Tay, an AI chatbot that learns from previous interaction, caused major controversy due to it being targeted by internet trolls on Twitter. The bot was exploited, and after 16 hours began to send extremely offensive Tweets to users. This suggests that although the bot learnt effectively from experience, adequate protection was not put in place to prevent misuse.
In one particularly striking example of how this rather limited bot has made a major impact, U-Report sent a poll to users in Liberia about whether teachers were coercing students into sex in exchange for better grades. Approximately 86% of the 13,000 Liberian children U-Report polled responded that their teachers were engaged in this despicable practice, which resulted in a collaborative project between UNICEF and Liberia’s Minister of Education to put an end to it.
1. AI-based: these ones really rely on training and are fairly complicated to set up. You train the chatbot to understand specific topics and tell your users which topics your chatbot can engage with. AI chatbots require all sorts of fall back and intent training. For example, let’s say you built a doctor chatbot (off the top of my head because I am working on one at the moment), it would have to understand that “i have a headache” and “got a headache” and “my head hurts” are the same intent. The user is free to engage and the chatbot has to pick things up.
I will not go into the details of extracting each feature value here. It can be referred from the documentation of rasa-core link that I provided above. So, assuming we extracted all the required feature values from the sample conversations in the required format, we can then train an AI model like LSTM followed by softmax to predict the next_action. Referring to the above figure, this is what the ‘dialogue management’ component does. Why LSTM is more appropriate? — As mentioned above, we want our model to be context aware and look back into the conversational history to predict the next_action. This is akin to a time-series model (pls see my other LSTM-Time series article) and hence can be best captured in the memory state of the LSTM model. The amount of conversational history we want to look back can be a configurable hyper-parameter to the model.
“Utility gets something done following a prompt. At a higher level the more entertainment-related chatbots are able to answer all questions and get things done. Siri and Cortana you can have small talk with, as well as getting things done, so they are much harder to build. They took years and years of giant company’s efforts. Different companies that don’t have those resources, like Facebook, will build more constrained utility bots.”
A chatbot works in a couple of ways: set guidelines and machine learning. A chatbot that functions with a set of guidelines in place is limited in its conversation. It can only respond to a set number of requests and vocabulary, and is only as intelligent as its programming code. An example of a limited bot is an automated banking bot that asks the caller some questions to understand what the caller wants done. The bot would make a command like “Please tell me what I can do for you by saying account balances, account transfer, or bill payment.” If the customer responds with "credit card balance," the bot would not understand the request and would proceed to either repeat the command or transfer the caller to a human assistant.
Despite the fact that ALICE relies on such an old codebase, the bot offers users a remarkably accurate conversational experience. Of course, no bot is perfect, especially one that’s old enough to legally drink in the U.S. if only it had a physical form. ALICE, like many contemporary bots, struggles with the nuances of some questions and returns a mixture of inadvertently postmodern answers and statements that suggest ALICE has greater self-awareness for which we might give the agent credit.
Smart chatbots rely on artificial intelligence when they communicate with users. Instead of pre-prepared answers, the robot responds with adequate suggestions on the topic. In addition, all the words said by the customers are recorded for later processing. However, the Forrester report “The State of Chatbots” points out that artificial intelligence is not a magic and is not yet ready to produce marvelous experiences for users on its own. On the contrary, it requires a huge work:
In 2000 a chatbot built using this approach was in the news for passing the “Turing test”, built by John Denning and colleagues. It was built to emulate the replies of a 13 year old boy from Ukraine (broken English and all). I met with John in 2015 and he made no false pretenses about the internal workings of this automaton. It may have been “brute force” but it proved a point: parts of a conversation can be made to appear “natural” using a sufficiently large definition of patterns. It proved Alan Turing’s assertion, that this question of a machine fooling humans was “meaningless”.
Logging. Log user conversations with the bot, including the underlying performance metrics and any errors. These logs will prove invaluable for debugging issues, understanding user interactions, and improving the system. Different data stores might be appropriate for different types of logs. For example, consider Application Insights for web logs, Cosmos DB for conversations, and Azure Storage for large payloads. See Write directly to Azure Storage.
Companies most likely to be supporting bots operate in the health, communications and banking industries, with informational bots garnering the majority of attention. However, challenges still abound, even among bot supporters, with lack of skilled talent to develop and work with bots cited as a challenge in implementing solutions, followed by deployment and acquisition costs, as well as data privacy and security.
Our team of IT marketing professionals and digital enthusiasts are passionate about semantic technology and cognitive computing and how it will transform our world. We’ll keep you posted on the latest Expert System products, solutions and services, and share the most interesting information on semantics, cognitive computing and AI from around the web, and from our rich library of white papers, customer case studies and more.
One pertinent field of AI research is natural language processing. Usually, weak AI fields employ specialized software or programming languages created specifically for the narrow function required. For example, A.L.I.C.E. uses a markup language called AIML, which is specific to its function as a conversational agent, and has since been adopted by various other developers of, so called, Alicebots. Nevertheless, A.L.I.C.E. is still purely based on pattern matching techniques without any reasoning capabilities, the same technique ELIZA was using back in 1966. This is not strong AI, which would require sapience and logical reasoning abilities.