AI do it yourself. Are you ready to start?
Hello, friends!
I explain right away - the article is practical. But first, an allegory.
So, have you ever paid attention to the sad picture - the dead under the wheels of an animal car? Sorry little animals. For millions of years of evolution, they have perfectly adapted to the environment, but no one has given them time to adapt to the technosphere .
However, do you have a feeling that we gradually find ourselves in a very similar situation? We invented a bunch of all sorts of technologies, but with the advent of more and more new and more difficult to control them. And it is not entirely clear whether they are working for us, or forcing us to do something that we did not plan ... To be delighted or horrified? Happiness, eternal life and glory to expect, or death?
The fact that events develop with acceleration is noticed long ago. Thanks Ray to our Kurzweil for packing this observation in the so-called. "The law of exponential development":
At first, there was nowhere to hurry - the time from the society of hunter-gatherers to the agricultural industry was infinitely long, from it to the emergence of the first states already faster, much less it passed before the beginning of the industrial revolution ... Well, the rest has already happened (and is happening!) During the whole life then several generations of people ...
Yes, and what is there after 2045 ? Wait, there's nothing left. If you find it difficult to answer, Ray will tell you with pleasure, if you want - believe, if you want - no.
In any case, please do not forget about the unfortunate little animals! We do not want to repeat their fate, is not it?
Yes, it is he - the notorious human factor. Something is not taken into account, or not provided, or simply did not have time to click the toggle switch. Or they could not make a reliable conclusion on the basis of the available, critical information. Or even sadder - they were unable to appreciate its real value. As a result - the fall of the aircraft, explosions of the challengers, the global financial crisis, Chernobyl / Fukushima, 11.09.2001, etc. etc.
But the cars on the road that we cross with all the discretion available to us are not traveling at the highest speed yet. Take a look at the chart again, please.
It's like programming - you may have enough time to write and test a program, but after that all control over its work is limited only to what you have laid in it - you cannot control the whole process simply because your the brain is a little slower than the executable code, is it?
Yes, we have a brain, and it is capable of much, but it is slow, very slow. Previously, this was enough for us, but now it seems to be gone. There is so much information around that in most cases there can be no question of the possibility of making a really effective decision. Do you get it? Well, then either you are a genius, or you do not understand what I mean. Personally, I constantly get the feeling that my decisions are random, non-systemic . Yes, perhaps they are effective, but effective not in general , but only in the context of the strictly limited information and time (the latter especially) resources available to me at the moment .
And as if it is not enough that he (the brain) is slow, he is also not dimensionless! You teach, you teach all your life, and still you cannot learn even the minimum of what is necessary. And if a light suddenly appears at the end of the tunnel - get it - sign it - again, they will think of something like this, that without this you just can't do it ...
And where, pray tell, find for all this time ?!
And how great it would be if you could just shove a block with the necessary information there and get everything at once here and now, without spending years studying the given subject area!
')
The technological holy grail may be much closer than you think. It is likely that the road to where it is waiting for you (like all of humanity) begins side by side, literally a couple of clicks from here.
Yes, I, of course, the same about the great-and-terrible AI, who are all waiting-not-waiting.
Meanwhile, the prototype of the neocortex is already there. So far - about 60 thousand neurons. The human brain contains, according to various estimates, from 10 to 100 billion. Well, somewhere around half a trillion synaptic connections. But most importantly, it works precisely on the principle that the gray matter inside your skull.
Of course, everything has just begun, but the first real step has already been taken.
On Habré several times published materials dedicated to the company Numenta . Personally, I have been following their activities from the very beginning, from the moment I read here the first article about it .
I will not talk here for a long time in detail about the approach used, much has been written about this, including here . The basic algorithm is published, the document (including, and in Russian) is available here: CLA White Paper . (CLA - Cortical Learning Algorithm). If you hear about the company for the first time, then it’s best to begin your acquaintance with the book by J. Hawkins “ On Intelligence ” (there is a Russian translation, you can easily find on the web). In any case, this is an incredibly interesting book, the result of 25 years of research by the author of the principles on which our brain works :).
