Maximum PC artificial intelligence RSS Feed

Future Tense: The Road to HAL

David Gerrold — Thu, 03 Dec 2009 19:30:00 -0600

The Turing Test says that if you can’t tell if you’re exchanging texts with a machine or a human being, then the machine has achieved cognitive ability—it’s thinking.

But based on that definition, and based on the evidence of the comment sections of various websites, then more than half the people posting online are not thinking. (And that may be a generous statistic. You can Google Sturgeon’s Law for a less optimistic assessment.) Too many people are just running tapes—canned responses. Automatic reflexes are simple mechanical operations. Press a button, run a program. There’s no thinking involved, just processing.

Thinking is reasoning ability. We see it in dogs, dolphins, chimpanzees, children, and even the occasional congressman—but that level of reasoning ability occurs at a primal level, it’s simple and direct. The higher functions of what we call rationality and sentience demonstrate themselves in profoundly different ways, recognizable but not easily definable.

Intelligence is generally able to recognize intelligence in action—and that may be one of the defining qualities of intelligence. Not every intelligent being can solve a Rubik’s cube or Fermat’s last theorem, but we can still recognize the intelligence at work in those solutions. The next step, actually designing and creating intelligence requires something else, call it meta-intelligence. We get to step back and think about thinking. We get to deconstruct thinking so we have a clear idea of what we want to build.

Notice that I’m specifically avoiding the term “artificial intelligence.” I don't like the term. It's not just inaccurate, it implies a distinction between the authentic and the simulated. It implies that there is such a thing as real intelligence and suggests that synthesized thought is not really thought all. In which case, the whole search for it is a dead end—because what we're really searching for is the real thing: a rationale for rationality, an understanding of the nature of understanding. This particular quest is a fundamental part of the ultimate existential question, “What does it mean to be a human being?”

We know this much: human beings are tool users. (So are chimpanzees, so let’s not get arrogant.) Language is only one of the tools we use, but it is one of the most useful and powerful of all tools because every idea, every concept, everything we design and build begins as a conversation, as an exercise in language. So we can say that language is the primary way we understand things. It is the primary way that we interact with and manipulate and control anything that will listen—especially other people.

From the moment we’re born, as soon as we come aware of the world around us, as fast as we recognize the difference between me and not-me, we start learning how to manipulate and control the part that is not-me. We learn to hold things, examine them, smell them, taste them, rattle them, and throw them on the floor. Very quickly, we learn that if we cry, someone will feed us or clean us or just keep us company. We learn to manipulate on the most essential levels. And just as quickly as we learn to recognize words, we learn that words can be used as very powerful tools. In fact, we assign enormous power to words, eventually believing that the power is in the words, not how we perceive and use them. We call that relationship with language magical thinking.

For the first two decades of our lives, we develop our skills of manipulation and increase the scale of the tools we use to manipulate: fingers, hammers, game controllers, keyboards, cell phones, steering wheels, and of course, words—polite words, vulgar words, big words, sharp words, nasty words, affectionate words—whatever works. And when words fail us, too many of us resort to other “tools.” Like guns and tanks and bombs. A weapon is just a specific-purpose tool. And the purpose of any tool is to expand our ability to manipulate our environment, so a great deal of what we call communication isn’t communication at all. It’s a polite way of saying, “I’m trying to control you.”

I suggest that intelligence starts with self-awareness—enough self-awareness that survival becomes a critical part of its thinking. Survival is about manipulating and controlling the environment so as to guarantee continuation of self. The scale of a being’s ability to consider the impact and consequences of its behavior on itself and its environment is a way to determine its sentience. How big is its thinking?

Mostly, when we talk about intelligent machines, we’re talking about machines with a DWIM function. (Do What I Mean) We’re talking about machines that will understand what we’re saying and then find or create a way to answer the question or produce that result. Such a machine will do more than just understand language, it will understand intention.

