The biggest problem in AI? Machines have no common sense.
Correlation doesn't equal causation — we all know this. Well, except robots.
Dr. Gary Marcus is the director of the NYU Infant Language Learning Center, and a professor of psychology at New York University. He is the author of "The Birth of the Mind," "The Algebraic Mind: Integrating Connectionism and Cognitive Science," and "Kluge: The Haphazard Construction of the Human Mind." Marcus's research on developmental cognitive neuroscience has been published in over forty articles in leading journals, and in 1996 he won the Robert L. Fantz award for new investigators in cognitive development.
Marcus contributed an idea to Big Think's "Dangerous Ideas" blog, suggesting that we should develop Google-like chips to implant in our brains and enhance our memory.
GARY MARCUS: The dominant vision in the field right now is, collect a lot of data, run a lot of statistics, and intelligence will emerge. And I think that's wrong. I think that having a lot of data is important, and collecting a lot of statistics is important. But I think what we also need is deep understanding, not just so-called "deep learning." So deep learning finds what's typically correlated, but we all know that correlation is not the same thing as causation. And even though we all know that and everybody learned that in Intro to Psych, or should have learned that in Intro to Psych they should have learned that you don't know whether cigarettes cause smoking just from the statistics we have. We have to make causal inferences and do careful studies. We all know that causation and correlation are not the same thing.
Have right now as AIs-- giant correlation machines. And it works, if you have enough control of the data relative to the problem that you're studying that you can exhaust the problem, to beat the problem into submission. So you can do that with go. You could play this game, over and over again, the rules never change. They haven't changed in 2,000 years. And the board is always the same size. And so you can just get enough statistics about what tends to work, and you're good to go. But if you want to use the same techniques for natural language understanding, for example, or to guide a domestic robot through your house, it's not going to work. So the domestic robot in your house is going to keep seeing new situations. And your natural language understanding system, every dialogue is going to be different. It's not really going to work. So yeah, you can talk to Alexa and you can say, please turn on my over and over again, get statistics on that. It's fine. But there's no machine in the world T=that we're having right now. It's just not anywhere near a reality and you're not going to be able to do it with the statistics, because there's not enough similar stuff going on.
Probably the single best thing that we could do to make our machines smarter is to give them common sense, which is much harder than it sounds like. I mean, first you might say, what is common sense? And what we settled on in the book that we wrote is that common sense is the knowledge that's commonly held. that ordinary people have. And yet machines don't. So machines are really good at things like, I don't know, converting metrics -- you know, converting from the English system to the metric system. Things that are nice, and precise, and factual, and easily stated. But things that are a little bit less sharply stated, like how you open a door, machines don't understand the first thing. So there's actually a competition right now for opening doors. So if somebody uploaded data sets from 500 different doors, and they're hoping that robots will experiment with all 500 doors and then get it. But what's more likely is that they'll get to door 501 and they'll actually have a problem or at least door 601. So every ordinary person in the west has opened a bunch of door knobs and gets the idea, right? I need to turn something, jiggle something -- might be different for the next one -- until the door itself is free.
So you can give a definition of something like that. It's hard to give a perfect definition. But we all know that. And yet nobody's ever built a robot that can do that. We made a joke about it in the book. We said, you know, in the event of a robot attack do the following six things and number one was close the door. And we added that you might need to lock it. And that was before this whole database came out. So, you know, the field advances. People are working on that. Maybe next year they'll work on teaching robots about locks. But I bet you it will take a while before robots understand all the little ways it can jiggle the key, and maybe you need to pull in the door to make it just right. We don't know how to even encode that information in a language that a computer can understand. So the big challenge of common sense is to take stuff like that-- like, how do you open the door, why do you want to open a door-- and translate into the language of the machine. It's a lot harder than a metric converter. It's a lot harder than a database. And right now the field's not even really trying to answer that question. It's so obsessed with what it can do with these big databases, which are exciting in themselves, that it's kind of lost sight of that, even though the question itself that goes back to the 1950s, when one of the founders of A.I., John McCarthy, first started writing about it in the late '50s.
