Editor's Blog
There is a common perception that, although computers may be able to perform calculations faster than us, they will never be able to “think” or be “truly creative”. Even among those who reject this perception, there is a commonly held belief that it will take “hundreds” or “thousands” of years for computers to match the brain’s abilities. Supporters of this view often cite the failed AI projects of the 1970s and 1980s — concluding that, like nuclear fusion, AI is the technology of the future and always will be.
However, unlike nuclear fusion, AI technologies have already become a part of our daily lives. Thirty years ago, if you wanted to read some papers written in another language, your only real options were to find a friend who knew it, or hire a translator. Nowadays, Google Translate, an AI technology, provides us with instant, free, automatic translation of any web page or document, between more than two thousand language pairs. The technology is, of course, not perfect, but citing imperfections in a wonder of the modern age as a reason why “computers can’t really think” brings to mind those who said that “this ‘telephone’ has too many shortcomings to be seriously considered as a means of communication”.
Facial recognition is another AI technology that has hit the mainstream. A laboratory curiosity until recently, Toshiba now offers dozens of laptop models with “login through facial recognition”; instead of typing in a password, you can simply look at the computer’s camera, and it will match your face to your profile and log you in automatically. It would be relatively straightforward to extend this technology to eliminate passwords altogether; there would be no more “wrong password” error messages or memorizing cryptic strings of numbers and symbols, because you could just look into the camera, and the computer would know it was you.
Many more AI technologies work quietly, behind the scenes, supporting our infrastructure and preventing disasters. For instance, in the late 1990s, the new online payment company PayPal was hit with a wave of credit card fraud schemes. Losses ran into the tens of millions of dollars. Given the millions of payments flowing through the system, there was no way that a few dozen PayPal employees could stem the losses, and it looked like the fraud problem was going to grow until it bankrupted the company. However, several of the people at PayPal built an artificial intelligence system that could sift through the millions of payments and predict, with fairly high accuracy, which ones were part of a criminal operation and which ones were innocent. Coupled with human investigators, AI managed to put a cap on the fraud problem, and PayPal was bought by eBay in October 2002 for $1.6 billion.
Yet another “technology of the future,” quietly operating behind the scenes, is AI-driven robotics. Human-shaped, anthropomorphic robots remain in the laboratory, because merely copying the human form as closely as possible is not an effective way to get work done. But factory robotics, a less-known technology, is one of the key drivers behind our increased manufacturing ability over the past thirty years. Robot labor in factories has so completely replaced human labor that the First World is, in a very real sense, slowly approaching a post-scarcity society, where a large number of physical goods — from books, to toys, to clothes — can be had for pennies on the dollar, plus the cost of shipping.
When compared to other branches of engineering, like electrical, chemical, nuclear, and mechanical engineering, AI’s impact on the world is admittedly modest… but this is mostly accounted for by the fact that AI, as a field, is only fifty-five years old. Fifty-five years after its discovery, electricity was still mainly seen as a laboratory curiosity. A hundred years later, it formed the backbone of our industrial civilization, and if anything, AI has even greater potential. Electricity powers our computers and light bulbs, but intelligence powers almost everything we do — from our ability to read and write, to our artistic and scientific endeavors, to our engineering and technological capabilities. If the distinguishing feature of a bird is its ability to fly, and the distinguishing feature of a cheetah is its ability to run very fast, the distinguishing feature of our species is clearly our intelligence and high mental capacity.
When we used machines to enhance our physical strength, we found that the combination of man and machine could move rivers and build skyscrapers. If we can use machines to augment our intellectual strength, the sort of great breakthroughs that used to only happen once in a lifetime might become commonplace. There is, as far as researchers in the field can tell, no technological reason why everyone can’t discover new laws of physics as quickly as Einstein, or play chess as well as Kasparov, or compose music as well as Mozart, if we have sufficiently good training programs and AI-based intelligence augmentation technology. Given these abilities, AI might propel our civilization to a whole new level, as different from our 21st-century society as our modern society is from hunter-gatherers.
See also:
Sign up for the Humanity+ newsletter:
8 Comments
You’re shifting the definition of AI. What people mean by “intelligence” isn’t simply the execution of an algorithm, even a very sophisticated one. If that was what they meant, then all computers count as AI. At a minimum, intelligence should cover both a general-purpose power of devising new algorithms, and the ability to reflect about which goals are worth pursuing in the first place, and how and when they should be traded off with other goals.
Put that way, it’s not clear why we should want an AI, since it could decide it wants something we don’t. But a near-AI which can’t reason that far back is also dangerous; for example, if we tell it to make us happy it might leave us sessile and unable to think or experience anything but a drug-style high.