As an instant-alternative, Hawkins’s TED talks performance (with captions in Russian, about 20 minutes).
For a long time, the process was neither shaky nor shaky, but some time ago the situation changed. The company has now released its first commercial product, Grok , based on its NuPIC platform. However, the most interesting thing is that now their development has the status of Open Source and anyone can download the project with all the source codes, and directly participate in it. Github: https://github.com/numenta/nupic/wiki .
The principle of the system is described here: https://github.com/numenta/nupic/wiki/Online-Prediction-Framework .
The project is written in Python, deployed on Linux, but if you only have Windows, it is possible to do this after installing Virtual Box .
Now directly about the conditions. Firstly, in order to become a project participant, you must agree with the so-called. “ Contributor License ” (CL).
Secondly, a project participant may have 4 statuses (assigned in a hierarchical sequence).
Next is my own translation from English. (the original is here: https://github.com/numenta/nupic/wiki/Contributor-Model ):
Status of project participants:
As I have already noted, recently the company's activity has intensified, almost every week they publish new materials on their YouTube channel , and not only they - some developers post their own materials. There is also content dedicated directly to the NuPIC platform and the current Grok product.
I highly recommend watching the video directly with the speeches of J. Hawkins - this will give you a visual representation of the company's approach and the solution of this mega-task and the author’s vision / assessment of the concept of prospects for creating intelligent machines. Nument's own media collection is here: http://numenta.org/media.html .
Since readers doubt the “practicality” of the post and require examples, an explanation about the scope and example.
Any field of application where it is impossible to describe the upcoming development of events by a strictly deterministic algorithm.
The system itself creates a certain model and builds predictions on its basis. In order for it to be able to start doing this, it must first “learn”, that is, obtain a set of input data and create a number of patterns based on them.
Let's look at a “practical” example. You have decided to create a service that will allow people to find out - what is this song that they have heard somewhere, but do not know the name? That is, there is a melody, and nothing more, even words.
And this person comes to Google and google there this melody. That is, it just humming into the microphone what it can, and the system tries to determine which REAL melody corresponds to these attempts. As you understand, most people are fake, that is, it is impossible, most likely, in this case, to create some kind of algorithm that can identify a song by a falsely sung melody.
However, our system “completed a course of study” (suppose you listened to 100,000 melodies). Due to this she, having heard a certain sequence of notes can:
1. Find out which sequence exactly corresponds to the melodies.
2. Make guesses about which melodies the fake notes can match.
3. Choose the “best” candidates from these tunes
Or, extremely simplifying the picture and giving an analogy - there used to be such a transfer - “guess the melody” (I don’t know, maybe even now, just don’t watch TV for a long time).
And imagine that the one who guesses is the system.
How would you guess the melody? That's right, trying to remember which of the previously heard (ie, preserved in your memory ) looks like what you just heard. Then you create several models - and what kind of notes can there be further? If what you have heard before is exactly the same ONE of the melodies in your memory, then you will have only one model, if more - as many as you have melodies, which the previous sequence corresponds to. And with each new note, the number of these models will decrease until there is one left.
In general, this is how HTM (Hierarhical Temporal Memory) works.
I explain right away - the article is practical. But first, an allegory.
So, have you ever paid attention to the sad picture - the dead under the wheels of an animal car? Sorry little animals. For millions of years of evolution, they have perfectly adapted to the environment, but no one has given them time to adapt to the technosphere .
However, do you have a feeling that we gradually find ourselves in a very similar situation? We invented a bunch of all sorts of technologies, but with the advent of more and more new and more difficult to control them. And it is not entirely clear whether they are working for us, or forcing us to do something that we did not plan ... To be delighted or horrified? Happiness, eternal life and glory to expect, or death?
Tepericha - not that just now.
The fact that events develop with acceleration is noticed long ago. Thanks Ray to our Kurzweil for packing this observation in the so-called. "The law of exponential development":
At first, there was nowhere to hurry - the time from the society of hunter-gatherers to the agricultural industry was infinitely long, from it to the emergence of the first states already faster, much less it passed before the beginning of the industrial revolution ... Well, the rest has already happened (and is happening!) During the whole life then several generations of people ...
