The great step of artificial intelligence: use what you learned in one task for the next

Shortly after we told you how robots learn by reinforcement, imitation and deep learning, there was a expected news in the world of artificial intelligence: DeepMind, the AI of Google, I could already use what I learned from one task and take it to another.

Until now, AIs have broken records by squeezing to the maximum a series of rules or norms in a specific field, but it is impossible to bring the gains from one application to another.DeepMind can do so thanks to a mechanism called consolidation elastic weight .AI now have memory , or the beginning of one.

AI and learning

Without repeating ourselves in what we have already said about reinforcement, imitation or profound learning, we can summarize that until now artificial intelligences have learned as humans do in the early stages of life .And they have done it in watertight compartments without any permeability.

To understand it, let's think of a baby crawling on the floor and whose goal is to reach a bottle.Being a human, it already has a brain as a series that allows deep learning , of course.But even if you have reached the coveted prize ( reinforcement learning ) ten times before, and even if you have optimized your crawling algorithm by looking at your parents ( learning by imitation ), you will not know how to open the bottle if you have never tried.For the baby, the bottle will be a new problem with new rules.

That was the state of the art of artificial intelligence until March 2017: he knows how to crawl but that doesn't help him open the bottle.What he learned by solving each problem didn't help them for anything.Next problem: With each new challenge you had to return to the starting position, fail, iterate, fail again, take a different path, fail, try again...

The great step of artificial intelligence: use what you learned in one task for the next

But now artificial intelligence has one more human tool, the elastic weight consolidation (EWC), which allows you to go further in your learning.Or, at least, reach that away in less time and using fewer resources.What is already an achievement in itself.

Elastic weight consolidation, what is that?

Let's do a quick review through the great milestones of artificial intelligence in the discipline "beating human beings." No, we do not talk about Cylons seeing how humans are inside , but of board games that are now AI domain:

1990: AI Chinook is second to the ladies against Marion Tinsley.

1995: after his death, Chinook comes to mind still unbeatable.

1997: the AI DeepBlue wins Kasparov to chess, and without the option of excuses.Logistello defeats Murakami 6 to 0 in the same year.

2016: AI AlphaGo defeats the game go to Lee Sedol.

2017: AI Libratus wins poker without limit Texas Hold'em.

Yes, the news could be that «artificial intelligences are leaving us without specialized jobs and without board games», but there is something else.

Until now, artificial intelligences have had to be designed expressly to win each of these games, however, we can say with certain certainty that Go Sedol himself could become a good chess player.of AI, it is not necessary for your parents to generate another Lee for it.

That that humans had but not AI (among other elements) is the elastic weight consolidator (EWC): a mechanism that rescues fragments of code used in applications similar to applications present .

In the case of games, the EWC is that part of the human software that sneaks us: "Hey! Have you noticed that the chess pawn moves in a similar way to the ladies?" It is the same type of algorithm that whispers to babies: «Hey, you, little one, the legs that you have used to sweep the floor help you stand up, and they will work more or less as badly».

The great step of artificial intelligence: use what you learned in one task for the next

As neurologist David Eagleman says, "the human brain is a multipurpose machine." Although AI processors are technically also (the same is solved by one operation than another), their starting software is not.

The EWC of artificial intelligence and the EWC of humans

The elastic weight consolidator is the mechanism that activates the stored memory, even time ago, and uses that prior knowledge to build more knowledge in solving a problem.But even humans have the EWC parameter set very under (except in exceptional cases).

Just as a human uses 10% of his brain (if we used 100% we would not be LUCY, but an epileptic), he usually cannot access 100% of his memory and what he has learned., nobody remembers everything.Reason why many times we find the optimal answer hidden in our memory a long time after having patched a problem with a second solution.

"Ah, it's true, it was like that !", we thought.And we lied to ourselves: "I remember that next time."

Imagine the access to the EWC as a percentage from 0 to 100.A very low elastic weight consolidator, 0.01%, does that the AI or the human does not relate a previous learning with a new problem.A very high one, close to 99.99%, causes it to block or even end up bleeding from the nose, because an elevated EWC would make any process heavy .

The great step of artificial intelligence: use what you learned in one task for the next

"What a clever AI", we could say by looking at the processor power consumption data.Without realizing that you spend all your computing power checking the relationships between the new problem and all the previous problems instead of solving the problem.What is known as Paralysis by analysis , applied to artificial intelligence.

So, are AI more clever now?

Yes and no, at the same time, although rather maybe .It is unlikely that we will reach a consensus, especially considering that it is difficult for us to define words widely used today as intelligence .