That’s one of the reasons I’ve always been puzzled by how much some folks want an AI, as opposed to augmenting our own minds (perhaps with silicate hardware). Given a choice between making yourself vastly more intelligent, and creating something vastly more intelligent that you, the former strikes me as obviously preferred in every way. Even if the AI works perfectly, what does that leave us as? Pets?
I think this is a great article and I want to thank you for your thoughts and concerns. There has been so much technology come about in the past, even, 20 years that it is hard telling how soon or long as computer that actually thinks can come about.
These slots and poker games at this CASINO ONLINE will provide you with hours of gambling entertainment and a chance to win big.
I consider Intelligence to not only be the execution of a fairly complex algorithm, but the ability to make associations between different pieces of data.
For example, if we figured out that we could use mathematics to model the trajectory, velocity, position, etc. of a stone which we have thrown, maybe we can use mathematics to model how birds or planes fly. This process is what we call creativity and is an essential part of any intelligent system.
Why would the creative process be impossible to “code” into computers? Or replicate some how? It’s quite a simple concept: take a problem, apply a solution which you’ve used in a different situation and see if the problem is solved.
I think the one shifting the definition of A.I. is you. You can’t debate on how intelligence is obtained (i.e. through algorithms) because we can’t define intelligence. Who’s more intelligent? An A.I. cabable of beating the hell out of you in a strategic game or a person affected by the Parkinson’s desease who can’t even speak coherently? You will say: yes, but the computer beats me at his own game. No. He beats you in an environment he knows by heart. He knows the rules, he knows how to mathematically take advantage of them, he can adapt to your strategy, he wins. The point now is to produce a set of algorithms complex enough to reproduce human intelligence, something not too specific like the A.I. we have today that works only in very specific contexts because of parallel calculations limits / memory constraints.
Intel and IBM are confindent this can be done, and I trust what the specialists say.
And, on your second point, I find amusing the most dangerous thing you can think about when discussing A.I. is that we can end up being “pets”. Have you ever thought what will be of us if we don’t research into complex A.I.s? Will we ever be able to keep under control increasingly sophisticated economics, energy crisis, climatic problems, transports, security issues, research challenges into health, genetics, science? Think about it.
If we can’t define intelligence then the project of attempting to create it will always be incoherent. I offered at least a first step, or core cases, of what we mean; a game-playing program obviously fails, in that I could invent new games, and other humans can play as soon as the rules are explained, but the program could not even play badly. Your whole argument here is nothing more than highly metaphorical language treated as if it were literally the case: “he” for it, “knows how” for a machine executing an algorithm.
I explicitly called the pet scenario the result of an AI working perfectly. Where does “the most dangerous thing you can think about” come from? Don’t answer: it doesn’t matter. Your litany of stuff we need an AI to figure out for us fails twice. First, none of it is any more difficult than problems we’ve been solving all along. For that matter none of it is any more difficult than inventing an AI. That sort of litany is already a well-worn trope, of course, to suggest by its length and variety (you should have included a few slightly outdated news stories, the sort of thing a person would recognize but not have had in his consciousness) that the world is spinning hopelessly out of control (usually to talk up nostalgia, not abdication, but one of the advantages of intelligence is thinking up new uses for things, even if those uses are stupid), giving this impression because no individual listener could ever master all those topics at once, or even really care about them all. But nobody has to because we have specialization; this rhetorical devise is therefore always misleading.
But the bigger failure in your litany is this: I advocated augmenting human cognition. You did read that, didn’t you? Even if you had managed to come up with problems beyond human ken, that doesn’t address my argument, which is that making ourselves more intelligent is preferable to making something else more intelligent.
>If we can’t define intelligence then the project of attempting to create it will always be incoheren
No, we can’t define exactly intelligence because it’s not so precisely defined like you seem to believe. Instead, there are infinite degrees of intelligence and we are at the top of the intelligence scoreboard, for now.
A cat is not intelligent like us but he *is* intelligent in his own way. A.I. s are not intelligent like us, but they are improving and, above all, they have this big advantage: they are not biological entities. We have this great amount of intelligence, obtained in thousands of years of biological evolution, but we can’t get faster than this, at last these A.I. will surpass us. How since we can’t exactly define what intelligence is? Because we can’t define intelligence with precise contours, but we can define what intelligent behaviour is. Replicating that is enough to reproduce intelligence, we don’t need to figure out what exactly make US what we are, we don’t need an A.I. to behave exactly as we do (actually, we do need them not to do it), we need them to solve problems, not to ask ourselves metaphysical (and a bit useless) questions such as what we are, what makes us human et cetera. This is the wrong approach to the whole matter. This whole and obvious introduction is justified by the fact that you seem to fail to recognize that “not so intelligent” is still intelligent, at the point that even in your example you contradict youself. You say that a person can learn how to play any game given he knows the rules, but that’s also true for an A.I. An A.I. can learn how to play many games if we explain him, through a programm, how to learn computer games’ rules in general. Eventually, A.I. is about schematizing how to learn and classify anything, and given enough computing power, we can abstract the way the brain learns. You can read about the projects “Blue Brain” and “Blue Matter” for an insight on this.
