There is also progress in a less advertised field, bionics, that doesn't require a lot of intelligence but it improves our ability to do things.

There are two major centers of research in bionics. One is at Brown Univ, especially Arto Nurmikko's laboratory that in 2008 develped BrainGate in collaboration with Cyberkinetics: a wireless transmitter for paralyzed patients with a neural implant that bypasses the spinal cord. In 2011 Leigh Hochberg of that team used BrainGate to make a paralyzed woman operate a robotic arm simply by thinking about the movement. The other one is at the Federal Institute of Technology (EPFL) in Switzerland that in 2015 built a robotic wheelchair for paralyzed people. This chair combines brain control with artificial intelligence. In 2016 Gregoire Courtine at EPFL used Brown's BrainGate to restore movement to a primate's paralyzed leg. In 2016 Facebook's Building 8 launched a project directed by Mark Chevillet to build a neural prosthetic that will let people type out words via a brain-computer interface (BCI). In 2017 ARM, maker of the most popular chip for smartphones, opened a Center for Sensorimotor Neural Engineering. In 2016 Nick Ramsey in the Netherlands inserted wireless electrodes into the skull of a paralyzed patient (unable to speak or move) so that she can now control a computer mouse simply by thinking of moving her fingers. Also in 2016 Niels Birbaumer in Germany equipped victims of complete motor paralysis with a neural device that allowed them to answer yes/no questions with their thoughts. And also in 2016 Bin He at the University of Minnesota demonstrated an EEG cap fitted with 64 electrodes that could convert the "thoughts" of a person into the movement of a robotic arm: no brain implant required. In 2017 Bill Kochevar, a man with complete paralysis, fed himself using Bolu Ajiboye's brain-controlled arm made at Case Western Reserve University. In 2018 Stefan Harrer of IBM Australia announced GraspNet, a system that uses Deep Learning (running on an embedded Nvidia chip) to decode EEG signals and control a robotic arm. This had been done before but it was extremely difficult to decipher the very weak EEG signals. By using A.I., Harrer's team managed to get a clearer signal. In 2017 bionic projects were carried out by surgeon Eric Leuthardt at Washington University by Newton Howard at Oxford University (formerly the director of the MIT Mind Machine Project), whose neural implant used technology from Intel and Qualcomm, and Dong Song at University of Southern California, whose brain implant boosted human memory. In 2016 DARPA's Neural Engineering Systems Design project, run by Phillip Alvelda, funded several bionic projects, including the project by Edward Chang at UC San Francisco to treat mental illnesses. In 2016 Elon Musk started Neuralink and Bryan Johnson started Kernel, two widely advertised startups. But they were not the first ones. Among the pioneers were Thomas Oxley's Synchron and Matt Angle's Paradromics.
This is certainly an exciting field that can restore movement to paralyzed people, but there are concerns for how this will be used. Most people are concerned that governments could use this technology to control people, but maybe we should be more concerned about how ordinary people will use it. Here is an example from a very similar field. In 1946 a surgeon called Robert Heath at Tulane University invented a technique to implant electrodes into brains through small holes drilled into the skull in order to stimulate specific brain regions. There is a book about Heath's invention: Lone Frank's "The Pleasure Shock". Why "the pleasure shock"? Because this procedure tends to induce a feeling of pleasure in the patient. After it was approved in 2002 in the USA, this procedure is now called "deep brain stimulation" and has been performed on about 100,000 people with brain diseases such as Parkinson's disease. Deep brain stimulation costs about $50,000 so it is done only in emergency cases, but one can see a future in which it will become cheaper and people will start using it not to treat mental illness but simply to get pleasure. In other words, there is a risk of creating an epidemics of "pleasure addicts" who spend their spare time self-stimulating their brain with this procedure the same way that some people take drugs or watch pornographic videos.
In 2018 Andrea Stocco's team at the University of Washington, the pioneers of non-invasive direct brain-to-brain communication, demonstrated a network of brains interacting via electroencephalograms (EEGs), which records electrical activity in the brain, and transcranial magnetic stimulations (TMSs), which transmits information into the brain ("BrainNet - A Multi-Person Brain-to-Brain Interface for Direct Collaboration Between Brains", 2018).

