The Dawn of The New AI Era.
For the
last two years I have been sporadically writing on the issues of AI. The full picture comes with the reading of all major pieces
on the matter: e.g.
On The Definition of AI
Relax! The real AI is Not Coming Any Soon!
Who Will Train Our Artificial Puppies? (the second piece on AI training)
Applying AI to Study Learning And Teaching Practices
October 7,
1913 changed the automotive industry forever. This is the day when the assembly
line for making the Ford Model T car began to operate.Relax! The real AI is Not Coming Any Soon!
Who Will Train Our Artificial Puppies? (the second piece on AI training)
Applying AI to Study Learning And Teaching Practices
Before
that, the industry just did not exist. Automobiles were manufactured like art
pieces, by hand; automakers were researchers, and developers, and manufacturers
in one. Everyone new how an automobile should operate, what major parts and
functions are involved, and everyone could craft those parts in a shop. Ford
ended pre-industrial phase and gave the birth to the true industry. He did more
than just invented a good car. He invented the industrial manufacturing process
for building cars on a large scale. Before Ford, the most important engineers
were the ones who have been inventing and advancing cars and the related parts.
After Henry Ford, the new profession has been born – production engineers,
responsible for designing the whole process of car manufacturing, and for the
corresponded devices and procedures.
Another
big impact of Ford’s invention was the birth of a vast amount of a new-type of
people – drivers, then the development of the set of new regulations, and the
formation of another new professional field – driving schools.
A similar
transition is being seen in the field of AI.
The news
regularly tells us about new achievement in AI technologies. The list is long:
1. beating
the World Chess Champion
2. beating
the World Go Champion
3. beating
the World Jeopardy Champion
4.
teaching a car how to ride
5.
automatic translation
6. face
recognition
7.
handwriting recognition
8.
learning patterns in human behavior (social media, regular media, sport and
other fans)
9.
learning patterns in financial transactions
10.
winning over professional poker players
11.
analyzing heart beating patterns
naturally,
there is more.
What we
see in that list is the solid proof that today everyone (in the field, of
course) knows how AI should operate, what major structural blocks, parts,
devices, and functions are involved, and can design a “personal” AI tailored
for solving a specific problem. It is inevitable that very soon the focus will
gradually shift from technical aspects of manufacturing AI, which will become
as “simple” as manufacturing a car, to all aspects of its “post-production”,
starting from training AI to do the job it is supposed to do.
Let’s
switch for a moment from talking about AI to talking about HI (human
intelligence). Albert Einstein was one of the smartest people in the history of
the mankind. Obviously, he had one of the most powerful brains in the world.
But let’s use our imagination, what would happen if Albert Einstein – an infant,
was left in jungles and was raised by monkeys? Would he really become Mowgli, a
human among animals, or would he grew up as an animal? All currently known
facts tell us that most probably he would become a monkey in a human body.
Because humans become humans not as a mere result of birth, by via social
interactions. For a large number of years, the core of those interactions is
training, learning, schooling, educating. Take training out of the picture, and
even the most powerful brain will remain a baby in an adult body. Of course, a
person whose brain is very powerful may need less initial training than a
regular person, may start learning from books and begin his or her own
productive (creative, critical) thinking much sooner than others, but he or she
still needs at least some initial training in reading, writing, counting, etc.
The same
is true for any I (Intelligence), including AI.
Everyone
who dreams about making a true, strong, actual AI, should start from treating
it as it was a child, and ask questions like: what happens if AI has bad
parents/teachers; what does make a teacher/parent good or bad; how to assess
the quality of a teacher/parent (and many more which people have been asking
for many years about actual adults raising actual kids).
That is
why I expect to see very soon the growing demand for professionals who can
train AI in the most effective way (within a specific professional field). To
better understand how to train AI, AI designers will inevitably turn to study
people who train HI, a.k.a. teachers (including teachers who train teachers).
What they will find is that there are not many teachers who are both: good at
teaching and good at explaining what makes them to be good at teaching.
My
encounters with some of AI experts (e.g. https://teachologyforall.blogspot.com/2017/04/vc.html)
shows that so far they are not aware of the transition their field is entering.
The approach is “all we need is data, good data, a lot of data, and we will
learn everything what we need from that data”.
This
reminds me an old Russian tale called “Porridge from an Ax” (the Russian
version is from http://lukoshko.net/story/kasha-iz-topora.htm,
the translation is by the web-based Google Translate, with my editing – a lot,
by the way; this AI still needs more training).
- Please, let the traveler rest!
The old woman opened the door.
- Come on in, solder.
- And don’t you have something to eat, mam?
The old woman had plenty of food, but she was
very stingy, did not want to feed the soldier, and pretended to be poor.