>I explicitly called the pet scenario the result of an AI working perfectly. Where does “the most dangerous thing you can think about” come from? Don’t answer: it doesn’t matter
I didn’t answer because it seemed to me the answer was superfluous. “We can end up being pets”. Yes, this is one of the possible outcome of the whole superhuman A.I. scenario. But there’s worse: we can end up being annhilated. So what? Do you want me to speculate on a solution to a possible problem whose possible cause has not even yet arisen? Let’s try to make A.I. more intelligent first, we will discuss this later. Of course the possibility of being reduced to pets won’t stop technology’s development: nothing did so far and nothing will.
>That sort of litany is already a well-worn trope, of course, to suggest by its length and variety (you should have included a few slightly outdated news stories, the sort of thing a person would recognize but not have had in his consciousness) that the world is spinning hopelessly out of control
Nah… I’m not that sort of pessimistic jerk. Ok, let’s say you can manage all these problems the way you do now: in a mediocre way more or less. Sometimes things go really bad, but we can recover someway. Yeah, well done. You don’t seem to recognize that in every field I cited simple A.I.s are already being deployed to make managing more efficient. A.I.s can only become more efficient and independent in the decision making process in the future, and at the end this is the only thing that can reduce the margin of error, there’s no other way, we can’t be as efficient and we can’t adapt as fast as we need. I hope so because a good A.I. doesn’t mean, as you think, to marginalize our role, but to save human beings. I don’t care a damn if my car is controlled by an A.I., if this means a very low chance of getting involved in a car incident. This is just an example.
Finally, I think you missed the point of my whole post: I’m as down to earth as one can be, and I see the whole matter in a very realistic light. I just think we have to stop to think that A.I. will never be intelligent as we are because this kind of approach isn’t going to help anybody, not the ones researching into A.I. and not the ones trying to tackle the problems the could arise from it.
By the way, know what? I said the whole “risk to being reduced to pets” was an idle question, but if you want me to play at this game, I will enjoy it.
When you say we’re going to be pets I don’t think your imagine of that is us being carried about with a collar by a robotic master, isn’t that right? So what do you think when you think about this possibility? For us treating an animal like a pet means taking care of every need he can have. If this is proportionally true for us, in order to treat a human being like a pet an A.I. should be so advanced to satisfy the infinite and complex needs the human beings have (in this case I would allow the A.I. to treat me like a pet…). So what’s wrong with it? I’ts not a mystery that no animal in the wild live as long as in captivity, that’s the cruel law of the state of nature. It was to evade it that we got out of the forests, built our cities, made the social pact, wrote our laws. If you think an A.I. will alter your state of nature you can as well get back to the jungle fighting for a scrap of food, so you will be as “natural” as you can get. But we are far from our original state of nature, everything artificial we have built IS part of our current state of nature. So an A.I. will never be a deviation from the human nature, but the task of the human intelligence being accomplished.
Quote: “Because we can’t define intelligence with precise contours, but we can define what intelligent behaviour is. Replicating that is enough to reproduce intelligence, we don’t need to figure out what exactly make US what we are, we don’t need an A.I. to behave exactly as we do (actually, we do need them not to do it), we need them to solve problems, not to ask ourselves metaphysical (and a bit useless) questions such as what we are, what makes us human et cetera.”
You are making a non-sequitor typical of most scientists i.e intelligence is simply a statistically defined set of behaviors we observe from current humans can therefore if we create a program that reproduces those defined behaviors we can classify it as intelligent. It’s like traveling back in time seeing a potter and making a automatic pot maker and declaring that you’ve an A.I that’s as smart as the potter.
Intelligence is about the ability to change behaviors when faces with new problems and therefore you can never define by a statistical process. The best example of the failure of statistical models was failure of Neoclassical economics to predict the financial crash despite the AI driven models.
Intelligence isn’t about solving problems it’s about finding them. You can’t even define new problems based on statistically defined behaviors because by definition they are outside of the programmed parameters.
How do you teach a computer to defining human problems without considering the metaphysical questions of human existence?
Only humans can define what is useful to humans and anything we make is subject to that consideration. A computer is a tool and it will never anything but no matter how much humans anthropomorphise that tool.
The greatest danger to humanity is creating a machine that superficially “thinks” and acts like so much like us that we actually believe that it can do our thinking for us and we give up our creative and thinking culture and when Murphy’s law breaks down the system we will find ourselves without the ability to regain the culture of true creativity.
Do not make AI into a graven image, the blind god of the blinded scultors.
Pygmalion and Galatea is just a myth and so is super human AI.