Sent from Yahoo Mail on Android
On Tue, Jan 16, 2018 at 9:19 PM, Brent Meeker<meeke...@verizon.net> wrote:
On 1/16/2018 8:55 PM, 'Chris de Morsella' via Everything List wrote:
--What is the craziest AI application you can think of?
A machine learned pet translator perhaps... they're actually working on that
app, Amazon amongst others. So, it seems the big players Google as well, are
running in that race... think of the potential market of pet owners forking
over their hard earned money to hear what the Google machine is telling them
their dog is telling them. I can imagine the marketing folks dreaming about
that market. As an aside also a commentary on how out of touch, we humans have
become from the world in which we exist. People already understand dog language
:)
Of course teaching the AI requires lots of training examples, so you will need
people to translate what their dog is saying to create the training examples.
Google will probably try to get people to do this online, similar to the way
they got visual identification training examples. But the really interesting
point is that not only do people understand dogs, it's also the case that dogs
understand people. So when Google's dog->human translate says, "Fido says the
mailman is here." will Fido be able to listen to that and say, "Rowf" ->
"That's right."?
Brent
------------------------------------------------------------------------------------We
might not want to always hear what our animals are saying about us behind our
backs... I see a potential law suit hehe :)
I believe, only half joking here... that a training set already exists somewhat
in the public domain. In the ever growing historical repository comprised of
all those pet videos uploaded online, and that dataset probably contains vast
numbers of clips of people trying to understand their pet vocalizations as well
as dogs (and to a lesser degree more aloof cats) listening intently to what
their people are saying. In fact I bet that a substantial body of raw video
feed exists even for more exotic human-other-species interactions... say
parrots... tegu lizards perhaps... cute little rodents.. gold fish...
tarantulas... you name it.A vast body of historical feed already exists.
The raw dataset would need to be cleaned, normalized, meta-described of course,
but heck there's machine learned systems that are even now getting pretty good
at parsing video stream data for some Darwinian evolved desired outcome, which
in this case would be to select out from the vast available but of spotty
value... those spots of value in the vast desert of cute pet video sameness.
Machine learned systems, becoming applied to evolving other machine learned
systems, is a self accelerating process.
Machine learning techniques can be applied to the entire pipeline of distinct
activities. Each granular step along the arc of information driven self
learning systems, from data sourcing, location etc., to actual retrieval (can
in practice be a huge headache, road block), normalization, formatting,
technical signal processing etc. On to activities such as meta-mining, symbolic
tagging & categorization, indexing etc. To the actual preparation of
experiments training and test sets.
Each of those granular activities, and many others as well not mentioned in
that off the cuff data pipeline can represent significant work, pose real
challenges. The whole long chain of activities that must occur even before an
experiment can begin has historically strangled the process somewhere along the
chain. It is slow hard work... it has historically been a hard nut to crack.
This is changing, and rapidly so, as each of these specialized activities,
which have in the past been potential bottlenecks becomes amenable to being
automatically ingested at near real time speeds by machine learned systems. For
example to tag and quantify correlating data, (an important activity in
preparing machine learned datasets to squeeze out as much signal as possible,
while minimizing the geometric explosion of over all uncertainty arising from
the introduced error from having too many dimensions that either duplicate (are
highly correlated), or do not contain any appreciable useful signal - but do
introduce potential bias, error etc.) Bucketization/classification of data is
another typixal example.
What used to be laborious and hence slow is increasingly being performed at
impressive rates. And by this, I intend the quite extensive array of
specialized activities as well as the web of pipelines between them (e.g. the
bus, as it is often called... and the queue/repository-cache based architecture
underpinning these things) All of it is now not only becoming automatically
processed, but the processing rate is becoming both more hi-if and also much
faster.
The cost of getting high quality, clean datasets out of raw data is coming down
as well. This is opening up the field of things that had been prohibitively
exp3nsive, but that can now be run through machine learned evolutionary
directed processes to produce the fittest outcomes... but hopefully not over
fit (inside joke)
The whole thing... which is the vast body of information being accumulated and
the growing body of evolved well fit outcomes, the desired goals, themselves
the outcome of generations of lineages trained over histories and then graded
against held off sets... this vast information planet earth information blob is
increasingly inspecting itself, and doing so in near real time.
And humans, incrwasingly are spectators to this process. We cannot even follow
what is going on in the hidden layers of the deep mind machines that Google
(and others) operate, we can know the input and output layers of neurons, but
what is happening in all those in-between layers, is increasingly hard to
reconstruct from logs even.
It seems to me that we are alive in the era when the evolved pattern
recognition systems, quorum peer node consensus networks and so forth that have
been unleashed upon both the voluminous incoming stream as well as the deep
stratum of reposited datastores of information... all becomes a self
accelerating, introspecting, self aware process.
As more stuff becomes accessible to machine learning techniques an increasing
body of highly fit systems will become plugged into an available body of
existing available added value... and often contribute to the adding of value
to other higher order sysrems, perhaps for, even as yet unforeseen potential
things to do.
In fact, increasingly machine learned systems are also travelling up the value
chain, putting together large numbers of factors and sub products, in near real
time and deriving subtle insights out of noisy raw data. It is a bottoms up
process that is really now taking off.
The train has left the station.... (for good or bad)Tempted to say... g-d help
us all... ;)
-Chris
What I think would be a wild application of machine learned systems is in
tackling the decoding/deciphering of lost ancient human languages and record
keeping systems (such as the Inca knotted strings for example). Wouldn't that
be cool... AI helping us humans learn about our own lost cultural heritage.
-Chris
On Tue, Jan 16, 2018 at 5:29 AM, K E N O <lucky@kenokeno.bingo> wrote:
Oh, no! As an media art student, I don’t believe in strict rules oft
usefulness (of course!). It was a rather suggestive or maybe even sarcastic
approach to get unusual thoughts from everything. Maybe I should rephrase my
question: What is the craziest AI application you can think of?
K E N O
Are you suggesting that fun is useless?
I can agree that the idea that fun has some use is not much funny, but that
does not make it false.
“Useful” is quite relative, also. Flies have no use of spider webs.
Bruno
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an
email to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to everything-list+unsubscr...@googlegroups.com.
To post to this group, send email to everything-list@googlegroups.com.
Visit this group at https://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