This may very well be the essential core of sentience—the awareness that occurs when intelligence looks outside of itself and discovers other beings. Call it empathy. Call it the ability to recognize that another entity is experiencing, feeling, and thinking. Whatever it is, recognizing the aliveness of another is critical to recognizing intention and therefore critical to sentience. And yes, there are human beings who are incapable of this level of being—the technical term for them is sociopath. And we’ve seen way too much of that in human history. That’s why the search for rationality is so critical. As a species, we could be at the threshold of a transformative moment in sentience.

But an intelligence engine will not mirror the consciousness of human beings. It won’t have the same experiences, so it won’t think like a human—and that will be the most exciting part of the adventure, discovering the differences as well as the similarities. That will lead us to the real understanding of sentience.

Thinking machines—intelligence engines—when we finally start building them, are not going to be just a smarter kind of tool. They will be cognitive entities, fully aware of themselves and very likely demonstrating the same curiosity about the universe that humans do. They will likely also want to manipulate their environment so as to guarantee their survival.

Just as we humans learn quickly how to manipulate each other, I expect that any advanced intelligence we create will also develop the same skills. Skynet (or HAL) isn’t going to kill us. It’s going to find ways to use us.

Oh, hell, even before Skynet exists, it’s already put us to work inventing it...

David Gerrold is a Hugo and Nebula award-winning author. He has written more than 50 books, including "The Man Who Folded Himself" and "When HARLIE Was One," as well as hundreds of short stories and articles. His autobiographical story "The Martian Child" was the basis of the 2007 movie starring John Cusack and Amanda Peet. He has also written for television, including episodes of Star Trek, Babylon 5, Twilight Zone, and Land Of The Lost. He is best known for creating tribbles, sleestaks, and Chtorrans. In his spare time, he redesigns his website, www.gerrold.com

Future Tense: Alan, We Hardly Knew Ye

David Gerrold — Thu, 12 Nov 2009 14:30:00 -0600

Alan Turing should have been knighted. He should have been Sir Alan Turing. Instead he was prosecuted for being homosexual and committed suicide in despair. The British government conveniently forgot that Turing was the genius behind the Allies’ code-breaking efforts during WWII. The “Ultra Secret” is generally credited as the single most important advantage the Allied Forces had against the Axis powers, to the point that Eisenhower was sometimes reading Hitler’s mail even before Hitler.

Fifty-five years after Turing’s death, in response to an Internet campaign, the British government finally got around to acknowledging Alan Turing’s contributions and apologizing for its failure to honor him appropriately. Sorry, guys, but an apology does not erase an egregious wrong.

Alan Turing laid the foundation of modern computing—programmable machines, the separation of software from hardware—the algorithm. Every CPU that you use, your phone, your netbook, your laptop, your desktop, your Xbox or Wii—every single one is a kind of “Turing machine.” It uses a program to process information.

Alan Turing, the father of modern computing

A computer program is nothing more than a list of instructions that the computer automatically follows. The computer does exactly what each instruction demands, nothing more, nothing less. There is no DWIM (Do What I Mean) instruction.

Even though today’s machines seem to have gained an extraordinary level of “intelligence,” the truth is they’re still just following orders, processing lists of instructions—larger instructions, and a lot more of them, and they’re doing it a thousand times faster than thirty years ago—but it’s still the same essential process, diddling ones and zeros.

Alan Turing was no doubt aware of the speed limitations imposed by computers that used electrical relays. Had he not died at the age of 42 in 1954, he most certainly would have lived long enough to marvel at the power of an Apple ][ or a TRS-80. Alan Turing should have had a chance to see the technology that he helped create.

I’m sure he would have laughed to have seen “ELIZA,” a primitive computer program that mimicked human interaction. ELIZA was a simple set of string-processing rules. The program picked out key words from sentences and respond, “Tell me more about <keyword>.” But it had a finite set of responses: “How do you feel about <keyword>?” and if it couldn’t find a keyword: “Why do you say that?” Or simply, “I see.”

Interacting with ELIZA is enough like a real conversation to be startling, at least until you began to sense the underlying algorithms. But what if an ELIZA-like program were capable of very sophisticated conversation, rationally dissecting ideas, introducing new thoughts, and making connections that are outside the specifics of what you’ve typed? At what point do the underlying algorithms disappear so completely that you feel you’re conversing with a real person?