But it's not-- common sense is not getting the attention that it deserves and that's one of the reasons we wrote the book. Common sense is just one step along the way to intelligence. People talk about artificial intelligence. And sometimes they talk about artificial general intelligence. There's also narrow AI. And narrow AI is the stuff that we're doing pretty well now. So, do the same problem over and over again, just solving one problem. You could think about idiot Savants that can do a calendar but can't do anything else and can tell you what day you were born on if you give them your birthday. We have a lot of narrow AI right now. We can't do narrow AI for everything that we want to do. But the dream is to have broad AI, or general AI, that can solve any problem. You think about the Star Trek computer. You can say, you know, please give me the demographics in this galaxy and cross-correlate it with this, and with that, and tell me this. And Star Trek computer says, OK. And it figures it out. So the Star Trek computer understands everything about language, and it understands pretty much everything about how the world works. And it can put those together to give you an answer.
It's not like Google, right? Google can search for pages that have the information. But then you have to put the information together. The Star Trek computer can synthesize it. And it's a great example of general intelligence. If I ask you, did George Washington have a cell phone? Well, if you have a kind of common knowledge of when cellphones were introduced, when George Washington was alive, the fact that he's dead, then you could compute the information and give me the right answer. If you Google for it, you might get a wacky answer. If it works for George Washington, maybe you search for, did Thomas Jefferson have a cell phone? And the answer should obviously be the same. But maybe that one won't be on a Google page. So if you have common sense about things like time, and space, and a lot of sort of everyday factual knowledge, that gets you a long way to intelligence. It doesn't give you all the way there. So general intelligence, first of all, has many dimensions to it. So you can think like, the SAT has verbal and math.
Those are two of the of intelligence that's Really not well-established right now in the AI community is common sense. There are other aspects of intelligence, like doing pure calculation, where the A.I. community has done a good job. But there are other things that go into intelligence as well. For example, the ability to read a graph is partly about common sense, and it's partly about understanding what people might intend, it's partly sometimes about expert knowledge. So reading a graph is another form of intelligence. And that's going to require putting together better perceptual tools than we have now, better common sense tools, probably some knowledge about politics, for example, if you're reading a graph that's relevant to the latest political campaign, and so forth. So there's a lot of stuff Right now we're trying to approximate it all with statistics. But it's never general knowledge. So you could learn to read one graph, Going help you read the next graph. So general intelligence is going to be putting together a lot of things, both some that we already understand pretty well in the A.I. community and some we just haven't been working on and really need to get back to.
- There are a lot of people in the tech world who think that if we collect as much data possible, and run a lot of statistics, that we will be able to develop robots where artificial "intelligence" organically emerges.
- However, many A.I.'s that currently exist aren't close to being "intelligent," it's difficult to even program common sense into them. The reason for this is because correlation doesn't always equal causation — robots that operate on correlation alone may have skewed algorithms in which to operate in the real world.
- When it comes to performing simple tasks, such as opening a door, we currently don't know how to encode that information — the varied process that is sometimes required in differing situations, i.e. jiggling the key, turning the key just right — into a language that a computer can understand.
Gary Marcus is the author of Rebooting AI: Building Artificial Intelligence We Can Trust.
- Human-like AI could emerge in 5 to 10 years - Big Think ›
- The next Stage of Machine Learning: Teach Robots to Think Like ... ›
- What animals is A.I. currently smarter than? - Big Think ›
Once a week.
Subscribe to our weekly newsletter.
A simple trick allowed marine biologists to prove a long-held suspicion.
- It's long been suspected that sharks navigate the oceans using Earth's magnetic field.
- Sharks are, however, difficult to experiment with.
- Using magnetism, marine biologists figured out a clever way to fool sharks into thinking they're somewhere that they're not.