- Oh, good man, I myself did not eat anything
today: nothing.
- Well, not to worry - the soldier says. Then
he noticed an ax under the bench.
- If there's nothing to eat, we can cook
porridge from an ax.
The hostess clasped her hands.
- How can you make porridge from an ax?
- Here's how, I’ll show you, just give me a
pot.
The old woman brought a pot, the soldier
washed the ax, put it into the pot, poured water and set it on the fire.
The old woman looks at the soldier and does not take her
eyes off.
The soldier took out a spoon, stirred the brew and tried
it.
- Well, how is it? - asked the old woman.
- It will be ready soon - the soldier replies, - it's a
pity, though, that there is no little salt in it.
- I have salt, take some.
The soldier took some salt, salted, tried the brew again.
- Good! If only it had a handful of grains in there!
The old woman began to fuss, ran out, and brought from
somewhere a bag of grains.
- Take it, fill it as you need.
He filled a brew with grain. Brewed, cooked, stirred,
tried.
The old woman is staring at the soldier, she cannot move
her eyes away from him.
- Mmm, the porridge tastes good! - The soldier licked his
lips - if there were a bit of
butter in it, it would be a real treat.
And butter was there, the old woman quickly found it.
- Now, old woman, just put some bread and get the spoon:
we'll eat porridge from
this ax!
They ate porridge.
- I really didn’t think that you could make such a good
porridge from just an ax - the
old woman says in awe.
Then the old woman asks:
- Solder! When are we going to eat the ax?
- You see, mam, it did not boil enough, still hard - said
the soldier - somewhere on
the road I'll cook it more and eat for my breakfast!
And at once he hid the ax in his knapsack, said good-bye
to the lady, and went on
walking to another village.
So the soldier and the porridge ate and the ax got!”
This tale
resembles the situation with data mining (including in AI development; aside
any possible moral implications :) ).
Data
mining specialists say to us (public, administrators, politicians, financiers):
“Give us the access to your data and we will solve all your problems”. But then
they say: “You know, if only we had a logistics manager here, to tell us what
he thinks; and a psychologist would not hurt, just as a on-a-side consultant,
and since we are here, let’s also call on … (fill the blank: and check https://www.ibm.com/watson/education/pearson)”.
And then –
“See what your data can do for you! (but also, what you can do to your data)”.
This
approach also demonstrates a common misconception, that collecting data is the
same as doing science.
Yes,
science is based on a collection of reliable data, but mining data does not
yet mean doing science – it means, though, enacting a scientific practice. To
make a transition from a scientific practice to science, data mining needs to
be molded using at least one specific model. The most famous example of
such transition is astronomy. The growing amount of data on the motion of
celestial bodies had led to the formation and the common acceptance of the
Ptolemaic system, which then has been eventually replaced by the Copernican system.
The
simplest (streamlined) model of the evolution of a scientific field includes
three stages/phases:
1. data
collection – a.k.a. measuring; which involves establishing measurable
parameters, standards (etalons), measuring devices, procedures, protocols.
2.
empirical research – searching for correlations.
3.
scientific research – a.k.a. research, establishing a paradigm, accepting the
set of fundamental models, using models to make successful predictions. The
mission of a science (any science) is to make successful predictions; until
then the field is a scientific field, but not yet a science (in the true
meaning of this word; no predictability = not yet a science).
Although,
in order to make all us to feel better about what we do, we could invent a specific
language to call everyone who is involved in any scientific practice “a
scientist”, and “a researcher”. For example, in accordance with the three
stages of the evolution of a scientific field, we could call ourselves:
1. a
researcher of the first level (involved in a research activity of the first
level); or a scientist of the first level (involved in a scientific practice of
the first level).
2. a
researcher of the second level (involved in a research activity of the second
level); or a scientist of the second level (involved in a scientific practice
of the second level).
3. a
researcher of the third level (involved in a research activity of the third
level); or a scientist of the third level (involved in a scientific practice of
the third level).
To make a
transition from a scientific practice to science, data mining needs to be
molded using at least one specific model. That has to be done by a
professional in the field, not by a data mining professional. Only a
professional in the field can develop the model to study (the scientific
hypothesis is - this model will work, more or less), state the criteria for the
model to be acceptable, modify the model based on the preliminary results, etc.
A model must include the list of measurable parameters, the description of
the possible values for those parameters, etc. and only a professional in the
field has the relevant knowledge (that is why the importance of people who can
design models will be growing faster than the importance of people who can
build AI).
That
finally brings us back to the topic at hands, i.e. the evolution of the AI
development.
Data
mining professionals are not experts in the field of training. That is why they will
need to forge a closer collaboration with such professionals. But that
collaboration has to be focused on solving a specific problem – training AI
how to train.
