to look like the discrete three-dimensional structures of a brain [beanangel, Oct 25 2018]
Skip forward to 5:40
https://www.youtube...watch?v=en5_JrcSTcU
Pretty good summary of captchas [doctorremulac3, Oct 26 2018]
as above, but starting at at 5:40
https://www.youtube...ture=youtu.be&t=340
ho hum [not_morrison_rm, Oct 28 2018]
Please log in.
If you're not logged in,
you can see what this page
looks like, but you will
not be able to add anything.
Annotation:
|
| |
I think I read that Paul Allen (company), is working on an AI
that can pass a science test for either 8 year olds or third
graders. |
|
| |
// currently in a stage of deveolpment similar to that of a 3 year old human // |
|
| |
... so time to hand over control to it from the petulant toddlers you delight in appointing to run your nation-states. |
|
| |
So, people do this already, mainly in order to inspire
investments. They make sentences like "we're at a 3-year
old level" or we're working on <passing this very specific test
that happens to be designed for humans of a certain age>". |
|
| |
At which point a human reader thinks, well, numbers are
adjacent, and if you just add enough really small steps
surely we'll be dealing with SkyNet by next Tuesday. And the
public relations people get away with it, too, as long as
their readers don't have any experience with real AI
systems and real 3-year-olds, at which point, well, if you're
dealing with a 3-year-old, you're probably too worn out to
argue with public relations departments. |
|
| |
Humans are really, really complicated, and we're nowhere
close to this very specific, very convoluted thing, and I don't
personally think it's a useful idea to even try, other than as
art. |
|
| |
It would be like measuring the quality of an Italian
restaurant by how closely it can reproduce a dish of
spaghetti in which the individual noodles have been
interwoven to resemble a particular, specific ball of yarn
which I have here in my bag and won't let you look at in
advance, but you're allowed to weigh the bag. Maybe shake
it a little. Only with a much, much bigger ball of yarn, and
it's got hundreds of different kinds of wool in it. |
|
| |
What sort of a sauce do you serve with that ? |
|
| |
the possibility of a utopian post-scarcity future is entirely
dependent on whether or not behaviours that today we
associate with consciousness and sentience can in fact be
automated without sentience. |
|
| |
If they can be, then this utopia is possible, but likely easily
hackable, so it's unlikely to survive. If not, then SkyNet is in
our future, as we are sure to try anyway. |
|
| |
I agree that to say a computer can pass a test an 8
year old can pass is a bit like saying a wind up toy
car can go the same speed as an 8 year old walking
therefore it's at the 8 year old level is a bit of a
useless comparison. |
|
| |
But the presumption is that by throwing out an
admittedly arbitrary milestone, the time when you
can have a conversation with an AI entity and a
human and not tell the difference COULD be
compared to a human's growth cycle. |
|
| |
Usefully? Eh, maybe. For an interesting point of
reference or even debate? Again, maybe. |
|
| |
I'm making the assumption here that there will be
synthetic human beings someday. You always see
them in science fiction but that doesn't mean we'll
definately go there. Maybe synthetic humans are
the "flying car" of the future. Yea, we can do it but
it's not really necessary so we don't. |
|
| |
// we can do it but it's not really necessary // |
|
| |
That's never stopped your species from doing many really, really stupid things in the past; besides, flying cars - or rather personal drones - are fairly close to commercialization, certainly Baked and WKTE. So that's probably not the best analogy. |
|
| |
Travel to your Moon is probably a better way of thinking of it; it's now nearly fifty of your Earth years since you first went there, and the technology still exists. Some aspects have improved significantly, it's more a lack of a socio-political motivation. But another reason is that although chemical reaction engines do allow exoatmospheric travel, the cost and risk are both very high, and a fair number of those involved in such endeavours spend quite a lot of time scribbling on napkins and muttering "There's GOT to be a better way ..." |
|
| |
Sometimes progress is incremental; sometimes it's a sudden lurch. The development of aviation was essentially incremental, with a visible linear descent from Cayley's and Lillienthal's gliders to your current "state of the art". Brunel would immediately recognize a contemporary bulk carrier as a much bigger version of his SS Great Britain, and a quick walkround of a modern diesel-electric train would allow him immediate understanding. |