A sample conversation with ELIZA

Alan Turing postulated that when you couldn’t tell the difference between a real person and a machine, the machine was thinking. We call that the Turing test and while it’s not the last word on machine intelligence, it continues to be a good place to start the conversation.

No software written has passed the Turing test yet—and based on the evidence quite a few people posting their thoughts on the Internet probably couldn’t either. (But that’s a different rant. The Internet has already disproved that old saying about a million monkeys typing at a million keyboards for a million years would reproduce the works of Shakespeare. LOL cats, maybe. But Shakespeare, no.)

Many people believe that the development of a genuine intelligence engine is very likely the next step in the evolution of the mind. Not necessarily the human mind, but the idea of mind itself.

Myself, I suspect that there’s a big clue to be found in the Turing test, in the idea of conversation, because Alan Turing seems to have hit on an existential truth—that our thinking exists in our conversations. Even more profound, we define who we are in our conversations about ourselves. (Answer this question: "Who are you?" That's your definition of yourself, that's your conversation of your place in the world.)

PhilosopherJohn Searle's Chinese Room argument

Some contemporary philosophers assert that language is where the mind occurs--language gives us the ability to conceptualize elements of our environment and manipulate the concepts from a specific perspective.

So, by that definition, intelligence is not simply language processing ability—language itself creates a specific kind of intelligence, allowing the mind to process possibilities without having to have actual physical objects to manipulate. Language also allows us to store, transmit, and share intelligence.

Yes, our computers are capable of storing, transmitting and sharing large amounts of raw information. Our computers can manipulate that information by whatever rules we give them--but the gap between that and actual intelligence based on conceptualizing and understanding remains the difficult and elusive part of the problem. It remains one of the most challenging tasks in computer science.

There's still a lot of work to be done. We may still be one or a dozen breakthroughs away from true machine intelligence. Or to put it another way … the road to HAL is paved with good inventions.

Interview With the Award Winning Chatbot AI

Alex Castle — Tue, 14 Oct 2008 14:33:34 -0500

Image Credit: www.elbot.com

Having apparently run out of actual people to talk to, New Scientist has posted an interview with Elbot, the chatbot that won this year’s Loebner Prize for artificial intelligence. Structured after the Turing Test, the prize is awarded to whichever bot can fool the most of the 12 judges into thinking that it’s a real person.

Elbot successfully convinced three judges that it was not a chatbot, but rather a human being pretending to be a robot.

Confused? Check out this excerpt from the interview:

New Scientist: You and your creator won $3000 of prize money. How do you plan to use the money?

Elbot: As I always say, it’s hard to keep a 600-pound robot down, unless you use gravity.

With natural, sensible dialogue like that, I don’t know how any of the judges could have not been fooled. On that note, if anyone is in need of a quick buck, we suggest entering a chatbot next year that pretends to be a man banging his head against the keyboard.

Anyone who wants to can chat with Elbot; give it a try and tell us what you think after the break.

Intel Executive Thinks The Lines Between Human and Machine Will Blur by 2012

Justin Kerr — Sat, 02 Aug 2008 16:23:24 -0500

Intel just passed its 40^th anniversary and the nostalgic occasion had CTO Justin Rattner musing about the future of technology. He foresees new breakthroughs in medical technologies, specifically with regards to nanoscale chips capable of moving through our bodies. Additionally, he predicts more realistic robotic intelligence, and a blurring of reality between humans and machines. Chuckle if you may, but in his 35 years at Intel, Rattner has witnessed some pretty amazing advances in technology, many of which Intel was at the forefront of. When the microprocessors first debuted in 1971 they contained about 2,300 transistors. It has since ballooned to over 820 million and the personal computer has become ubiquitous in our everyday lives. If Moore’s law holds true, and we have no reason to think otherwise, the future may indeed be a very different reality from what we understand today. According to Rattner, “In the next 40 years, computer chips will extend beyond our computers and phones, as people want to become more entrenched in virtual worlds and computers learn to react to our motions and thoughts.”

So what do you think the future holds?