In 2018 Gregoire Courtine's spinal implant at EPFL in Switzerland helped three paralyzed men walk again. It wasn't a neural implant, but it still says a lot about how we can heal paralyzed people by using electrical signals: it was an electrical device that boosts signals sent from the brains to the legs.

People complain that the printed newspaper and the printed magazine are disappearing, but what is disappearing is not the writer, it is the reader: 30% of the readers of texts published on the web are robots. Most of those robots belong to big corporations. I am not comfortable when i think that in the future the robot readers will outnumber the human readers. Most of the content is still written by humans, but soon it will be read mostly by machines. Anything you do online is read and analyzed by machines; and not because they "like" what you write. They do it because your private life is a business opportunity. Human readers read my writings because they are interested in what i write. A small percentage of them might also do it professionally, but not many. It is too costly and time-consuming for humans to read all the texts published on the Internet. Machines, instead, read "professionally". They don't read what i write because they are interested in the subject: they are interested in checking if they can "use" what i write for their purposes. There is something sinister in this new kind of reader.

I am worried about what will happen to humanity in a fully automated world. Today we live in a partially automated world. Our interaction with other humans is increasingly limited because machines perform many of the functions that used to be performed by humans. Who gives you cash at the bank? An automatic teller machine. Who hands you the ticket at the parking garage? A machine. We tend to look at the machines that replace humans purely in economic terms: the service is now available 24/7 and it is cheap or even free; a job is lost; we can create more jobs elsewhere because we saved money here; etc. But to me there is a much more important story: if the people around me are replaced by machines, it means that i will interact less with humans. Every time a human is replaced by a machine, it decreases the interaction that i will have with other humans. We talk a lot about human-machine interaction, and tend to ignore the fact that a consequence of human-machine interaction is the decline of human-human interaction. This trend has been going on for at least a century (there used to be armies of telephone operators to direct phone calls, there used to be armies of secretaries typing documents, etc) and will continue into the age of Artificial Intelligence to the point that many individuals, especially the older ones, will only interact with machines. Machines will take care of your house, of your errands, of your health, of your entertainment. This will dramatically reduce your interaction with other human beings, even with your own family (family support will become less and less important). You will make fewer friends. Your coworkers will be robots. Your friends will be robots. Maybe your lovers will be robots. What happens to humanity when you don't interact with humans anymore?

I grew up when sociologists were talking about the virtual movement of information replacing the physical movement of people, but instead the car is becoming a second home, a second center of personal life, equipped with communications, entertainment and, soon, ecommerce. In 2014 Rinspeed in Switzerland demonstrated the "XchangE Concept", an autonomous car equipped with all sorts of digital gadgets. Something like that will put 8 billion people on the road, even toddlers and certainly a lot of elderly people.

Of course i am not too happy that people, especially young people, are becoming so dependent on "smart" devices. Those devices are not particularly more intelligent than the old ones, but sometimes young people look a lot less intelligent than their parents and grandparents, who had to use their brains instead of smart devices. It would be great if smart devices were used for more intelligent activities but instead they are mostly used for e-commerce, entertainment and socializing. Maybe it is not correct to say that smart devices are making us "stupid", but it is true that many people are happy to be "mediocre" and let the devices do the "thinking". Humans always wanted to become omnipotent but now they seem content with letting machines become omnipotent while they (the humans) settle for mediocrity.

In 2014 Fei-Fei Li at Stanford Univ demonstrated a computer-vision algorithm that can describe photos. There were already software algorithms capable of recognizing faces (typically, within your group of friends, but Stanford's system (as well as a similar system developed at Yahoo/Flickr) can recognize what is in a picture, eg two young people playing frisbee in a park.

In 2014 Vladimir Veselov's and Eugene Demchenko's software Eugene Goostman, which simulates a 13-year-old Ukrainian boy, managed to fool 33 percent of the judges of the Turing Test at the Royal Society in London. It was not the first time that a software had fooled a significant number of humans but it was the first time that the humans were allowed to ask any kind of questions.