Yes, and what is there after 2045 ? Wait, there's nothing left. If you find it difficult to answer, Ray will tell you with pleasure, if you want - believe, if you want - no.
In any case, please do not forget about the unfortunate little animals! We do not want to repeat their fate, is not it?
Fast changes, slow brain
Yes, it is he - the notorious human factor. Something is not taken into account, or not provided, or simply did not have time to click the toggle switch. Or they could not make a reliable conclusion on the basis of the available, critical information. Or even sadder - they were unable to appreciate its real value. As a result - the fall of the aircraft, explosions of the challengers, the global financial crisis, Chernobyl / Fukushima, 11.09.2001, etc. etc.
But the cars on the road that we cross with all the discretion available to us are not traveling at the highest speed yet. Take a look at the chart again, please.
It's like programming - you may have enough time to write and test a program, but after that all control over its work is limited only to what you have laid in it - you cannot control the whole process simply because your the brain is a little slower than the executable code, is it?
Yes, we have a brain, and it is capable of much, but it is slow, very slow. Previously, this was enough for us, but now it seems to be gone. There is so much information around that in most cases there can be no question of the possibility of making a really effective decision. Do you get it? Well, then either you are a genius, or you do not understand what I mean. Personally, I constantly get the feeling that my decisions are random, non-systemic . Yes, perhaps they are effective, but effective not in general , but only in the context of the strictly limited information and time (the latter especially) resources available to me at the moment .
And as if it is not enough that he (the brain) is slow, he is also not dimensionless! You teach, you teach all your life, and still you cannot learn even the minimum of what is necessary. And if a light suddenly appears at the end of the tunnel - get it - sign it - again, they will think of something like this, that without this you just can't do it ...
And where, pray tell, find for all this time ?!
And how great it would be if you could just shove a block with the necessary information there and get everything at once here and now, without spending years studying the given subject area!
')
Good news
The technological holy grail may be much closer than you think. It is likely that the road to where it is waiting for you (like all of humanity) begins side by side, literally a couple of clicks from here.
Yes, I, of course, the same about the great-and-terrible AI, who are all waiting-not-waiting.
Meanwhile, the prototype of the neocortex is already there. So far - about 60 thousand neurons. The human brain contains, according to various estimates, from 10 to 100 billion. Well, somewhere around half a trillion synaptic connections. But most importantly, it works precisely on the principle that the gray matter inside your skull.
Of course, everything has just begun, but the first real step has already been taken.
Who is at the forefront?
On Habré several times published materials dedicated to the company Numenta . Personally, I have been following their activities from the very beginning, from the moment I read here the first article about it .
I will not talk here for a long time in detail about the approach used, much has been written about this, including here . The basic algorithm is published, the document (including, and in Russian) is available here: CLA White Paper . (CLA - Cortical Learning Algorithm). If you hear about the company for the first time, then it’s best to begin your acquaintance with the book by J. Hawkins “ On Intelligence ” (there is a Russian translation, you can easily find on the web). In any case, this is an incredibly interesting book, the result of 25 years of research by the author of the principles on which our brain works :).
As an instant-alternative, Hawkins’s TED talks performance (with captions in Russian, about 20 minutes).
For a long time, the process was neither shaky nor shaky, but some time ago the situation changed. The company has now released its first commercial product, Grok , based on its NuPIC platform. However, the most interesting thing is that now their development has the status of Open Source and anyone can download the project with all the source codes, and directly participate in it. Github: https://github.com/numenta/nupic/wiki .
The principle of the system is described here: https://github.com/numenta/nupic/wiki/Online-Prediction-Framework .
The project is written in Python, deployed on Linux, but if you only have Windows, it is possible to do this after installing Virtual Box .
Now directly about the conditions. Firstly, in order to become a project participant, you must agree with the so-called. “ Contributor License ” (CL).
Secondly, a project participant may have 4 statuses (assigned in a hierarchical sequence).