For some time, scientists have suspected that sharks belong among the growing number of animals known to navigate using Earth's magnetic field. Testing anything with a shark, though, requires some care.
The key was selecting the right candidate. Keller and his colleagues chose the bonnethead shark, Sphyrna tiburo, a small critter that summers at Turkey Point Shoal off the coast of the Florida State University Coastal and Marine Laboratory with which Keller is affiliated.
Bonnetheads elsewhere have been known to complete 620-mile roundtrip migrations. As the lab's Dean Grubbs puts it, "That's not bad for a shark that is only two to three feet long. The question is how do they find their way back to that same estuary year after year." There's a report of a great white shark migrating between two locations, one in South Africa and another in Australia, year after year.
The research is published in Current Biology.
Keller and his team rounded up 20 local juvenile bonnetheads and transported them into a holding tank at the marine lab. For the tests, the researchers simulated three real-world magnetic fields. As the various magnetic fields were activated, the sharks' movements were captured by GoPro cameras and their average swimming orientations calculated by software.
The first simulation, serving as a control, mimicked the magnetic field of the nearby shoal from which the sharks had been captured. When this field was activated, the sharks essentially acted like they were "home," just swimming around as they do.
A second field was the magnetic equivalent of a location 600 kilometers south of the lab within the Gulf of Mexico. When this field was activated, the sharks, apparently mistaking themselves for being far south in the Gulf, began swimming northward toward the shoal.
The opposite occurred with a field standing in for a location in continental North America 600 km north of their home shoal — the sharks began swimming southward.
"For 50 years," says Keller, "scientists have hypothesized that sharks use the magnetic field as a navigational aid. This theory has been so popular because sharks, skates, and rays have been shown to be very sensitive to magnetic fields. They have also been trained to react to unique geomagnetic signatures, so we know they are capable of detecting and reacting to variation in the magnetic field."
His team's experiments confirm what's long been suspected, Keller says: "Sharks use map-like information from the geomagnetic field as a navigational aid. This ability is useful for navigation and possibly maintaining population structure."
A machine learning system lets visitors at a Kandinsky exhibition hear the artwork.
Have you ever heard colors?
As part of a new exhibition, the worlds of culture and technology collide, bringing sound to the colors of abstract art pioneer Wassily Kandinsky.
Kandinsky had synesthesia, where looking at colors and shapes causes some with the condition to hear associated sounds. With the help of machine learning, virtual visitors to the Sounds Like Kandinsky exhibition, a partnership project by Centre Pompidou in Paris and Google Arts & Culture, can have an aural experience of his art.
An eye for music
Kandinsky's synesthesia is thought to have heavily influenced his painting. Seeing yellow summoned up trumpets, evoking emotions like cheekiness; reds produced violins portraying restlessness; while organs representing heavenliness he associated with blues, according to the exhibition notes.
Virtual visitors are invited to take part in an experiment called Play a Kandinsky, which allows them to see and hear the world through the artist's eyes.
Kandinsky's synesthesia is thought to have heavily influenced his 1925 painting Yellow, Red, Blue.Image: Guillaume Piolle/Wikimedia Commons
In 1925, the artist's masterpiece, "Yellow, Red, Blue", broke new ground in the world of abstract art, guiding the viewer from left to right with shifting shapes and shades. Almost a century after it was painted, Google's interactive tool lets visitors click different parts of the artwork to journey through the artist's description of the colors, associated sounds and moods that inspired the work.
But Google's new toy is not the only tool developed to enhance the artistic experience.
Artist Neil Harbisson has developed an artificial way to emulate Kandinsky by turning colors into sounds. He has a rare form of color blindness and sees the world in greyscale. But a smart antenna attached to his head translates dominant colors into musical notes, creating a real-world soundtrack of what's in front of him. The invention could open up a new world for people who are color blind.
A Harvard professor's study discovers the worst year to be alive.