Alan Turing wrote a paper in 1950 where he toyed with the idea of a machine that can fool people into believing that it is a human being. Turing said that some day a machine may well fool 30% of people. So if you create a program that can fool 30% of people, you passed the Turing test.

Most of this progress is really about "recognizing". And that伍西s because the progress has been in neural networks, which Artificial Intelligence has always used for tasks of the recognizing type. AI was founded in 1955. Initially neural networks were the favorite technique. But it was difficult and expensive to build large neural networks. Our brain has billions of neurons and billions of connections. Creating a neural network of that size was impossible in the past. So neural networks were abandoned in favor of another technique, called "knowledge-based". The simplest way to understand "knowledge-based" AI is to think of the difference between information and knowledge. In an information-based system there is a database that contains the answers. When someone asks a question, such as Who is the president of the USA or Where is Rome, the system looks up the answer: Barack Obama, Italy. But now imagine that I ask the questions "Who WILL be the next president of the USA?" or "Where is Atlantis"? Aknowledge-basedsystem cannot find the answer anywhere: it has to THINK. It has to use the available knowledge, which is incomplete and imprecise, and "guess" who could become president of the USA; just like we do.

Or think of the doctor. What does the doctor do? When you are sick, is there a place where the doctor can find the disease? No medical encyclopedia is the same as a doctor. The doctor uses knowledge, experience, common sense, intuition, etc to "GUESS" what your disease is. If you drank water from a polluted source, he guesses that maybe the water made you sick. If there is an epidemic of flu, he guesses that maybe you have the flu. Plausible reasoning.

This type of AI prevailed until the 1990s. But the results were not very exciting. It was too difficult to encode human knowledge. So AI actually did not have a good reputation. It was theoretical research that never kept its promises. When Hinton and others developed deep learning (which is an evolution of neural networks), AI became popular again because it can really do interesting things like recognizing faces or recognizing voices or recognizing scenes. But the limitations of neural networks is that they are really a form of "pattern matching",which ultimately really means "recognizing". So you have to translate all your problems into problems of "recognition". It is not impossible, it just feels a little unnatural to turn the question "Who will be president" into a question of recognizing a pattern.

Then there is common sense. Neural networks are statistical methods. Fundamentally, they work well if you have many cases and you want to guess the next case. Modern translation systems are a good example. They are NOT based on knowledge of the language (grammar). They are based on statistic analysis. The system "learns" by comparing thousands and thousands of texts that have been translated. When you ask to translate a new sentence, the system guesses based on previous translations. It "recognizes" the closest likely translation. In a knowledge-based approach, the system would know the grammar of the language and use reasoning to understand the sentence, and then it would know the other language伍西 s grammar and create the translation.

The neural network approach was abandoned in the 1960s because we didn伍西t have computers fast enough. But now we can use thousands of servers, each of which is very powerful, to run big computations. So in a sense "deep learning" has been enabled by cheap computing. Of course, there has also been progress in the math of neural networks. George Hinton and others have come up with more and more efficient computational methods. But it would be pointless if we couldn伍西t use thousands of powerful computers to implement those neural networks.

Nothing with this method is wrong. I am just not sure that abandoning the knowledge-based approach was a good idea just like in the past I thought that abandoning neural networks was not a good idea.If you want to guess who will be the next president of the USA, you don伍西t have a statistical sample to work on. What we do is to analyze all the factors that help politicians win the presidential elections. We can spend hours or days arguing who will be the next president. We use reasoning based on knowledge.That isi not what a neural net does.

Deep Learning has become extremely popular, and this year Google/DeepMind's AlphaGo beat a go/weichi master. But if you analyze all these great successes you realize that they depend on 2 factors: 1. thousands of very fast processors; and 2. large collections of human examples. The success in recognizing images started after the creation of ImageNet, where thousands of students upload images and "tag" them. The success in playing chess started after humans created a huge database of chess games played by masters. The success in playing weichi came after humans created a huge collection of weichi games played by masters.