|
| |
On the other hand, sometimes a single, often theoretical, breakthrough causes a very fast change. |
|
| |
What's a 'modlel' - something related to a 'covfefe'
perhaps? |
|
| |
You just gave me an idea for a game. “How quickly
can you turn this to a Trump conversation?” |
|
| |
The players have a deck of cards, each with
pictures of random objects. They all have pads of
paper. Once a card is drawn, they write something
insulting about Donald Trump associated with the
picture. These are scored two ways, fewest words
used and it must pass an overall plausiblity test. So
if you draw a card with a frying pan on it and
write: “Trump’s probably got one of these up his
ass.” You’d get zero points. But if you wrote “An
empty frying pan like we’ll all have when Trump
takes all the poor people's money and gives it to
the rich.” you’d get 50 points minus how many
words you used to make your Trump association. |
|
| |
// I'd venture that it's about at the 3 year old toddler mark
currently// I'm with [jutta] on this one. AI can answer
questions as well as a 3-year old? Fine. Say "Fuzzy jelly
elephants" to the AI and see if it laughs. |
|
| |
(of course, if it doesn't, you may be dealing with a 3-year-old
AI accountant) |
|
| |
//Say "Fuzzy jelly elephants" to the AI and see if it
laughs.// |
|
| |
It won't unless we tell it to, but then it's really not a good
analog of a human. |
|
| |
Like the flying car thing, is it really worth it to go down
that road? I'm sure someday we might be able to create a
totally mirror image synthetic human, which is great until
somebody says "So we can spend billions doing that or you
can get a partner and have sex." |
|
| |
I think the job that needs to be done is going to largely
dictate how AI evolves, but then again I predicted that
sex robots would never catch on so what do I know. |
|
| |
I'm leaving the question mark off of that sentence. I'm
making a statement, not asking a question. |
|
| |
//I think the job that needs to be done// Thing is, in most
situations, the job
that needs to be done needs most of a human. |
|
| |
At the moment, we think of AI for solving very specific, AI-
ready problems (like
facial recognition, or analysing pharmacological data). But
those problems are
insignificant in the big picture. |
|
| |
A truly useful AI would be able to parse "Go down the shop
and see if they have
any of those apple cakes." It would understand that it had
to go horizontally,
not down, and that when it got there it should actually buy
an apple cake
instead of just seeing it. It would know how to open and
close the door, avoid
obstacles en route, which shop you meant, who to ask
where the cake aisle
was, whether it had enough cash, how to get cash if it
hadn't, whether it
needed a bag for the cake, whether to put the cake
horizontally or vertically in
the bag to avoid the topping sliding off it, and about 13
trillion other minor
things. And that's only to tackle one out of a trillion
possible requests. |
|
| |
Even the very, very, very best AI we have at present is
much less than
1/1000th human in the areas where it would be really,
broadly useful. The AI
that we're building now tries to mimic only the thinnest and
most newly-
evolved rind of the human brain, and completely bypasses
the millions of
years-worth of basic ability to function as a human being.
It's like we're
struggling to mimic the "Fasten Seatbelt" sign of a plane,
but have not tackled
the significant issue of building the underlying plane. |
|
| |
Maybe we need to start from scratch with something
different than complicated mosaics of on/off switches. |
|
| |
Until we get an analog system on the head of a pin that
can compute flight trajectories, sense pheromones,
control angle and velocity of flight surfaces, landing
surface evaluation, fuel search and chemical analysis,
harvesting
and processing as well as reproduction tasks, threat
assessment, weapon deployment etc. we're just spending
a lot of time trying to make a honey bee out of a box of
hammers. Yea, we can do it, but maybe there's a better
basic approach we should be looking for. |
|
| |
I think AI should be absent sentience, unless it talk us into
it. |
|
| |
That said, I think I read there is a part of the brain about
the size of a quarter that when interrupted (TMS?) people
have no knowledge they exist while doing normal things.