If solving the same problem faster with less power is a sign of intelligence, then yes, artificial intelligences are probably at the root of DeepMind more lists .

But, except for television programs, we do not consider a human with quick answers more intelligent (even if they are correct).We say that he has a good memory and, if he is solving a problem, we simply say that it is resolutive.We relate intelligence to speed, and we do not even say that it is wiser.

The elastic weight consolidator mechanism is a system that helps to reach a partly known solution .So at most we could say that AI is more wasted realizing that part of the problem was already solved.

For many, that AI becomes intelligent is a technophobia attack.But according to experts these fears are unfounded.First, because our artificial intelligences are in honeycombs and, second, because the most intelligent thing an AI can do is to pass from humans (and for now they continue to play our arcade when we ask for it).Very intelligent they are not.

Images | iStock/Voyagerix, iStock/IPGGutenbergUKLtd, iStock/IPGGutenbergUKLtd, iStock/solvod

Comments

Internet of plants: a tomato knows as much about networks as you

The possibility of plant species communicating through complex networks, what some scientists have called the internet of plants , is an issue that has been discussed for a long time, at least 20 years.It was in 1997 when the scientist Suzanne Simard , from the University of British Columbia in Vancouver (Canada), published a study in the journal Nature on how Forest plants use complex systems where species exchange nutrients, send warning signals and relate to the environment. Although there is no scientific consensus, several investigations seem to support this hypothesis.If confirmed, as we read in a fascinating report by the Sinc Agency, the forests would act as huge structures that are articulated under the soil through a network in which factors invisible to the human eye interact and can determine the future of the climate.Understanding how it works is the challenge that even science faces. internet of plants and the wisdom of forests The premise of Suzanne Simard ...

New setbacks for conspiranoicos: radiofrequencies do not affect health

New evidence published by the Scientific Advisory Committee on Radio Frequency and Health (CCARS) in its 2016 Radio Frequency and Health Report indicates that there are no adverse health effects derived of exposure to radio frequencies emitted by mobile phone antennas or radio and television broadcasting.No evidence has been found on the danger in wireless systems (Wi-Fi) or airport scanners. The new 2016 Radio Frequency and Health Report collects, updates and analyzes the scientific evidence on the subject for the period between January 2013 and June 2016. radio frequencies are not dangerous According to the conclusions of the report " the critical analysis of the evidence supports that there are no technical or sanitary reasons that justify the arbitrary and discretionary imposition of more demanding exposure limits than those recommended by WHO-ICNIRP and the Union European ", while the application of more restrictive limits" would imply increasing the n...

GIF history: where they come from and where they are going

Funny, witty, curious... GIFs have already become a regular resource in social networks and messaging applications and have become part of our daily lives and jokes.It is an image format that has evolved significantly throughout its history and is experiencing a new awakening. In fact, to find its origins, we must go back more than twenty years ago .A moment in which they had little to do with what we know today and an evolution that we could not stop review.So let's find out where they come from and try to catch a glimpse of where they are going. What is a GIF The word GIF responds to the acronym of Graphics Interchange Format , or what is the same, graphic exchange format.A term for the graphic format used in the network for both static images and animations.In more concrete terms, it supports 8 bits per pixel and allows you to display images with up to 256 colors simultaneously-limited by a palette of 16 million-. This particularity makes it a perfect ally of infogr...

Amazon Prime Video arrives in Spain: this is all you need to know

That during the last years there has been a change in the way we enjoy audiovisual content is an obvious reality, as reflected by the huge popularity of streaming services like Netflix and the like.A context in which new competitors are constantly added and that, after the landing of this company in our country just over a year ago, is in full transformation. In fact, today we have learned of the incursion of a new protagonist in this area.Yes, because Amazon Prime Video has just arrived in Spain .A landing that takes place only two weeks away that of HBO, which culminates in a 2016 sown of changes.But, what exactly does it consist of? What advantages and disadvantages does it have compared to the "of its kind"? WHAT AMAZON PRIME VIDEO OFFERS The type, quality and quantity of movies and series offered by the service will be key when determining your success.A catalog a priori smaller than that of HBO and Netflix, but in which we find titles like Seinfield , Into...

