Wise of the machines

Every year around November since 1991, contestants get together to vie for the Loebner Prize. In the contest, ‘judges’ sitting in front of a computer screen must decipher whether the ‘person’ they are having a conversation with is a person or a machine. Any machine which can fool the judge into thinking it is human gets to win. The contest is based on a practical playing out of a proposition put forward by Alan Turing, the father of research into artificial intelligence.

So far confined to the realms of science fiction or Hollywood, artificial intelligence or machines that ‘think’ are finally beginning to make their mark. One landmark event that has drawn immense attention is the triumph in a ‘Jeopardy!’ challenge quiz of an IBM computer, Watson which beat veteran Jeopardy! champs Ken Jennings and Brad Rutter. Watson is a product of IBM’s DeepQA project which is looking at developing machines that have the smarts to answer human queries by deciphering the nuances of language and context rather than having the need for the question to be put in their terms.

In ‘The Last Question’ the sci-fi master Isaac Assimov constructs a scenario where mankind keeps putting an existential question in front of ever improving generation of computers called Multivac. Whether or not the question is answered is hidden in the cosmic confusion of Assimov’s plot construct but the theme has been a recurring one – humans looking at machines as possibly as possessing a higher intelligence.

Marvin Minsky of MIT, one of the pioneers and a most respected AI researcher put forth a very rational approach to understanding the concept of what it means for machines to be intelligent. In his seminal article “Why people think computers can't,” first published in AI Magazine in the Fall of 1982, he says that we ought not to be intimidated by the idea that computers will never have that ‘special’ gift of a Beethoven or an Einstein to weave together symphonies or grand theories because the fundamental point is that we are “so accustomed to the marvels of the unusual that we forget how little we know about the marvels of ordinary thinking.” And this ordinary thinking is what machines of Watson’s ilk are trying to master (read IBM engineer Aditya Kalyanpur’s account on page 50). In the case of Deep Blue, it managed to win by outpacing Kasparov’s thinking in an otherwise mathematical and methodical game. For the next generational leap in AI embodied by the likes of Watson, the idea is to outsmart other humans.

The biggest question however will remain – will these smart machines supplant us or support us? 
 

The Future of Intelligence


Stephen Baker, who has chronicled Watson’s journey from an idea to a Jeopardy! winning machine in the book “Final Jeopardy: Man vs. Machine and the Quest to Know Everything" (see excerpts), talks about the impact of the advances that Watson’s achievements represent in the field of artificial intelligence.

Do you foresee the computing power of Watson (or a significant fraction thereof) being replicated on a more manageable scale in the near future to make this kind of ‘intelligence’ far more accessible?

Initially, I see Watson-like services being sold to large companies to provide specialised services. Many of the customers, I would imagine, would subscribe to it as a service, getting the answers through the network. Watson would be in the "cloud." And some of them would provide those services to us as information that we could access through the computer or the phone. With the years, question-answering computers will become much smaller and cheaper. That's the pattern with all computing. It has to do not only with smaller and cheaper semiconductors, but also smarter algorithms.

In an Isaac Asimov story, ‘The Last Question’, humankind keeps asking a supposedly unknowable question to Multivac and its descendent computers. The theme has found its way into other stories as well. Why this fascination that a machine’s answer might be more accurate or ‘final’ than what humans can abstractly conceive? Is there any merit to this argument?

We always look for a smarter and more knowing power. Some people see it in a God. Some await smarter aliens. Some even believe that animals have wisdom that we lack. A smart machine fits this role, and it raises the question: If we produce machines that are more intelligent, what happens to us? Will they change the way we think and live? Cure diseases? Enslave us? As any science fiction reader can attest, it leads to lots of interesting scenarios.

Is the development of artificial intelligence driven by this sentiment or should it be seen as a supplement for the more mechanical things that we are not so good at doing simultaneously and quickly?

I think that the development of a "sentient" computer intelligence is many decades away, and more likely centuries. That said, the machines we build will be dramatically more effective and "smart" in coming years. And some of the breakthroughs may well come from what we learn about the workings of the human brain. The IBM researchers, who built Watson, did not focus on the human brain. They were not interested in creating a sentient being. They just focused on results. Their path clearly made sense. But there's no reason for other researchers not to pursue the loftier goal of building machines that think more like us.

Should we be bracing ourselves for a Skynet kind of scenario (in the movie Terminator, computers become self aware and take over the world) soon or Terminator like flourishes of the imagination are best left confined to Hollywood?

Watson is a severely limited computer that cannot do any cognitive work outside the realm of answering questions. If you ask Watson what time it is, it will search its data for all kinds of answers about time, but will not know to look at the clock. My point is that Watson is far from the Skynet vision. I don't see that type of computer coming any time soon. That said, humans can use ever more powerful computers to monitor our lives and optimise us as shoppers, voters, patients. I wrote a previous book about that called The Numerati. In our lifetime, I would say, computers will exert great and growing power over our lives. But they will be following the commands of fellow humans. 
 

The Ideas Frontier

Only four years earlier, in 2006, Watson was a prohibitive long shot, not just to win at "Jeopardy," but even to be built. For more than a year, the head of IBM Research, a physicist named Paul Horn, had been pressing different teams at the company to pursue a "Jeopardy"- playing machine. The way Horn saw it, IBM had triumphed in 1997 with its chess challenge. The company's machine, Deep Blue, had defeated the reigning world champion, Garry Kasparov. This burnished IBM's reputation among the global computing elite while demonstrating to the world that computers could rival humans in certain domains associated with intelligence.

