Sejem inovativnih tehnologij

Tudi letos nas lahko srečate na sejmu IT tehnologij – Feel the Future. Lahko vam bomo predstavili najbolj avtomatizirana produkta za izdelavo šolskega urnika – iUrnik in za razporejanje delavcev – WoShi. Sestavljanje ja naloga umetne inteligence in to že 15 let, ko ostali o tem še sploh razmišljali niso.

Dobili boste lahko tudi informacije o umetni inteligenci – kaj si lahko pomagate z njo.

Preverite kaj zmoreta iUrnik – program za sestavljanje šolskega urnika in WoShi – razporejanje in planiranje dela delavcev.

Več o ukazu.

Pregledali smo konkurenco, kjer se stanje ni kaj dosti izboljšalo. Avtomatsko sestavljanje je omejeno na povsem osnovne pogoje, kar pomeni da rezultat brez človeške pomoči ni uporaben. Če ne moreš zahtevati, da se v ponedeljek ne dela, če si delal v nedeljo ali da delaj 4x ponoči med tednom in 3x ponoči za vikend, da naj delajo nekateri delavci skupaj z Janezom, drugi pa z Micko, …., potem ni od razporeda kaj dosti uporabnega.

Umetna inteligenca in šolski urnik

Če so zahteve staršev in otrok na eni ter ponudba izobraževalnih vsebin s strani šol na drugi strani čedalje raznolikejše, je tako najbrž dobro in prav. Toda kako jih najbolj smiselno razporediti in ob tem upoštevati, da na primer neke učilnice zaradi obnove nekaj časa ne bo mogoče uporabljati? Morda bosta morala delo učiteljice, ki odhaja v pokoj ali na porodniški dopust, začasno prevzeti dve drugi, zato bo treba vpeljati nekaj nujnih sprememb. Kako s čim manj spremembami prilagoditi urnik, če se izkaže potreba po novem oddelku, združitvi dveh ali vpeljavi novega izbirnega predmeta v sredo, 7. uro? Bi bilo mogoče popoldansko izmeno ukiniti delno ali v celoti? Vse takšne in podobne ter še številne druge spremembe, želje ter potrebe zelo zapletejo in časovno podaljšajo sestavljanje šolski urnikov po klasični poti. Zato so si v podjetju Algit (www.algit.si) pomagali z evolucijsko umetno inteligenco (angl. EAI — Evolutionary Artificial Intelligence), ki samostojno in bolje od vseh drugih znanih procesov rešuje kombinatorične probleme, ter zasnovali urniški program iUrnik (www.iurnik.si). Odslej se za sestavljanje urnika zadolženi osebi ni več treba dneve in dneve mučiti s kombiniranjem, ampak se po vnosu vseh podatkov v računalnik lahko udobno nasloni nazaj in opazuje sestavljanje urnika kakor v znanstvenofantastičnem filmu.

Tiste šole, ki tega učinkovitega pripomočka še ne uporabljajo, imajo zdaj lepo priložnost, da si ga omislijo in si z njim olajšajo izdelavo urnika za prihodnje šolsko leto.

If there is too much marketing hype in the above statement, then just ignore this feeling for now.
We have a log of an (unsupervised) machine learning session. There is a training data at the beginning, with numbers between 3 and 122. Primes are labeled with 0, nonprimes with 1. The third column are all 1’s and that simply means “use this row as a part of valid training data. This table of three columns is the entire world, for our Spector-Profunder (SP) to see. Otherwise SP is a “blank slate”, except for some basic mathematical operations.

Now, the SP begins to observe the world around. A small matrix 3 by 120 of numbers, In the first column there are integers from 3 to 122, in the second column there is either 0 or 1 (prime or nonprime), and in the last column there is always 1 in this case.

SP is like a person who woke up from deep comma and total amnesion. It has no knowledge about itself of course. Only the built in mathematical skills kicks in. It’s thinking language is a set of Turing complete computer commands. This language is quite understandable to humans, insofar as any programming language is.

In the log bellow, there are chronologically ordered algorithms, with which the SP models the world (number table) it sees. The first one is probably completely random and does not explain a thing. After some random mutations, there comes a generation slightly better. A random mutation and a nonrandom selection, Darwinian way. The log file is too long (in TB) to be readable, so just some more interesting “fossils” are shown.

After 20 thousand generations (which take a fraction of second), we only get a consistent algorithm which mostly misses the desired output (0 or 1 for prime or nonprime). It’s error is measured in “percent squared” is about 10^5. There may be no single correct prime guessed. SP might assign a nonsensical 2 to every number or something equally wrong. But we have a large “damage number” in the form of squares of differences between desired and actual values.

After some 24 million generations, we got the first algorithm which correctly predicts primality of ebery number from the list of tripples. Had we used a neural network, we would say it’s trained by now. But it’s overfited. Too long and too cumbersome.

After 100 million generations, we have quite an understandable code, which behaves quite well. The Eratosthenes sieve of a kind, was reinvented. Evolved once again.