Next is my own translation from English. (the original is here: https://github.com/numenta/nupic/wiki/Contributor-Model ):
Status of project participants:
1. Users
The user is a community member interested in participating in the project. The user can be any person, for this there are no special requirements. The usual user input may include popularizing the project, informing the developers about the strengths and weaknesses of the product and / or their moral support :)
Users who continue to collaborate with other members of a project for a period of time often become more involved in its activities and may become contributors, as described in the next section.
Users who continue to collaborate with other members of a project for a period of time often become more involved in its activities and may become contributors, as described in the next section.
2. Contributors
Contributors are community members who make a specific contribution to the development of a project in one way or another, most often as creators of code and / or documentation. Anyone can become a contributor, and his contribution can be expressed in a variety of forms. We do not have any specific expectations about exactly what your contribution to the project will be, and we do not require any specific skills from you and do not set selection criteria for you.
Contributors participate in the project using our tracker and mailing list. They send their own changes through pull requests, which are considered for inclusion in the working version by current committers (see the next section). The mailing list or the #nupic IRC channel on Freenode are the most suitable places to get help in preparing for the start of your own contribution to the project.
All contributors must sign the appropriate project participant license. This is necessary insofar as this license confirms that everyone who contributes to the development of the project is either the author of the code or those who can represent its interests and speak on its behalf. This protects the interests of researchers using the NuPIC platform for their own purposes.
As the contributor gains experience and deepens his or her own understanding of the project, its weight within the community and its value for it will increase. Over time, they can be nominated for the next status by any of the current committers.
Contributors participate in the project using our tracker and mailing list. They send their own changes through pull requests, which are considered for inclusion in the working version by current committers (see the next section). The mailing list or the #nupic IRC channel on Freenode are the most suitable places to get help in preparing for the start of your own contribution to the project.
All contributors must sign the appropriate project participant license. This is necessary insofar as this license confirms that everyone who contributes to the development of the project is either the author of the code or those who can represent its interests and speak on its behalf. This protects the interests of researchers using the NuPIC platform for their own purposes.
As the contributor gains experience and deepens his or her own understanding of the project, its weight within the community and its value for it will increase. Over time, they can be nominated for the next status by any of the current committers.
3. Committers
Committers are community members who have made significant contributions to the project as developers and have sufficient experience working with other community members. They have the ability to place their code (push) in the project repository and must comply with the standards of the developer.
If a committer’s activity shows its value to the community and its commitment to the goals of the project, it can be considered as a potential reviewer (see the next section).
To become a committer, you need the desire and ability to participate in the project as a member of the team. As a rule, a potential committer must demonstrate a real understanding of the project, its goals and strategy. He also has to make a significant contribution to the project over time.
A project member can be submitted to the committer status by any of the current committers. From the moment of nomination, voting on his candidacy among the reviewers begins (see the next section). This vote is one of the few procedures of a closed nature. Each vote “for” updates their total value for a given applicant in the published NuPIC mail list. The nominee has the right to request clarification on any vote "against", regardless of the overall result of the vote. The answer should be provided by reviewers (see below) and is anonymous and constructive in essence.
It is important to realize that the status of a committer is a privilege, not a right that should be earned, but can be revoked by reviewers (see the next section) in extraordinary circumstances. However, as a rule, she remains with the project participant during the entire time when he takes part in it.
A committer who demonstrates a high level of contribution to the development of a project, especially in a strategic direction, can be nominated for the status of a reviewer described below.
If a committer’s activity shows its value to the community and its commitment to the goals of the project, it can be considered as a potential reviewer (see the next section).
To become a committer, you need the desire and ability to participate in the project as a member of the team. As a rule, a potential committer must demonstrate a real understanding of the project, its goals and strategy. He also has to make a significant contribution to the project over time.
A project member can be submitted to the committer status by any of the current committers. From the moment of nomination, voting on his candidacy among the reviewers begins (see the next section). This vote is one of the few procedures of a closed nature. Each vote “for” updates their total value for a given applicant in the published NuPIC mail list. The nominee has the right to request clarification on any vote "against", regardless of the overall result of the vote. The answer should be provided by reviewers (see below) and is anonymous and constructive in essence.
