Breaking Down the Technology of AI Chatbots
In order to fully leverage AI chatbots, it’s crucial to understand some of the technology used to help it outperform DIYl chatbots. Artificial intelligence helps the bot understand language, and learn over time.
The chatbot receives data and interprets, contextualizes and translates it so that it can provide the appropriate answer when prompted.
At a more granular level, here are some different techniques AI chatbots use to improve performance.
- Machine Learning – The ability for the system to improve functionality based on a variety of algorithms including pattern and text recognition. Over time, as it has more reference data, the machine learns to become more efficient.
- Natural-language Processing – A process that deals with a bot’s ability to analyze language through speech recognition, semantics and syntax. Just like a human learns a language through listening and reading while understanding the context, computers can attain a similar capability.
- Deep Learning – A broader version of machine learning, deep learning is the ability for a computer to process various pieces of information the way a human would to make informed decisions and judgements. Deep learning uses neural networks to prioritize data by assigning a numerical value to each data point or using true/false logic analysis.
- Neural Networks – Neural networks refer to the vast clusters of data within a computer system that leverage their proximity to other related clusters of data to increase each cluster’s ability to learn from the other, much the way the human brain and nervous system do.
- Support Vector Machines (SVM) – SVM technology allows machines to identify optimal solutions when faced with multiple options. Machines are typically fed a small set of data samples to help it ﬁnd an optimal solution.
- Supervised/unsupervised Learning – Machine learning often contains three types of learning: supervised learning, semi-supervised and unsupervised learning. In supervised learning, the machine’s output that it is supposed to learn for understanding rules are given as ground-truth during training. For unsupervised learning, the intended answers, or equivalently machine output, is not provided. Any rules or inferences the machine learns are determined strictly using machine learning algorithms, independent of having been provided the answers beforehand.
While many researchers hold out hope for a completely intelligent chatbot that can talk to a human in the same manner as a live agent, AI technology is not at the point where this desire is feasible. Instead, chatbots are best utilized when there are predetermined topics for which the bot can gain expertise and address before passing more sophisticated topics to a human. However, we can expect that in the coming years, that gap will continue to shrink as bots grow the capacity to handle more complex decision-making capabilities. The History of Chatbots The creation of robots and artiﬁcial intelligence dates back to 1950 when Alan Turing asked a question that led to a paper, known as the Turing Test. It may seem like chatbots have only been around for a few years, but they actually have a relatively long history. The paper proposed a test to determine whether humanity could tell the difference between a human and machine. This question is still performed regularly today as a critical benchmark in understanding the capacity for bot performance. www.ivy.ai 3200 Carbon Pl, Suite 103, Boulder CO 80301 11 Here is a list of other landmark events that helped define where bots are today. ELIZA, 1966: Named after Eliza Doolittle in George Bernard Shaw’s Pygmalion, the ﬁrst bot developed at MIT by Joseph Weizenbaum aimed to fool humans into thinking a psychotherapist was interviewing a patient. The project was ultimately successful in being the ﬁrst machine to use natural language processing.
PARRY, 1972: Psychiatrist Kenneth Colby took ELIZA a step further by creating a more conversational chatbot, that allowed it to converse with ELIZA. PARRY was intended to represent a paranoid schizophrenic and had greater language capabilities compared to ELIZA. JABBERWACKY, 1981-1988: British programmer Rollo Carpenter created the ﬁrst chatbot with a goal of creating artiﬁcial intelligence capable of passing the Turing Test. The bot could emulate natural human chat while being interesting and entertaining. Dr. Sbaitso and ALICE, 1991-1995: Dr. Sbaitso was another psychologist chatbot with a digital voice, powered by AI to show off an impressive range of digitized voices. It could ask questions similar to a therapist, such as “Why do you feel that way?” ALICE was a chatterbot that took the natural language processing ability of ELIZA to a new level. Elbot and Smarterchild, 2000-2005: Elbot was the ﬁrst chatbot to successfully use sarcasm and wit to converse with humans through AI. In 2008, it came close to successfully passing the Turing Test. Smarterchild was a bot for AOL Instant Messenger that offered personalized conversation and could offer information on movies, weather or current events. IBM Watson, 2006-2009: Watson was designed as an AI chatbot meant to compete against humans on Jeopardy using natural language processing and machine learning. It is now arguably one of the smartest systems in the world with use cases in healthcare, weather forecasting, advertising, tax preparation and much more. Watson is used to reveal a variety of insights with a large amount of data. Google, Siri, Alexa and Cortana, 2010-2015: Amazon, Apple, Google and Microsoft launched their ﬁrst voice assistants, which allowed the user’s device of their choice (mobile, PC, tablet) to provide information to answer questions and make recommendations. Alexa was the ﬁrst smart home speaker that enabled these capabilities, in addition to powering a variety of devices through a voice command. Messaging Bots, 2016-present: Facebook launched its ﬁrst messaging platform growing to over 300,000 active chatbots in 2018. It has since captured the imagination of industries like e-commerce, retail and higher education, with Ivy.ai launching an artiﬁcially inteligent self-service chatbot to help students across the country access the answers they need. The future of chatbots holds great promise, both for higher education and society at large. Between advances in technology, a change in student preferences and the COVID-19 pandemic, bots have become a necessity. It’s not a stretch to state that in the not-too-distant future most of our customer service interactions will be carried out by a bot