That伍西s why IBM's Watson of 2013 consumes 85,000 Watts compared with the human brain's 20 Watts and AlphaGo of 2016 consumes 440,000 Watts. Your brain can do an infinite number of things with 20 watts. AlphaGo can do only one thing with 440,000 watts. I will be impressed when someone builds a machine that uses only 20 watts and can do just 2 things. No cheating: you can build two machines that do two things and put them inside the same box, but that伍西s cheating. We don伍西t have one million brains to do one million things. We have one brain that can do one million things, and it uses only 20 watts. Obviously we are very very far from building truly intelligent machines. We are building better and better appliances: the light bulb, the washing machine, the refrigerator, the microwave oven, the chess player, AlphaGo, 雾熙 They can do one thing, and they do it better than me.

That伍西s why sometimes I joke about "the curse of Moore伍西s law". In the past the motivation to come up with creative ideas in A.I. was due to slow, big and expensive machines. You had to find a creative way to make machines "intelligent" when machines were slow, big and expensive. So AI scientists came up with a lot of interesting ideas. Today we are in the opposite situation: brute force (100s of supercomputers running in parallel) can find solutions using relatively simple mathematical techniques. Actually, you can find the answer to most questions by simply using a search engine: no need to think, no need for intelligence. In a sense, this has reduced our motivation to come up with creative ideas about intelligent machines.

Then there is common sense. The vast majority of what we do in a day is done without thinking. You don伍西t touch something that is very hot. You don伍西t walk out of a window if you are in a hurry. You don伍西t walk out in heavy rain without an umbrella. There are many actions that would make people laugh: common sense tells them that those actions are wrong. Most of what we do on a daily basis is due to common sense.

Statistical method yields a plausible result but it has not learned why. And that伍西s why the learned skills cannot be applied to other fields. Philosophers like John Searle have always argued that whatever the machine does雾熙 it is not what it "does"; meaning that the machine may have done something but it doesn伍西t know that it has done it. Searle explained it in 1980 with the "Chinese room" example. If you give me a book that has the answers in Chinese to all the possible Chinese questions, and then you ask me a question in Chinese, I will find the answer in Chinese. I give you the correct answer. But I still don伍西t know Chinese. In fact, I only know 3 sentences in Chinese. So when I answer in Chinese, I am NOT answering in Chinese. That applies to neural networks too: they may find the correct answer, but they don伍西t know why. They may translate correctly a sentence from English to Chinese, but they don伍西t know why. It is just that thousands of people translated it that way, so they guess it is the correct trnalsation. But the translating machine doesn伍西t know English and doesn't know Chinese.

1. Reality Check. A.I. has always been a bit optimistic. In 1965 the great mathematician Herbert Simon wrote "Machines will be capable, within twenty years, of doing any work that a man can do". Well... we are still waiting for that day. My reaction to recent achievements is the opposite to the enthusiastic headlines of the news media. Recognizing a cat is something that any mouse can do (it took 16,000 computers working in parallel to do the same). NASA's Curiosity is one of the most sophisticated robots ever built. In 2013 my friend the NASA planetary scientist Chris McKay told me "What Curiosity (robot) has done in 200 days a human field researcher could do in an easy afternoon". A video showing a robot that rides a bicycle went viral, but human-looking automata that mimic human behavior have been built since ancient times, particularly in China. It is true that machines can now recognize faces, and even scenes, but they have no clue what those scenes mean: we will soon have machines that can recognize the scene "someone picked up an object in a store" but when will we have a machine that can recognize "someone STOLE an object in a store?" A human being understands the meaning of this sentence because humans understand the context: some of those objects are for sale, or belong to the shopper, and a person walking away with those objects is a thief, which is very different from being a store clerk or a customer. We can train neural networks to recognize a lot of things, but not to understand what those things mean. The automatic translation software that you use from Chinese to English doesn't have a clue what those Chinese words mean nor what those English words mean. If the sentence says "Oh my god there's a bomb!" the automatic translation software simply translates it into another language. A human interpreter would shout "everybody get out!", call the emergency number and... run!