That triumph had left IBM's top executives hungry for an encore. Horn felt the pressure. But what could the researchers get a computer to do? Deep Blue had rifled through millions of scenarios per second, calculated probabilities, and made winning moves. But it had skipped the far more complex domain of words. This, Horn thought, was where the next challenge would be. The next computer should charge into the vast expanse of human language and knowledge. For the test, Horn settled on "Jeopardy."

The quiz show, which debuted in 1964, attracted some nine million viewers every weeknight. It was the closest thing in the United States to a knowledge franchise. "People associated it with intelligence," Horn later said.

There was one small problem. For months, he couldn't get any takers. "Jeopardy," with its puns and strangely phrased clues, seemed too hard for a computer. IBM already had teams building machines to answer questions, and their performance, in speed and precision, came nowhere close to even a moderately informed human. How could the next machine grow so much smarter?

Horn eventually enticed David Ferrucci and his team to pursue his vision. An expert in Artificial Intelligence, Ferrucci had a wide and ranging intellect. He was comfortable conversing about everything from the details of computational linguistics to the evolution of life on earth and the nature of human thought. This made him an ideal ambassador for a "Jeopardy"-playing machine. After all, his project would raise all sorts of issues, and fears, about the role of brainy machines in society. Would they compete for jobs? Could they establish their own agendas, like the infamous computer, HAL in "2001: A Space Odyssey," and take control? What was the future of knowledge and intelligence, and how would brains and machines divvy up the cognitive work?

For humans, knowledge is an entire universe, a welter of sensations and memories, desires, facts, skills, songs and images, words, hopes, fears and regrets, not to mention love. But for those hoping to build intelligent machines, it has to be simpler. Broadly speaking, it falls into three categories: sensory input, ideas and symbols.

Consider the colour blue. It's something that computers and people alike can perceive, each in their own fashion. Sensory perception is the raw material of knowledge. Now think of the three-letter word "sky." Those letters are a symbol for the biggest piece of blue in our world. Computers can handle such symbols. But how about this snippet from Lord Byron? "Friendship is love without his wings." That sentence represents the third realm of knowledge: ideas. How can a machine make sense of these? In these early years of the 21st century, ideas remain the dominion of humans – and the frontier for thinking machines.

Excerpts from “Final Jeopardy: Man vs. Machine and the Quest to Know Everything" by Stephen Baker, published by Houghton Mifflin Harcourt © by Stephen Baker. 

Elementary Watson


Aditya Kalyanpur, a research staff member at IBM Research and a member of the algorithms team that designed the Watson Question Answering System, explains what the machine is all about and how its ‘intelligence’ works.

Machines are adept at dealing with structured, well-formed data and doing complex exacting calculations over a predefined vocabulary of numbers and symbols. For example, tasks involving database operations or number crunching are relatively easy for machines. However, machines struggle when dealing with unstructured, imprecise and implicit data, which is exactly the problem of natural language understanding (i.e. making sense of words and phrases we humans use to communicate with one another). 

The Question-Answering (QA) problem requires a machine to go beyond just matching keywords in documents, which is what a web-search engine does, and correctly interpret the question to figure out what is being asked. The QA system also needs to find the precise answer directly without requiring the human to read through the returned documents. 

Since natural language is ambiguous, tacit and highly contextual, the computer needs to consider many possible meanings, preferably in parallel (to save time) to correctly interpret and analyse the question. This is the key underlying principle of the DeepQA software framework, which is used to build Watson, our Jeopardy! playing computer system.

The DeepQA architecture assumes and pursues multiple interpretations of the question, generates many plausible answers or hypotheses and collects and evaluates many different competing evidence paths that might support or refute those hypotheses. 

Under the hood, there are hundreds of algorithms (or programmes) in DeepQA, each of which is designed to perform a specific task and many of which run independently, allowing the system to achieve a massive scale-out through parallelisation. 

Watson AI vs. human intelligence

As you may have heard, Watson won the Jeopardy! Man vs. Machine Challenge played on Feb 14-16, 2011, beating two former grand champions. It was a great demonstration of the DeepQA technology underlying Watson. It also attracted broad attention and got people discussing the strengths and limits of language understanding technology and raised the question of whether Watson “thinks” like a human. 

We can only speculate that in some ways, the machine approaches the problem as a human does – it considers a vast number of potential interpretations in parallel and analyses and compares them based on evidence gathered, while using previous knowledge to learn how to weigh and balance information. 

In most every other way, however, the machine operates differently from humans. Humans get at the meaning of words, phrases and sentences, conceptualising and visualising information if necessary, and using intuition, experience etc to understand and internalise the text. On the other hand, the machine only “understands” the language by finding statistically significant patterns in the usage of words and phrases, considering additional semantics based on context and background knowledge. By training on these patterns, it comes up with answers that make sense to us in the given context. 

Future of deepqa

There is still a long way to go before machines can achieve full natural language understanding. Nevertheless, the DeepQA technology in Watson represents a huge leap in the way machines can make sense of vast amounts of unstructured, natural language text. The next step is to apply the technology to several application areas, especially information-rich domains with the need for critical and rapid decision-making. 

One such area is medical diagnosis, a task that is a great fit for DeepQA, given that doctors need to sift through a lot of data – medical records, lab tests, patient history, latest medical journals and books etc – to come up with a set of probable diagnoses. Precision, confidence and speed are all key factors in this process and are the strengths of the DeepQA framework.