Tag Archives: chatterbots

A.I. and the Mastery of Spoken Language

The question isn’t just whether we are capable of making simulations of human speech but, rather, if bots can replicate the singular mind that gives form to all speech.

In Steven Spielberg’s dystopian film, A. I. Artificial Intelligence, a software designer played by William Hurt explains to a group of younger colleagues that it may be possible to make a robot that can love. He imagines a machine that can learn and use the language of “feelings.” The full design would create a “mecha”—a mechanized robot–nearly indistinguishable from a person. His goal in the short term was to make a test-case of a young boy who could be a replacement for a couple grieving their own child’s extended coma.

The film throws out a lot to consider. There are the stunning Spielberg effects of New York City drowning in ice and water several decades in to the future. But the core focus of the film is the experiment of creating a lifelike robot that could be something more than a “supertoy.”  As the story unfolds, it touches on the familiar subject of the Turing Test: the long-standing challenge to make language-based artificial intelligence that is good enough to be indistinguishable from the real thing.

Should we become attached to a machine packaged as one of us? Even without any intent to deceive, can spoken language be refined with algorithms to leap over the usual trip wires of learning a complex grammar, syntax and vocabulary?  It takes humans years to master their own language.

The long first act of the film lets us see an 11-year old Haley Joel Osment as “David,” effectively ingratiating himself to the Swinton family.  In my classes pondering the effects of A.I., this first segment was enough  to stop the film and ask members what seemed plausible and what looked like wild science fiction. I always hoped to encourage the view that no “bot” could converse in ordinary language with the ease and fluency of a normal kid.  That was my bias, but time has proven me wrong. If anything, David’s reactions were a bit too stiff to reflect the loquacious chatter bots around today. Using Siri, Alexa or IBM’s Watson as simple reference points, it is clear that we now have computer- generated language that has mostly mastered the challenges of formulating everyday speech. There’s no question current examples of synthetic varieties are remarkable.

Here’s an example you can try. I routinely have these short essays “read” back to me by Microsoft Word’s “Read Aloud” bot, which comes in the form of a younger male or female voice that can be activated from the “review” section in the top ribbon. Not having an editor, it helps to hear what I’ve written, often letting me hear garbled prose that my eyes have missed. I recall the first version of this addition to Word was pretty choppy: words piled on words without much of attention to their  intonation, or how they might fit within the arc of a complete sentence. Now the application reads with pauses and inflictions that mostly sound right, especially within the narrower realms of word usage focused on formal rather than idiomatic English.  Here is the second paragraph of this piece as read back to me via this Word function:

Of course, language “means” when it is received and interpreted by a person.  An individual has what artificial Intelligence does not: a personality, likes and dislikes, and a biography tied to a life cycle. Personality develops over time and shapes our intentions. It creates chapters of detail revealing our social and chronological histories as biological creatures. A key question isn’t whether we are capable of making simulations of human speech. And that begs an even bigger question about whether bots can replicate the unique mind within each of us that gives form to human speech.

Even tied to advanced machine-learning software, chatterbots easily use similarity to falsely suggest authenticity. And there’s the rub. Generating speech that implies preferences, complex feelings or emotions makes sense only when there is an implied “I.” For lack of a better word, with Siri or Watson there is no kindred soul at home. The language of a bot is a simulacrum: a copy of a natural artifact, but not a natural artifact itself.

Even so, we should celebrate what we have: machines that can verbalize fluently and–with complex algorithms–might speak to our own unique interests.

Turing, And The Bogus Rivalry With Machine-Based Intelligence

IBM's Watson Wikipedia.org
                     IBM’s Watson           Wikipedia.org

In reality, humans have nothing to fear. Most measures of artificial intelligence use the wrong yardsticks.

We are awash in articles, books and films about the coming age of “singularity:” the point at which machines will supposedly duplicate and surpass human intelligence.  For decades it’s been the stuff of science fiction, reaching perhaps its most eloquent expression in Stanley Kubrick’s 1968 motion picture, 2001: A Space Odyssey.  The film is still a visual marvel. Who would have thought that Strauss waltzes and images of deep space could be so compatible?  Functionally, the waltzes have the effect of juxtaposing the familiar with a hostile void, making the film a surprising celebration of all things earthbound.  But that’s another story.

The central agent in the film is the HAL-9000 computer that begins to turn off the life support of the crew during a long voyage, mostly because it “thinks” the humans aren’t up to the enormous task facing them.

Kubrick’s vision of very smart computers is also evident in the more recent A.I., Artificial Intelligence (2001), a project started just before his death and eventually brought to the screen by Steven Spielberg.  It’s a dystopian nightmare. In the film intelligent “mechas” (mechanical robots) are generally nicer than the humans who created them.  In pleasant Haddonfield New Jersey, of all places, they are shot on sight for sport.

Fantasies of machine intelligence have lately given way to IBM’s “Big Blue” and “Watson,” mega-computers with amazing memories and—with Watson—a stunning speech recognition capability that is filtering down to all kinds of devices.  If we can talk to machines, aren’t we well on our way to singularity?

For one answer consider the Turing Test, the challenge laid down by the World War II code-breaker Alan Turing. A variation of it has been turned into a recurring world competitions.  The challenge is to construct a “chatterbot” that can pass for a human in blind side-by-side “conversations” that include real people.  For artificial intelligence engineers, the trick is to fool a panel of questioners at least 30 percent of the time over 25 minutes. According to the BBC, a recent winner was a computer from the University of Reading in the U.K. It passed itself off as a Ukrainian teen (“Eugene Goostman”) speaking English as a second language.

In actual fact, humans have nothing to fear.  Most measures of “human like” intelligence such as the Turing Test use the wrong yardsticks. These computers are never embodied. The rich information of non-verbal communication is not present, nor can it be.  Proximate human features are not enough.  For example, Watson’s “face” in its famous Jeopardy challenge a few years ago was a set of cheesy electric lights.  Moreover, these smart machines tend to be asked questions that we would ask of Siri or other informational databases.  What they “know” is often defined as a set of facts, not feelings. And, of course, these machines lack what we so readily reveal in our conversations with others: that we have a sense of self, that we have an accumulated biography of life experiences that shape our reactions and dispositions, and that we want to be understood.

Just the issue of selfhood should remind us of the special status that comes from living through dialogue with others.  A sense of self is complicated, but it includes the critical ability to be aware of another’s awareness of who we are.  If this sounds confusing, it isn’t.  This process is central to all but the most perfunctory communication transactions.  As we address others we are usually “reading” their responses in light of what we believe they already have discerned about us.  We triangulate between our  perceptions of who we are, who they are, and what they are thinking about our behavior. This all happens in a flash, forming what is sometimes called “emotional intelligence.”  It’s an ongoing form of self-monitoring that functions to oil the complex relationships. Put this sequence together, and you get a transaction that is full of feedback loops that involve estimates if intention and interest, and—frequently—a general desire born in empathy to protect the feelings of the other.

It’s an understatement to say these transactions are not the stuff of machine-based intelligence, and probably never can be.  We are not computers.  As Walter Isaacson reminds us in his recent book, The Innovators, we are carbon based creatures with chemical and electrical impulses that mix to create unique and idiosyncratic individuals.  This is when the organ of the brain becomes so much more: the biographical homeland of an experience-saturated mind.  With us there is no central processor.  We are not silicon-based. There are the nearly infinite forms of consciousness in a brain with 100-billion neurons with 100-trillion connections.  And because we often “think” in ordinary language, we are so much more—and sometimes less—than an encyclopedia on two legs.