- Harvard professor Michael McCormick argues the worst year to be alive was 536 AD.
- The year was terrible due to cataclysmic eruptions that blocked out the sun and the spread of the plague.
- 536 ushered in the coldest decade in thousands of years and started a century of economic devastation.
The past year has been nothing but the worst in the lives of many people around the globe. A rampaging pandemic, dangerous political instability, weather catastrophes, and a profound change in lifestyle that most have never experienced or imagined.
But was it the worst year ever?
Nope. Not even close. In the eyes of the historian and archaeologist Michael McCormick, the absolute "worst year to be alive" was 536.
Why was 536 so bad? You could certainly argue that 1918, the last year of World War I when the Spanish Flu killed up to 100 million people around the world, was a terrible year by all accounts. 1349 could also be considered on this morbid list as the year when the Black Death wiped out half of Europe, with up to 20 million dead from the plague. Most of the years of World War II could probably lay claim to the "worst year" title as well. But 536 was in a category of its own, argues the historian.
It all began with an eruption...
According to McCormick, Professor of Medieval History at Harvard University, 536 was the precursor year to one of the worst periods of human history. It featured a volcanic eruption early in the year that took place in Iceland, as established by a study of a Swiss glacier carried out by McCormick and the glaciologist Paul Mayewski from the Climate Change Institute of The University of Maine (UM) in Orono.
The ash spewed out by the volcano likely led to a fog that brought an 18-month-long stretch of daytime darkness across Europe, the Middle East, and portions of Asia. As wrote the Byzantine historian Procopius, "For the sun gave forth its light without brightness, like the moon, during the whole year." He also recounted that it looked like the sun was always in eclipse.
Cassiodorus, a Roman politician of that time, wrote that the sun had a "bluish" color, the moon had no luster, and "seasons seem to be all jumbled up together." What's even creepier, he described, "We marvel to see no shadows of our bodies at noon."
...that led to famine...
The dark days also brought a period of coldness, with summer temperatures falling by 1.5° C. to 2.5° C. This started the coldest decade in the past 2300 years, reports Science, leading to the devastation of crops and worldwide hunger.
...and the fall of an empire
In 541, the bubonic plague added considerably to the world's misery. Spreading from the Roman port of Pelusium in Egypt, the so-called Plague of Justinian caused the deaths of up to one half of the population of the eastern Roman Empire. This, in turn, sped up its eventual collapse, writes McCormick.
Between the environmental cataclysms, with massive volcanic eruptions also in 540 and 547, and the devastation brought on by the plague, Europe was in for an economic downturn for nearly all of the next century, until 640 when silver mining gave it a boost.
Was that the worst time in history?
Of course, the absolute worst time in history depends on who you were and where you lived.
Native Americans can easily point to 1520, when smallpox, brought over by the Spanish, killed millions of indigenous people. By 1600, up to 90 percent of the population of the Americas (about 55 million people) was wiped out by various European pathogens.
Like all things, the grisly title of "worst year ever" comes down to historical perspective.
A new study suggests that private prisons hold prisoners for a longer period of time, wasting the cost savings that private prisons are supposed to provide over public ones.
- Private prisons in Mississippi tend to hold prisoners 90 days longer than public ones.
- The extra days eat up half of the expected cost savings of a private prison.
- The study leaves several open questions, such as what affect these extra days have on recidivism rates.
The United States of America, land of the free, is home to 5 percent of the world's population but 25 percent of its prisoners. The cost of having so many people in the penal system adds up to $80 billion per year, more than three times the budget for NASA. This massive system exploded in size relatively recently, with the prison population increasing by six-fold in the last four decades.
Ten percent of these prisoners are kept in private prisons, which are owned and operated for the sake of profit by contractors. In theory, these operations cost less than public prisons and jails, and states can save money by contracting them to incarcerate people. They have a long history in the United States and are used in many other countries as well.