Nothing puts the progress in A.I. better in perspective than the progress in robots. The first car was built in 1886. 47 years later (1933) there were 25 million cars in the USA, probably 40 million in the world, and those cars were much better than the first one. The first airplane took off in 1903. 47 years later (1950) 31 million people flew in airplanes, and those airplanes were much better than the first one. The first television set was built in 1927. 47 years later (1974) 95% of households in the USA owned a tv set, and mostly a color tv set. The first commercial computer was delivered in 1951. 47 years later (1998) more than 40 million households in the USA had a computer, and those personal computers were more powerful than the first computer. The first (mobile) robot was demonstrated in 1969 (Shakey). In 2016 (47 years later) how many people own a robot that is at least as good as Shakey?

What i see around me is depressing. It is not machines that learned to understand human language but humans who got used to speak like machines in order to be understood by automated customer support (and mostly not even speak it but simply press keys).

What "automation" really means: in most cases the automation of those jobs has required the user/customer to accept a lower (not higher) quality of service. The more automation around you, the more you (you) are forced to behave like a machine to interact with machines.

A lot of "intelligent behavior" by machines is actually due to an environment that has been structured by humans so that even an idiot can perform. For example, the self-driving car is a car that can drive on roads that are highly structured. Over the decades we structured traffic in a way that even really bad drivers can drive safely. We made it really easy for cars to drive themselves.

We structure the chaos of nature because it makes it easier to survive and thrive in it The more we structure the environment, the easier for extremely dumb people and machines to survive and thrive in it. It is easy to build a machine that has to operate in a highly structured environment What really "does it" is not the machine: it's the structured environment

2. Accelerating progress. That伍西s the other dogma of the Singularity movement. Because progress is so rapid and it is even accelerating, then machines will become more and more intelligent more and more rapidly. My problem with this dogma is that it is just not true. There has been the same kind of rapid progress in the past, and maybe even more rapid. Think of the period between 1880 and 1915 when the car, the airplane, the telephone, the radio, the record and cinema were invented. Suddenly people were "flying", and one could talk to her mother who was thousands of kms away, and one could hear the voice of someone dead, etc. Those inventions must have looked like magic. They all happened in that short period of time. At the same time science invented Quantum Mechanics and Relativity. The USA produced 11,200 cars in 1903, but already 1.5 million in 1916. The Wright brothers flew the first plane in 1903: during World War I (1915-18) more than 200,000 planes were built. That伍西s accelerating progress. On the other hand, are we really sure that today there is so much progress? 47 years after the Moon landing we still haven't sent a human being to any planet . The only supersonic plane (the Concorde) has been retired. So I am not even sure that progress in our age is really so special. I am not denying that today there is a lot of change; but change is not necessarily progress. Sometimes it is fashion created by marketing. Sometimes it is change of a commercial type, that mainly benefits the big corporations. Maybe it is "progress", but progress for whom?

Argument number 3: non-human intelligence has always been around. We are surrounded by non-human intelligence, and frequently these non-human intelligences can do things that we cannot do. Bats can avoid objects in absolute darkness at impressive speeds. Birds are equipped with a sixth sense for the Earth's magnetic field. Some animals have the ability to camouflage, to change the color of their skin. Many animals have night vision. Most animals can see, sniff and hear things that we cannot. And flies can fly and land upside down on the ceiling. These brains are more powerful than ours in performing these tasks. We already built machines that can perform tasks that we cannot perform. For example, we cannot do what a light bulb do. One of the most influential inventions of all time is the clock, invented one thousand years ago, that can do something that no human being can do: keep time. I am not sure what they mean when they say that the Singularity will be a super-human intelligence: what is the difference between non-human and super-human? There already are so many kinds of intelligence that can do things that we cannot do. Nothing scary about it.

