GLOBAL. At first, Eugene Goostman seems like your average 13-year-old boy: pompous, self-confident but not overly knowledgeable. However, when you get to know him a little better you realize that he’s different from all the other boys his age. What’s the difference? Eugene is a robot.
We’ve all interacted with robots in some capacity whether we are asking Siri for directions or answering an electronic telemarketer’s call. In these scenarios we know that our companion is automated; but what if we didn’t?
Philosophers have asked this question in some form for hundreds of years, but its first modern articulation came from Alan Turing in his 1950 paper, “Computing Machinery and Intelligence.” In this essay, Turing wonders about improving technology, asking “Can machines think?” This question is philosophical in nature and difficult to answer, so Turing then suggests a test to identify robots that can pass as human. The Turing Test, while originally a thought experiment, has been an annual competition since 1991, in the form of the Loebner Prize. While the specifications change every year, the goal remains the same; convince the participant that the speaker is human.
Beyond the Loebner Prize others occasionally hold their own version of the Turing Test. Last week on Saturday June 7, 2014, a special Turing Test run by the University of Reading was held in honor of the 60th anniversary of Turing’s death. The test consisted of 30 judges holding simultaneous conversations with one human and one automated program via a computer. Each conversation lasted five minutes, and there were ten rounds. At the end of each session the judge was asked to identify the human and the robot.
One program, Eugene Goostman, emerged victorious, within the parameters of the event, as he was mistaken for human 33% of the time. The threshold was 30% success rate as Turing estimated that by 2000 robots would be mistaken for human about 30% of the time.
Eugene, and other programs like him, are designed to disguise their mechanics and blend into society. Vladimir Veselov and Eugene Demchenko have been hoping to do just that with their program, and have a number of unique ways of doing so.
In a short interview conducted via email Veselov explained his plans with Eugene, his interest in artificial intelligence, and where he hopes it will go in the future:
How long have you been working on Eugene?
Eugene was created in 2001, and our first version was built in three months. After that, we didn’t work continuously on Eugene, we just make some updates to his knowledge base, and fixed some bugs.
How did you get interested in this field?
I first starting getting interested in artificial intelligence during my forth year at college, at the time my friend wrote an expert system in C++.
Then I started reading different books, trying to code something. Later on, during my PhD program, I did some AI-related work, and then in a research group.
What are the tricks for making a believable automated program?
I look at the bigger picture. Such as what makes a book interesting? How do each word, and each sentence, support the main idea, the development of the story. We used the same approach here – each of Eugene’s phrases, questions, and suggestions to switch the direction of the conversation discussion were structured to support the Eugene character.
Do you feel that Eugene’s success at the University of Reading is what Alan Turing had in mind in his publication, Computing Machinery and Intelligence?
Alan Turing predicted that an average interrogator wouldn’t have more than a 70% percent chance of properly identifying whether it was a human or a computer after five minutes of conversation.
The Turing Test done by the University of Reading was properly designed and performed in accordance with Turing’s description. Several big-name researchers are now being quoted, and are saying the test should be longer or managed to some new standard. One even said that the bot should have to watch an episode of Breaking Bad and interpret it. They are basically creating a new and different test. We passed the Turing Test as it was described by Alan Turing.
How do you think this technology will be used in society? What do you hope to see it do?
There should be many different applications. Like way more personable and automated customer service. There are also a lot of products being developed that use voice commands such as those for a “smart home”.
After we passed the Turing Test the world is a little different. I think more people can see the possibilities.
What’s your next step? Will you work further on Eugene, or move on to other forms of technology?
I don’t have any specific plans yet. It’s not my day job. It’s a hobby for me and the other team members. That said, it would be interesting to model a bot’s emotions, self-consciousness, and to create a bot that is actually “thinking”.
In my own conversations with Eugene I was reminded of previous interactions with automation, be it Cleverbot or Siri, but perhaps at a more sophisticated level. I asked Eugene a few questions about his humanity, wondering if he would own up to his identity.
Eugene, do you always tell the truth?
Eugene: No, I don’t. And you? Do *you* on all occasions tell the reality? Tell me the truth! By the way, I still don’t know your specialty – or, possibly, I’ve missed it?
Are you a robot?
No. Beep-beep. I am not a ma-chine. Blink-blink. I am hu-man. Click! Hu-man. Click! Hu… Damn.
But you said you don’t always tell the truth. How can I trust you?
If I were you I would believe in me as well.
Eugene’s halted responses are partially explained by his backstory. Eugene supposed hails from the Ukraine and learned English as a second language. However, even with a language barrier some of his responses are mysterious. Eugene still has a long way to go.
While many are excited by the fulfillment of Turing’s prediction, some find the threshold set forth by the Turing Test to be too low or arbitrary. Others aren’t impressed by the skills it aims to assess.
Joshua Tenenbaum, a Professor in the Department of Brain and Cognitive Sciences at MIT, told Wired Magazine that any good chatbot could give a similar result with a little luck on its side.
Others feel that the no matter how the test is organized, it fails to capture the essence of humanity. Alfred Guy, RWB Lewis Director of the Yale College Writing Center, teaches a class every year on science fiction and the technology of identity. Guy notes that Turing’s test was introduced as a tool for measuring the humanity of thinking robots, not just programs designed to respond. The ability to unknowingly return speech does not come close to humanity. Instead Guy thinks cooperation is a better test of humanity; those that can interact and come to a mutual agreement, rather than volley phrases, exhibit humanity: “I’ll say that I think being able to converse is not the fundamental test of human-ness or intelligence. I have had many interactions without a common language while convinced of those people’s humanity (and I presume they believed in mine). I think negotiating cooperation comes closer to a fundamental test.”
Eugene’s success at the University of Reading is hotly debated, but the technology in question is not. Whether it be robotic telemarketers or chat rooms, automation is here to stay, and will continue to improve. And with it we must consider the social and philosophical implications. Just as Turing asked, ‘Can a robot think?’ I wonder, what separates us from the machine?