IBM’sWatson became the first machine to beat a human in Go, a game that requires reasoning as well as strategy. Earlier, chatbots could only curate a few words to answer questions, but use of machine learning or artificial intelligence has helped bots talk easily by curating questions.
CHATBOTS HAVE BEEN in existence for long, but it is only recently that industries and services are using these to ease their work. That said, one area that has largely remained unexplored as far as the development of bots is concerned is using them for professional help. While this possibility has been talked about ever since the first bot cleared the Turing test—this proves if a bot can fool a human into thinking that she is talking to another human—Samsung’s NEON has centred the focus back on what chatbots can, or can’t, do. More so, as industries are also realising the limitation of bots. With Tokyo Olympics around the corner, the currency of such devices may increase. But that shall also depend on how successfully countries and cities can adopt the Tokyo model, where automation is going to drive the Olympics.
What are chatbots?
The first of the chatbots was created 50 years ago by MIT professor Weizenbaum, and it was called ELIZA. Ever since, chatbots have undergone radical transformations. In 2001, Richard Wallace created the first programme for chatbot. Around the same time, SmartChild—a chatbot available on AOL IM and MSN Messenger—was released.
In 2014, a bot cleared the Turing test, making it a benchmark for chatbots of the future. Since then, each year many bots clear both Loebner and Turing tests. How do they operate? An easy way to understand the function of chatbots is understanding how computer games are played. Ever since the advent of computers, there have been games where one can compete with a computer.
The concept was simple. For instance, in Chess, the computer was given strategies to follow that had been gleaned from on millions of games, and people could play single-player games. A similar logic was followed for card games. But, as technology evolved, companies like IBM were able to create programmes that went beyond playing poker, blackjack and chess with humans.
IBM’sWatson became the first machine to beat a human in Go, a game that requires reasoning as well as strategy. Earlier, chatbots could only curate a few words to answer questions, but use of machine learning or artificial intelligence has helped bots talk easily by curating questions. Mitsuku is one example. The bot is used by millions to fight depression and to discuss things. When it started in 2013, Mitsuku could not hold a conversation for more than a few minutes. Now, it can do more than five minutes, and even longer in some cases. What kinds of chatbots are there? One kind is the likes of Mitsuku, which have either won theLoebnerprize or cleared Turing tests.
Another are service industry bots, which are mostly used by companies to make processes easier and reduce human interface. In such cases, the bots can only perform certain tasks. But, over time, they learn more as they tend to answer more queries and become smarter. Next level of learning One of the basic learning levels for any bot is trying to pick up as much of language and communication as possible. So, the more you talk, the better the bot will become. But, that isn’t always the case, sometimes bots tend to represent biases. For instance, if a bot can only answer a question like “How are you doing?” or “How are things for you?”, it may never be able to answer to a similar question like“How’s life?”.They can only decipher so much from language For this purpose, national language processing becomes very important, where a bot cannot only understand the words, but sentences and framing of sentences.
It is also important for bots to understand intentions of the person asking the question and, for this to happen, there needs to be sufficient data on that person. If, say, one is suicidal, then the bot should be able to adapt and communicate accordingly. But all this requires data, and this is where companies are facing issues. While most consumers consent to giving up data rights in lieu of services, chatbots require much more data for machines to learn from humans. Whether NEON becomes successful in this experiment or not will prove how fast chatbots can advance in the future.