Apparently this makes people into P-zombies for the
duration of the experiment. |
|
| |
There are clumps of nervous tissue that form "mini
brains", although during 2018 they lack a circulatory
system. They are called organoids. [link] |
|
| |
Perhaps a combination of organoids and electronics could
do improved AI, carefully omitting sentience. |
|
| |
//a combination of organoids and electronics could do
improved AI, carefully omitting sentience// |
|
| |
[beany], [8th]. [8th], [beany]. |
|
| |
See follow up to your link. |
|
| |
Should we really expect that the development of AI will follow
the same path as the development of a human child? |
|
| |
"Ontogeny recapitulates phylogeny", or so they say. |
|
| |
// expect that the development of AI will follow the same path as the development of a human child? // |
|
| |
Only if it is subjected over a period of years to the perverse and irrational whims of parents, siblings, other relatives and members of its peer group, plus social pressures to conform and vicious, repressive indoctrination gratuitously inflicted by psychotic geography teachers. |
|
| |
// improved AI, carefully omitting sentience // |
|
| |
Intelligence without sentience ? Aren't there enough Democratic voters already ? |
|
| |
I think you mean to reverse that, [8th]. |
|
| |
I've decided to go all Wittgenstein on you...I think we
have already made AI but we just don't recognise it. |
|
| |
We always judge AI intelligence as how close it is
being akin to human intelligence. Assuming there is a
intelligence in humans. |
|
| |
// We always judge AI intelligence as how close it is being akin to human intelligence. // |
|
| |
... otherwise known as "Mistake #1". |
|
| |
// Assuming there is a intelligence in humans. // |
|
| |
... otherwise known as "Mistake #2". |
|
| |
" ... at the great unveiling, the group leader feeds the computer its first question: "Is there a god?" "There is now," the computer replies. " |
|
| |
//I think we have already made AI but we just don't
recognise it.// |
|
| |
Oh, just like with obscenity, we'll recognize it. |
|
| |
It is likely the reader has had a blank mind(momentarily,
a kind of nonsentience), as well as a mind full of thoughts
(active sentience). Thinking of how some would like to
raise AI to sentience is there something the human mind
can be raised to that exceeds sentience? |
|
| |
Having tried LSD I think it is possible
the psychedelic experience might be a qualitative step up
from what humans usually call sentience. Aside from
that, there might be other things a step up from
sentience at humans. |
|
| |
One can support the idea of raising humans to a good
feeling psychedelic experience. A qualitative step up.
There is this possibility of thinking up other things beyond
sentience. This would provide a branch on the original
idea, a developmental gauge of AI, into a new separate
area qualitatively different than sentience. |
|
| |
That also suggests that genetically engineering other
animals to have a happy psychedelic consciousness could
advance well being. So instead of raising fish to
sentience and risking producing dissatisfied organisms, we
could raise them to beneficial psychedelic experience. |
|
| |
"New from the Sirius Cybernetics Corporation, robots with Genuine People Personalities !" |
|
| |
// a blank mind ... a kind of nonsentience // |
|
| |
Just like we said, Democratic voters. |
|
| |
You mean Democrat voters. Although "hive mind
rubber stampers" might be more accurate still. |
|
| |
That's pretty harsh coming from a hegemonic swarm of space
zombies invented by a creative but drug-addled liberal. |
|
| |
We're OK with harsh ... harsh is good. Also, we do not lurch round, arms outstretched, muttering "Brains ... brains ...". |
|
| |
As for Mr. Roddenberry, he was as you point out a creative type. They're permitted a certain latitude. |
|
| |
And longitude. But if they venture eastward out of Los
Angeles county bad things start to happen. |
|
| |