Using the language
developed in my field – education – I state that the most important goal in
the field of AI development is teaching AI how to
teach. Naturally, this task cannot be solved using any abstract
theories of teaching or learning. The goal is to teach AI how to teach a
specific subject, and while doing that to research the process of teaching AI
how to teach. Since I am a physicist by trade, and a teacher by birth, I am
confident that the next goal in the AI development is developing AI which
can teach physics as good as the best physics teacher does (but as
the first stage of the project, this AI should be able to win Physics Olympiads).
Anyone who
is interested in a collaboration on this project can reach me at
TeachOlogy@teachology.xyz.
Thank you,
Dr.
Valentin Voroshilov
To learn a
little bit more about my, please visit:
P.S. since
I am a normal person, my views on things evolve over time. This link guides to
an older opinion on the intersection between education and innovations: http://www.gomars.xyz/30us.html
P.P.S.
There is a common misunderstanding of what does AI mean. The literal reading is
"Artificial Intelligence". But we need to keep in mind, that
currently a true artificial intelligence - as an entity of this world, as
something material, presentable, usable - does not exist. What does exist is an
artificially created pattern recognition system(s) (a device - in a general
meaning of this word), which is (a) trainable, and (b) has elements of
self-training. This system has several specific realizations, which differ by
(a) the underlying structure; (b) specific area/domain of recognizable
patterns. And currently, none of those realizations can operate outside of
the domain it was trained to analyze. Which makes those realizations
non-intelligent (or animal-level "intelligent"), because the
mission and the core ability of an intelligence/intellect is creating
solutions to problems which have never been solved before ( (c)
Valentin Voroshilov)
BTW: A human brain is a
composition of networks of networks of networks with a huge number of active
elements (which makes it able to create solutions to problems which have never
been solve before); currently manufactured AI is a network with a dismal number
of active elements. Like an excavator is better at digging than a man, current
AI is better at certain task than a man. But it cannot think and feel and will
not learn it any time soon. That is why all publications about emotions,
ethics, morality and danger of AI represent a nice scientifically packed
version of science fiction (even an exponential rise will take decades to get
from thousands of active elements to hundreds of billions).
For the definition of AI at:
https://teachologyforall.blogspot.com/2017/12/aidef.html
https://teachologyforall.blogspot.com/2017/12/aidef.html
Appendix
On
Wednesday, 02/14/2018, I was listening live a Congressional hearing on AI (https://oversight.house.gov/hearing/game-changers-artificial-intelligence-part/).
Everyone who has a slightest interest in AI should do it, too. I would like to point at only three (of many) interesting moments.
Everyone who has a slightest interest in AI should do it, too. I would like to point at only three (of many) interesting moments.
1. Despite
one of the first the stated goals of the hearing (clarify what is AI), no one
of the four panelists offered a clear definition, except saying “AI is what we
see in the futuristic movies” (meaning, basically, devices acting like people).
I would like to have a discussion about my definition of AI (which as an
artificially manufacture system which can create solutions to problems the
system has never solved before).
2. When
asked when AI could exhibit reasoning abilities similar to human, all four
panelists offered numbers between 20 and 30 years from now. Which makes a
perfect sense to me. If they said "fifty" congressmen could start
thinking "well, if it so far ahead, what's the all fuss, we have more
pressing matters to finance?". But they just could not say "ten"
because they all knew (and all in the field know, and they knew they know) that
"ten years from now" is just not realistic, not believable (and lying
to the Congress is bad – at least according to movies).
3. When
asked about the areas where AI can bring significant advances, NONE (!) of the
participants named education. Clearly “big fish” in AI don’t have
education on the list of their priorities (didn’t pop up in their mind), or at
least as a potential funding generated field. That is despite the fact that the
training procedures they use to “teach” AI, such as “supervised learning” and
“reinforcement learning”, are just simplest teaching approaches – way before,
say, John Dewey’s Constructivism. The reason behind this fact is very simple –
current AI does NOT require any complicated teaching strategy, current AI is
not really smarter than a dog (can recognize a face, a voice, a command), well,
very fast thinking dog. And since it will not be requiring such a strategy for
at least twenty years, why even bother? This is one of the reasons that all my
attempts to reach out to AI professionals failed. And this one of the reasons
for me to start an open search for collaborators interested into
merging advance in AI with education.
Thank you for visiting,
Dr.
Valentin Voroshilov
Education
Advancement Professionals
To learn more about my professional experience:
Dear
Visitor, please, feel free to use the buttons below to share your feelings
(ANY!) about this post to your Twitter of Facebook followers.
No comments:
Post a Comment