It is important to realize that the status of a committer is a privilege, not a right that should be earned, but can be revoked by reviewers (see the next section) in extraordinary circumstances. However, as a rule, she remains with the project participant during the entire time when he takes part in it.
A committer who demonstrates a high level of contribution to the development of a project, especially in a strategic direction, can be nominated for the status of a reviewer described below.
4. Reviewers
Reviewers are individuals defined as “project administrators” (hosted on GitHub). They are subject to additional responsibility, the level of which is higher than that of committers. This implies control over the smooth operation of the project. Reviewers are required to analyze the code generated by contributors, confirm changes to the current document and monitor intellectual property rights within the project.
The contribution of any reviewer can be explored by other reviewers, but this is not a necessary condition. Reviders do not have significant powers over other members of the community, although it is their votes that determine the possibility of appointing the status of a committee member to participants. They also have the right to make a final decision within the community on any of the issues, if a general consensus is not reached. In addition, the reviewers have access to the closed mail list and its archive. This sheet is intended for important communications, such as votes for / against the candidate candidate and consideration of legal issues not discussed publicly. This sheet is never used for project management, planning or technical discussions.
A committer is invited to receive the status of a reviewer by another, existing, reviewer. The result of the nomination is determined by discussion and voting among the reviewers.
The contribution of any reviewer can be explored by other reviewers, but this is not a necessary condition. Reviders do not have significant powers over other members of the community, although it is their votes that determine the possibility of appointing the status of a committee member to participants. They also have the right to make a final decision within the community on any of the issues, if a general consensus is not reached. In addition, the reviewers have access to the closed mail list and its archive. This sheet is intended for important communications, such as votes for / against the candidate candidate and consideration of legal issues not discussed publicly. This sheet is never used for project management, planning or technical discussions.
A committer is invited to receive the status of a reviewer by another, existing, reviewer. The result of the nomination is determined by discussion and voting among the reviewers.
As I have already noted, recently the company's activity has intensified, almost every week they publish new materials on their YouTube channel , and not only they - some developers post their own materials. There is also content dedicated directly to the NuPIC platform and the current Grok product.
I highly recommend watching the video directly with the speeches of J. Hawkins - this will give you a visual representation of the company's approach and the solution of this mega-task and the author’s vision / assessment of the concept of prospects for creating intelligent machines. Nument's own media collection is here: http://numenta.org/media.html .
UPD
Since readers doubt the “practicality” of the post and require examples, an explanation about the scope and example.
Any field of application where it is impossible to describe the upcoming development of events by a strictly deterministic algorithm.
The system itself creates a certain model and builds predictions on its basis. In order for it to be able to start doing this, it must first “learn”, that is, obtain a set of input data and create a number of patterns based on them.
Let's look at a “practical” example. You have decided to create a service that will allow people to find out - what is this song that they have heard somewhere, but do not know the name? That is, there is a melody, and nothing more, even words.
And this person comes to Google and google there this melody. That is, it just humming into the microphone what it can, and the system tries to determine which REAL melody corresponds to these attempts. As you understand, most people are fake, that is, it is impossible, most likely, in this case, to create some kind of algorithm that can identify a song by a falsely sung melody.
However, our system “completed a course of study” (suppose you listened to 100,000 melodies). Due to this she, having heard a certain sequence of notes can:
1. Find out which sequence exactly corresponds to the melodies.
2. Make guesses about which melodies the fake notes can match.
3. Choose the “best” candidates from these tunes
Or, extremely simplifying the picture and giving an analogy - there used to be such a transfer - “guess the melody” (I don’t know, maybe even now, just don’t watch TV for a long time).
And imagine that the one who guesses is the system.
How would you guess the melody? That's right, trying to remember which of the previously heard (ie, preserved in your memory ) looks like what you just heard. Then you create several models - and what kind of notes can there be further? If what you have heard before is exactly the same ONE of the melodies in your memory, then you will have only one model, if more - as many as you have melodies, which the previous sequence corresponds to. And with each new note, the number of these models will decrease until there is one left.
In general, this is how HTM (Hierarhical Temporal Memory) works.
Source: https://habr.com/ru/post/194182/
All Articles