Number 4: I am more concerned about the future of human intelligence than about the future of machine intelligence. The Turing Test was asking "when can machines be said to be as intelligent as humans?" I always joke that this "Turing point" can be achieved in 2 ways: 1. Making machines smarter, or 2. Making humans dumber! If machines get just a little smarter while humans get a lot dumber, then, yes, we will have machines that are a lot smarter than humans. So that伍西s the danger: not that we create machines that are too intelligent, but that we create people who are too dumb. People make tools that make people obsolete, redundant and dumb. In fact, the success of many high-tech projects depends not on making smarter technology but on making dumber users. "They" increasingly expect us to behave like machines in order to interact efficiently with machines: we have to speak a "machine language" to phone customer support, automatic teller machines, gas pumps, etc. In most phone and web transactions the first question you are asked is a number (account #, frequent flyer#雾熙) because you are talking to a machine. The machine performs its task because YOU spoke the machine伍西s language (numbers), not because the machine spoke your language. Rules and regulations (driving a car, eating at restaurants, crossing a street) increasingly turn us into machines that must follow simple sequential steps in order to get what we need. I am afraid that we talk about Artificial Intelligence while humans are moving a lot closer towards machines than machines are moving towards humans?

And finally #5 there are philosophical objections to the Turing Test. Who is supposed to be the judge of the test? 33% of the members of the Royal Society jury were fooled by the machine in 2014. What if we replaced the Royal Society jury with a group of mountain villagers? Is it possible that the result of the test simply tells us that people who hang out at the Royal Society are not very smart? And what does it tell us that 33% of the jurors thought the human was a machine? What can we conclude from the fact that a human failed the Turing Test (s/he was mistook for a machine by 33% of the jurors)? If a machine fails the test (i.e. the jury thinks the machine is a machine), then we are supposed to conclude that the machine is not intelligent;but what are we supposed to conclude if a human fails the test (if the jury thinks that the human is a machine)? That humans are not intelligent? I think we need a better way to measure the intelligence of a machine. The whole discussion on the Singularity is really very vague.

Some people assume that neural networks simulate the brain so well that machines will soon "think" the same way we think, but that assumption is based on a gross misunderstanding of how the brain works. First of all, we still know very little about the human brain. We can't repair even the most basic of brain diseases. It will take decades or maybe centuries to fully understand how the brain works. So we only have very superficial models of the brain structure. Secondly, the neural networks that we have today are rough approximations of those superficial models. For example, a neural network has only one type of "neurotransmitter", only one type of communication between neurons, whereas the human brain has 52 types of neurotransmitters (and maybe even more). Neural networks assume that the neuron is a simple zero-one switch, but neuroscientists have discovered a very complex structure inside the neuron. Today's machines are very far from simulating a human brain, and we are very far from understanding how the human brain works, so i think that we are very far from the day when we can have a machine equipped with the equivalent of a human brain. When that day comes (probably very far in the future), you can ask me this question again...

The modern mind, however, looks down on many of these systems of symbols. Our society are increasingly based on "rational" rules and regulations that get rid of those "useless" and "expensive" rituals. Life is increasingly programmed to be efficient. Children are sent to school according to a program. Then they are expected to find a job. Even entertainment is highly regulated. This is what i call the "robotic" mind: a mind that has to obey rational rules. Think of the difference between the traditional wedding (in India it could last 3 days) and the secular wedding that takes place in a government office and consists in simply signing a contract. We are genetically programmed to be "symbolic minds" (minds that indulge in rituals and legends) but somehow we increasingly like societies that create "robotic minds".

Now you may begin to understand why i told you that maybe A.I. abandoned the knowledge-based approach too soon. Knowledge-based A.I. was all about systems of symbols: that's how knowledge is represented, that's what knowledge is. If you "know" something, it means that you created symbols about it. The A.I. that prevails today is about the "robotic mind", not the "symbolic mind". The "deep learning" networks are good at recognizing and performing tasks, not at creating complex systems of symbols. That's why i used the term "robotic mind": the robots that we are designing emulate the robotic mind, not the symbolic mind. A robot that has low battery does not start dancing around the fire or praying supernatural beings.