What we know and what we suspect in the new Facebook algorithm change

"In my time, we read the newspaper in Facebook ".By 2020, nostalgia has taken over billions of people.In its memory, 2018, when Facebook stopped being Facebook to become Facebook again.Paraiso of engagement for brands and media, land paid for clickbaits and fake news, the social network of Mark Zuckerberg has given a rudder to his algorithm. He wants to return to his origins .And this is what will happen. Related If something goes well, to change it.40.653 million dollars entered in 2017.More than 2.100 million active users per month, a quarter of the world's population.But Zuckerberg is not one of those, of which they are.Want a Facebook with more social connections between people and less presence of brands and media .And what do your users want? Nearly two-thirds of American adults consume news on social networks.More than half do so on their Facebook walls.Pew Research Center data points to a clear trend: Facebook is becoming a media platform.54% of Facebook u...

No kidding: selfitis exists and can be a problem

Sometimes one thinks that social networks were created to share photos of faces putting on little bones and images of dishes overflowing with appetizing foods or, better, of faces putting little bones in front of plates overflowing with appetizing foods in the same shot.is to see faces, faces, faces...all of them smiling, as if looking at the smartphone mirror was the height of happiness.And it does not seem that there are so many beautiful and happy people, or maybe yes, that you go Namely.What envy. Well, what we're going to do: Janarthanan Balakrishnan psychologists from the Thiagarajar School of Management in India; and Mark D.Griffiths, of the Nottingham Trent University, in the United Kingdom, have published an article in the International Journal of Mental Health and Addiction entitled 'A study on' Selfitis', which names this mania of going around the world taking pictures of oneself, as if beyond our curls the world had been destroyed by that meteorite th...

Twitter shows the best of ourselves (it wasn't always going to be the worst)

In the era of social networks, accustomed to interact daily with Facebook, Twitter, Instagram, Google or WhatsApp, we forget that these tools are phenomenal in the History of Humanity capable of mobilize hundreds of thousands of people in very different geographical locations in a matter of hours or even minutes. Already in the past, Twitter has asserted its more social dimension to mobilize people, companies or organizations and also, from a broader perspective, feelings of solidarity and support or also of rejection and repulses towards causes with an important mediatic impact.Recently, Twitter was once again the engine of solidarity after Dortmund's explosions as the Borussia Dortmund team bus passed on its way to the local stadium on the occasion of the Champions League match between the Borussia and the Monaco. These explosions did not have the devastating effect that they could have caused , so that everything remained in a major social commotion at international lev...

Most apps for children violate your privacy

Do you know if apps with which children enjoy their safety? A team of researchers from the International Computer Science Institute (ICSI) of the University of Berkeley has studied compliance with the Online Privacy Protection Act (COPPA) in these mobile applications for the little ones, through a scalable dynamic analysis that allows the automatic evaluation of privacy behaviors of Android apps. After studying 5,855 applications aimed at children, the research concludes that most of these apps, 57%, violate the privacy of their users mainly due to the use of data by third parties. apps for children and privacy Although many apps offer options in their configuration to respect the federal law that protects minors in the United States, by disabling tracking, their results suggest that a majority of applications either do not use these options or propagate them in wrong way. In addition, researchers believe that 19% of apps for children send user IDs to third parties to show ...

Are social networks damaging democracy?

The revelations according to which Russian agents inserted ads on Facebook that tried to influence the 2016 US elections raise a disturbing question: is Facebook bad for democracy? As an expert in the social and political repercussions of technology, I think that the problem is not exclusively from Facebook, but that it is much broader: Social networks are weakening some of the conditions that have historically enabled the existence of national states democratic. I understand that it is a dramatic statement, and I do not expect anyone to believe it immediately, but considering that almost half of all potential voters received false news promoted by the Russians in Facebook is an argument that must be debated. Associate Professor of Philosophy, Director of Center for Professional and Applied Ethics, University of North Carolina-Charlotte How we create a shared reality Let's start with two concepts: the "imagined community" and the "bubble filter". ...

This is how visual challenges star in some of the most viral content

Constantly, we all receive in our mail, our WhatsApp or our Facebook account, games that try to test our ingenuity, imagination or visual skill.There are a thousand types, more fun or darker.But they have one thing in common: we cannot avoid trying them and see if we are able to solve them .This is how the visual challenges star in some of the most viral content of the network. Visual challenges, why are we crazy? The visual challenges are games.So, they attend to this playful and fun part that people have no choice but to feed from time to time.They are pure and hard entertainment, with no more pretensions, and that desire to Having fun represents one of the main features that are sought when surfing the Internet . Another of the qualities for which they succeed is because they challenge the mind directly.Human beings love that feeling of feeling a little troubled, somewhat lost, but convinced that there is a solution to that enigma to which, sooner or later, it will come.Rathe...

Network One Online

Search This Blog