"Hello. What appears to be your problem?"

“Hello. What appears to be your problem?”

"Hello. What appears to be your problem?"

Chatbots and RolePlaying. A conversation with Eran Hadas

ELIZA is an “ancient” chatbot, one that simulates the responses of a psychologist talking with a client. From the conversation above, which I held with a Hebrew version of ELIZA, you can see that it is not particularly sophisticated, but when the original version was 1st created, it passed the Turing Test – that is, it convinced users they were chatting with a human and not a bot. Today’s chatbots don’t always try to pass the Turing Test – they are here to provide very specific services with the conversational interface defining the type of relationship to be established with the user.


“There are two things that make this technology unique,” explains Eran Hadas, who translated ELIZA into Hebrew for the Bloomfield Science Museum in Jerusalem, and who also invented the Hebrew term for chatbots, botpetanim. “The first is practical – it allows a saving in manpower by creating an interactive experience (in education, art or commerce) through software alone. The second is psychological, what we call the ‘Eliza effect,’ named for the 1st chatbot created by Joseph Weizenbaum. It turns out that, when we are being spoken to through text that appears or feels as though it is directed at us, we are psychologically prepared to abandon the distinction between man and machine, and we pay attention to the computer’s text, as though it were a human being talking to us. This allows greater engagement with the subject matter.”


This is very relevant for educational chatbots, explains Hadas: “When a student talks with a teacher, there is a greater level of involvement, in comparison with simply reading content material. If a student is talking to a bot, but psychologically relating to it as though it were a quasi-human conversation partner, that’s where the advantage comes in. Bots can be available 24/7 and they don’t get tired. Also, there is something that arouses curiosity, when we know that there is no person on the other side. We want to test the limits of the bot, to repeat things and see if the response is the same, or to try to get it to do something funny. The conversation becomes a kind of game, which is a preferred method of learning.”

For the full Magazine >> Click Here

So it encourages students to go a little crazy, in both a positive and a negative sense?

“I think that today, there is a certain amount of appreciation for hacking, for attempting to uncover the infrastructure. In learning, many people talk about understanding the process, rather than memorizing dry facts; this ‘going crazy’ is exactly the same thing. The desire and curiosity to understand how things work from the inside. On the other hand, this is a relatively closed computer system. Of course, with a chatbot such as ELIZA, which reflects back to the user what he himself has written, if you type in a coarse expression, you will get one back. Chatbots can filter words and even moods. The chatbot is, in fact, an interface; you could present a search engine as a chatbot, where instead of writing search terms, you could ask a question in natural language, and the response will be the outcome of the search, but in the form of a natural language answer. The same applies to knowledge bases. A significant portion of the chatbots today are simply collections of questions and answers; the bot compares the text entered into it with all the questions, ‘decides’ what the user’s intention is, and then responds to the question that closest matches it. As the technology advances, bots will have the opportunity to choose how open or rigid they will be.”


Hadas teaches “Tools for Web Interactive StoryTelling,” digital media track at Tel Aviv University about ways to tell stories that are unique to the internet age and taught computational literature at the California Institute of Technology (CalTech). “One of the things pushing chatbot technology, particularly in a world that is still feeling its way with the worlds of virtual and augmented reality, is the ability that bots offer in the context of role playing,” says Hadas. “When you give a person a situation, and he has to take part in a conversation, then that person enters into a character – the highest level of involvement. Of course, in the world of virtual reality, the bots can be represented by characters (either imaginary or perhaps real or quasi-real), which allows the immersion of the user within the situation. One can learn about a historical battle by participating in it and conversing with characters from that time and place.”


Hadas has created chatbots for artistic purposes in the Turing Girls group, Batt-Girl and Deganit Elyakim; created Lizetush, a female chatbot that converses with people in internet slang using a unique internet language that makes use of specialized expressions and graphic symbols that replaces letters and decorates the text; with Maayan Shalef and Gal Eshel he created Frankie, a female documentary robot that interviews people to learn and understand what it means to be human.


“Today there is still a prejudice in relation to chatbots, understood as operating purely on the basis of a rigid collection of instructions. For the most part, we are talking about the identification of key words and the operation of appropriate rules to get to the answer”. “Today’s technology is actually more advanced. Alongside statistical methods, looking for similarities between texts, or learning from past examples as to what would be a popular answer to a particular question, the neural network trend is also moving towards the field of natural language processing (NLP) and particularly that of chatbots.”


“The big breakthrough in deep learning networks is actually in the area of image processing, but the motivation still exists in the field of text. Today an enormous amount of research is being carried out in relation to text. The basic architecture of networks that provide a response to problems in the field of text is what is called Recurrent Neural Networks (RNN). What makes this kind of network is that it takes in a sequence of elements (letters, words) and returns another sequence based on a very large number of samples. The challenge of chatbots is very much similar to the challenge of automatic translation, or to any task that takes in a sequence of words and has to return another sequence of words. The need for enormous investment in the field therefore becomes clear.”


“In terms of the significance of the results, it may reasonably be assumed that in the foreseeable future we will get to a situation in which we will not be able to understand what rules have been used by the network” claims Hadas. “Today we already have difficulty in following the logic and rules used in artificial intelligence. As Prof. David Weinberger, a senior researcher at Harvard University’s Berkman Klein Center, said at MindCET’s Shaping the Future 4 conference, ‘It’s troubling that we have machines that are making moral decisions, or decisions that have moral consequences, in ways that we cannot question or interrogate.’ And when these machines are talking regularly with children at critical stages of their development, the shaping of their consciousness and world view – then God, or the deus ex machina, help us.”

For the full Magazine >> Click Here