Ultimately, what is the purpose of having machines? I think the answer should be "to make us happier". Right now we are at the stage where we interpret "intelligent" as "useful": the more useful a machine is, the more intelligent it is. But being useful is not the same as making us happy. What is it that machines could do for us that would make us happier? This is not an easy question to answer. We have often interpreted happiness in a material way with the result that we got less happy. Suicide rates tend to be high in countries like Japan and Scandinavia that provide a high living standard to their citizens, and tend to be very low in poor countries. What is it that makes people truly happy? When i travel in poor African countries, i am surrounded by people who smile and laugh all the time. When i walk in Western cities, hardly anybody smiles. The reason for this apparent contradiction is that we often confuse goods with values. Many of the greatest philosophers, including Jesus and Buddha, warned that material wealth does not translate into happiness. That's what the great systems of symbols provide: a path to happiness. That's why it is dangerous to get rid of the "symbolic mind"; and machines that speed up this process then become truly dangerous.

When people ask me about immortality, i remind them that the longest living bodies on the planet have no brain: bacteria and trees. Are they happy? Would you like to be a tree?

(Note: Actually, Narnia would like to be a tree, but i don't think she represents the average reader).

There are two general advices that we can give to younger people (and also to the older people who are afraid of losing their job): knowledge and context. Machines can store a lot of information, but they are really bad at turning that information into knowledge. That's why the machine cannot have a conversation about who will be the next president of the USA, something that every person in the USA is doing during a presidential campaign. People have knowledge. They don't have a lot of information (in fact, some of them are very ignorant) but they have knowledge. They know the problems of the country, they know what other people complain about, they know what the previous politicians did, they know a lot of things about politics and elections. Knowledge is not information. The names of all the presidents of the USA from Washington to Obama is information, not knowledge. Knowledge is that Roosevelt presided over a Great Depression, and what that means. Knowledge is that George W Bush started two wars, and what that means. Here is an example of the difference that knowledge makes. Machines are getting better and better at translating from German to English because there are so many books translated from German to English. Machines learn the statistics of those translations. Machines learn that "Ich bin" is usually translated "I am". The machines will get better and better at this. But what happens if tomorrow we discover a new language? We discover a number of books in Mongolia that are written in a language never seen before. What does the machine do? Absolutely nothing. What does the human expert do? He uses her knowledge to decipher the language. Of course the human expert will use computers to do statistical analyses of the terms. Of course the human expert will use computers to compare this new language to the existing ones. But computers cannot do what this expert does: she is trying to find out the logic behind the writing, she is using her "knowledge" of what a language is. The machine can help the expert with fast computations, but the expert has the knowledge that makes it possible to decipher a new language. A translating machine doesn't even know what a language it. A translating machine is simply a statistical tool to guess that a certain string of characters should be translated into another string of characters.

Secondly, people understand the context. If i ask you "Where is the library"? you may answer "The library is closed" or "The library doesn't have the magazine you want to read" or "The library is very crowded at this hour". This is actually not an answer to that question, but you know the context of the conversation and therefore you can guess important facts that are not asked by my question. The machines are trying to catch up: they display the address and then also the hours of operation. Waze knows the traffic on the road. And so on and on. But humans can still be far ahead of machines in understanding the context. We can listen to a person talking for six hours and turn those six hours into a context. A machine can do it only for a few sentences, then it gets lost in too much context.

If you are simply performing your job like a machine, you will be replaced by a machine. If you are acquiring knowledge and use common sense to understand the context, you will be promoted when a machine replaces you, because you can do something that the machine cannot do. And, when the machine arrives, you can use the machine to get dumber, or you can use the machine to get smarter. You can use the smartphone for WeChat or you can use the smartphone to acquire knowledge about your field of work.

Think of the simplest case in which a human is needed and the human is paid a lot of money: when the machine fails. If the machine breaks down (or it cannot operate because the building lost electricity), a human needs to take over. That human being is very valuable. If a machine takes your job, you want to become that person, the person who knows what to do when the machine fails.