However, despite the pervasiveness of private contractors in the American prison system, there is not much research into how well they live up to their promise to provide similar services at a lower cost to the state. The little research that is available often encounters difficulties in trying to compare the costs and benefits of facilities with vastly different operations and occasionally produces results suggesting there are few benefits to privatization.
A new study by Dr. Anita Mukherjee and published in the American Economic Journal: Economic Policy joins the debate with a robust consideration of the costs and benefits of private prisons. Its findings suggest that some private prisons keep people incarcerated longer and save less money than advertised.
The study focuses on prisons in Mississippi. Despite its comparatively high rate of incarceration, Mississippi's prison system is very similar to that of other states that also use private prisons. Demographically, its system is representative of the rest of the U.S. prison system, and its inmates are sentenced for similar amounts of time.
The state attempts to get the most out of its privatization efforts, as a 1994 law requires all contracts for private prisons in Mississippi to provide at least a 10 percent cost savings over public prisons while providing similar services. As a result, the state seeks to maximize its savings by sending prisoners to private institutions first if space if available.
While public and private prisons in Mississippi are quite similar, there are a few differences that allow for the possibility of cost savings by private operators — not the least of which is that the guards are paid 30 percent less and have fewer benefits than their publicly employed counterparts.
The results of privatization
The graph depicts the likelihood of release for public (dotted line) vs. private (solid line) prison inmates. At every level of time served, public prisoners were more likely to be released than private prisoners.Dr. Anita Mukherjee
The study relied on administrative records of the Mississippi prison system between 1996 and 2013. The data included information on prisoner demographics, the crimes committed, sentence lengths, time served, infractions while incarcerated, and prisoner relocation while in the system, including between public and private jails. For this study, the sample examined was limited to those serving between one and six years and those who served at least a quarter of their sentence. This created a primary sample of 26,563 bookings.
Analysis revealed that prisoners in private prisons were behind bars for four to seven percent longer than those in public prisons, which translates to roughly 85 to 90 extra days per prisoner. This is, in part, because those in private prison serve a greater portion of their sentences (73 percent) than those in public institutions (70 percent).
This in turn might be due to the much higher infraction rate in private prisons compared to public ones. While only 18 percent of prisoners in a public prison commit an infraction, such as disobeying a guard or possessing contraband, the number jumps to 46 percent in a private prison. Infractions can reduce the probability of early release or cause time to be added to a sentence.
It's unclear why there are so many more infractions in private prisons. Dr. Mukherjee suggests it could be the result of "harsher prison conditions in private prisons," better monitoring techniques, incentives to report more of them to the state before contract renewals, or even a lackadaisical attitude on the part of public prison employees.
What does all this cost Mississippi?
The extra time served eats 48 percent of the cost savings of keeping prisoners in a private facility. For example, it costs about $135,000 to house a prisoner in a private prison for three years and $150,000 in the public system. But longer stays in private prisons reduce the savings from $15,000 to only $7,800.
As Dr. Mukherjee remarks, this cost is also just the finance. Some things are a little harder to measure:
"There are, of course, other costs that are difficult to quantify — e.g., the cost of injustice to society (if private prison inmates systematically serve more time), the inmate's individual value of freedom, and impacts of the additional incarceration on future employment. Abrams and Rohlfs (2011) estimates a prisoner's value of freedom for 90 days at about $1,100 using experimental variation in bail setting. Mueller-Smith (2017) estimates that 90 days of marginal incarceration costs about $15,000 in reduced wages and increased reliance on welfare. If these social costs were to exceed $7,800 in the example stated, private prisons would no longer offer a bargain in terms of welfare-adjusted cost savings."
It is possible that the extra time in jail provides benefits that counter these costs, such as a reduced recidivism rate, but this proved difficult to determine. Though it was not statistically significant, there was some evidence that the added time actually increased the rate of recidivism. If that's true, then private prisons could be counterproductive.