****************** Generation#2,826,188,589 ****************

Algorithm’s error : 0
Algorithm’s cost : 0.0019871199949837033
Input register(s) : A
Output register(s): B

01 Nop
02 ID2 : B=B+1; C=B+1
03 Jmc : If (A%C==0) {24 Else 4}
04 Nop
05 Nop
06 Opc : If (B<A) {C=C+B}
07 Jmc : If (A%C==0) {20 Else 9}
08 Nop
09 Opc : If (B<A) {B=B+B}
10 Opc : If (B<A) {C=C+B}
11 Jmc : If (A%C==0) {20 Else 12}
12 Nop
13 Nop
14 Opc : If (B<A) {C=C+B}
15 Nop
16 Nop
17 Sq? : B=(A Is X^2)
18 Jmc : If (A%C==0) {21 Else 26}
19 Nop
20 Nop
21 Opc : If (B<A) {A=A Xor C}
22 Nop
23 Sgn : B=Sgn(A)
24 Nop
25 Nop
26 Nop

The log of events during machine learning

Pomembnejše dodelave:

• Vsaka izmena ima kategorijo, ki pove v katero obdobje spada – dopoldan, popoldan, ponoči
• Delavec je dobiv tri dodatna polja (zaradi lažjega vnosa mesečne obveznosti, dopustov in praznikov)
  • koliko je dnevna obveza dela – to se potem prenese v obvezo za praznik in dopust
  • kakšnega tipa je (od kod vzamemo mesečno obveznost dela)
  • enačba, s katero določimo mesečno obvezo dela (polovični delovni čas, 75%, fiksen čas, …)
• Praznik po novem omogoča vnos trajanja za vsakega delavca
• Ko se nekomu določi kdaj MORA delati, se določi tudi katero veščino. Dodatno je sedaj vpeljan isti sistem kot pri določanje prisotnosti (preko, vsebuje, natančno, znotraj)
• Vnos mesečnih obveznosti za eno leto
• Podatki o stanju delavčevih ur in dopustov

Več o tem.

Po dolgih letih končno IT sejem – Feel the Future. Z obiskom smo zadovoljni, obiskovalci so pri nas lahko dobili informacije iz prve roke, tudi o zmožnostih AI.

Predstavili smo program iUrnik in program za razporejanje delavcev WoShi.

Stephen Hawking says that we should augment ourselves as soon as possible, otherwise technological progress will overtake biology and then we’ll be in trouble! He says that we should use genetics based tech and whatever else preserves our biological essence. Either he sees something of intrinsic worth in biology, or he thinks that the death of biology means the death of us, which would be a shame.

Elon Musk says that in developing AI we are letting a genie out of the bottle whom, once free, we will be unable to contain. Interestingly, self-driving cars seem not to qualify as “true AI”, nor a significant step in that direction. It’s also interesting that he believes that if AI were developed in a more democratic manner it would be less dangerous. He even started a company centered around “open source, non-violent AI”.

Mark Zuckerberg, on the other hand, is more optimistic. He says that Musk is needlessly fear mongering, and that everything will be fine almost as if by default. It almost seems that he is of the opinion that with an advanced enough AI, the whole world will be like one big happy Facebook. Perhaps he will grow somewhat more pessimistic now that Facebook chatbots, which are supposed to negotiate with each other on behalf of users, have suddenly started speaking in tongues. Interestingly, a deal is usually made anyway. In the best interest of the users, of course.

Andrew Ng says that AI should concern us as much as a future overpopulation of Mars. He says that instead of efforts going towards technological advancement, some fear mongers and speculators are diverting funds towards counteracting nonexistent risks. As if he last name were part of the band name “Cool & The Gang”. Nomen est Omen.

Eliezer Yudkowsky is quite a bit more concerned. He says we have one chance to get it right, or it will end very badly for us. He also believes that convolutional neural networks, which began the ongoing AI spring, are a big deal. We live in dangerous times therefore, according to him.

Nick Bostrom is full of self-authored interesting stories. He likens us humans, who wish to create a digital superintelligence, to sparrows intending to domesticate an owl. All they need do is steal some owl eggs, hatch them and raise some young owls. Control over them would give the sparrows a good defense against cats and some help in the construction of nests for sparrow couples. Not to mention babysitting the chicks.

As you can see, different people have very different perspectives on AI. We mainly use it as a tool to automatize intellectually demanding processes. These optimizations, in turn, call for further recursive application of automation. Could it, some day, come to pass that we will no longer be able to control them? Not if we do it right.

So, in summation, we believe in automation everything, including automation.

Aegean, mythical king of Athens, consulting the Pythia

Before we get into more practical discussion about AI, let’s establish a bit more of the philosophical context.

The human need for communication is not limited to interactions with other people, but extends also to non-human interlocuteurs. Dogs, cats, and various other pets tend to make for relatively boring conversation companions. That’s why people went to Delphi, where gods would, however cryptically, communicate wisdom to them through the priestess. That’s why Moses climbed Mount Sinai to receive directions from a superior entity.

These examples only scratch the surface of the human desire to consult with someone or something smarter than ourselves. Interestingly, super-natural entities never had much to say about technicalities or science. However, at the dawn of the 19th century some sought to change this. Carl Friedrich Gauss decided to speak with Martians. He planned to arrange wheat fields into the shape of Pythagoras’s Theorem so that extraterrestrials might see them, and infer that we too are intelligent. To what end? So that Carl could discuss with them mathematics that perhaps only he and they could understand, and so that they could share new knowledge with him, which he would then (hopefully) relay to us.

In the 20th century, with the OZMA project, this notion saw a modernized approach. We thought now to harness our ability to send and receive data through radio waves in order to reach out to the universe beyond our solar system. However, faith in good natured and superhumanly wise aliens waned considerably when even the most advanced SETI failed to detect a signal. Chiefly due to the fact that biological evolution requires more permutations and longer periods than the universe can, on average, offer. More concisely: there is, as yet, nothing intelligent to be seen in the sky.

Therefore, there remains one last avenue of satisfying our desire to chat with something smarter and more interesting than ourselves – a superintelligent machine.

If for no other reason, satiating this hunger is why we need AI. It is very likely that, this time, we will succeed. Some of us can hardly wait to have long conversations with HAL. In our impatience we even write blog posts to future super intelligences. To SkyNet in the worse case scenario. Certainly we will get a reply before light or radio waves reach a distant galaxy and return back to Earth. Someday this dialog will take place in real time.