A personal AI chatbot that never misfires? How can we effectively train the artificial mind to act human-like? What are the most common chatbot challenges and how can we overcome them? And most importantly — why do we, humans, need bots? Since 1956 the world’s best engineers are racking their brains to find the answers.
We’ve got used to AI assistance in numerous ways. When we leave our home in the morning to commute to work, we check our phone to see the bus traffic delays, delivered by Google Maps machine learning algorithms. When we start our workday, we use smart reply and smart compose functions on Gmail, accepting AI sentence suggestions and draft our emails faster and easier. When we have to schedule tasks for the week, we use AI tools to automate our to-do lists and organize our schedules. When we start a meeting with a client, we use AI voice assistants that transcribe and highlight meetings notes. Artificial intelligence is all over us, so are the AI chatbots. They aren’t only a magic instrument used for delegating mundane tasks and cutting costs, but also an abundant multiplier that creates new value. Be it a virtual assistant, like Alexa, Siri or Google Assistant, or a charming chatbot friend like Mitsuku or AIBRO, AI-based chatbots are supporting us in various life situations, responding instantly to any our query. In this article, we’ll learn about chatbots past, present, and future and explore the top chatbot challenges.
In 2016 the world’s attention was drawn to the emerging chatbot hype. It was an earthshaking shift in human to machine communication, commemorating the change from point to click interactions to the intelligent conversations with machines. Only in the last couple of years, the neural network chatbots were sufficiently trained on huge datasets so they are capable to provide meaningful responses in chats. Never before they were able to do so.
2016 was a year when Google and Facebook publicly announced they are supporting businesses and individuals willing to build chatbots on their platforms. In 2018 the number of chatbots on Facebook Messenger exceeded 300,000. In 2016 Google DeepMind’s AlphaGo program won in its final game against South Korean Go grandmaster Lee Sedol, celebrating a tipping point of what can be accomplished with AI program. Google Translate kicked off neural networks for machine translation in 9 languages in 2016. Flo, a period tracker, formally announced that neural network has considerably improved cycles predictions thanks to 1.4 million new data points logged every day by women all over the world. 2016 was a year when neural networks and artificial intelligence chatbots have transformed from a sci-fi notion to the real thing.
Chatbots are now an indispensable part of our daily lives, forming new consumer habits and expectations. It’s easy to find a chatbot that can serve any possible function you can think of: be it a food delivery service, banking assistant, university teacher, travel helper. Some of these chatbots offer menu with options, others offer you to enter your questions or query. Some of the chatbots are capable to learn, while others do not.
The chatbots that don’t learn are comparatively easy to develop. They are built on decision-tree logic and their performance is just a matter of how meticulously the script is written. Often menu-based chatbots are quite limited, as they can’t respond to the questions that aren’t in the script. To make our life easier, we, humans, decided to create the chatbots that are capable to learn and can handle any questions we may think of so we won’t bother ourselves with drafting so many huge decision-trees. Such chatbots are called AI chatbots as they use machine learning and neural networks to form an answer. AI chatbots are the youngest type of chatbots, ss artificial intelligence technology was developed at a full-scale not a long time ago. AI chatbots are harder to develop and it takes more time to educate such chatbot, but once it’s done, the chatbot is capable to provide coherent answers to any possible question asked. As the technology is comparatively new, AI chatbots have to face some issues, the most common of which are the following:
1. AI chatbots lack singular personality
A great conversation with chatbot requires harmony — between being simple and elaborate; between sticking to the topic of conversation and switching to another one; between asking questions and providing replies. Chatbots that use machine learning are trained on tons of dialogues with diverse speakers. Sometimes the answers chatbots provide can be inconsistent and contradictory. That becomes especially obvious in multi-turn dialogues. To fix this issue chatbots can use NLI models (NLI stands for natural language inference) that can improve chatbot consistency and enhance semantic adequacy of the chatbot replies.
2. AI chatbots lack explicit long-term memory
Chatbots don’t remember all the past dialogues with one specific person and all his/her viewpoints. Chatbots are trained to yield replies within the latest dialogue history and it’s hard for them to grasp the whole context of the conversation. Sometimes it’s incredibly hard for a chatbot to track all partner’s peculiarities, backgrounds, personalities, speaking styles, and interests. Therefore, chatbots built with the memory-augmented neural networks can respond with more personalized, precise and appealing replies than the ones without. Sometimes chatbot may switch the whole conversation about the smartest person in the world to the joke about roast beef and pea soup, like Cleverbot.
However, deep learning algorithms can suppress context recognition restrictions.
3. AI chatbots respond with unspecific replies
Chatbots reply with “That’s all I have to say about it” which is basically a form of “I don’t know” etc. when they are unfamiliar with the topic. The “I don’t know reply” received from a human is justifiable, as we, humans, aren’t expected to know everything about everything. However, receiving such a reply from a chatbot turns us off. We may see chatbot as just a tool that has specific functions. Therefore the “proactive” bots that can switch the topic of conversation to another subject, etc are valued highly and considered to be “wiser”.
4. AI chatbots are expensive to train
Chatbots that use machine learning techniques require huge, refined and expensive datasets to be trained on. It’s not easy, nor quick. These datasets need to be cleaned and labeled properly. Moreover, chatbots need to learn for lengthy time slots to work well. A lack of sufficient training will affect chatbot performance and can lead to poor results.
Chatbots are here to stay. While we, humans, are restricted by slow biological evolution, artificial intelligence is not. As AI technology is getting stronger and more powerful every year, chatbots have fewer and fewer limits on what they can do. As a result, nowadays humans can communicate with chatbots as they would do with people.
Chatbots proved itself as a valuable technology in different use cases of peoples’ everyday lives, becoming essential not only for personal but also for business use. Unlike apps and websites, chatbots are more convenient and easy to use, as the UI of the messenger apps are well-known.
In many industries, chatbots work as a colleague for company employees. Chatbots are especially valuable for taking care of low cognitive and repetitive work. Thanks to chatbots, we, humans, are free from doing mundane and mechanical work and can focus on highly important, strategic and top-level cognitive level tasks. While you are concentrated on exciting and impressive assignments, or on a detox or a vacation, an intelligent AI chatbot AIBRO can completely simulate your behavior on Facebook Messenger, such as your sense of humor, sensitivity, empathy, and optimism. It protects you from messaging overload by taking care of your conversations and making your life easier and hassle-free.
Lots of businesses are very close to AI chatbot integration in their day to day operations but haven’t yet made the leap. According to MITSloan Management Review three out of four executives believe artificial intelligence would allow them to open the new business ventures and 85% believe AI technology implementation provides a competitive advantage on the market. The same survey discovers that just one out of 20 companies has integrated chatbots in its processes or offerings.
As per Moore’s Law, we expect the speed and capabilities of our computers to grow exponentially and their costs to halve. Indeed, the computing power has doubled every two years from 1959 until 2012, as Moore’s Law states. However, starting from 2012 we’re entering a new computing era, that considerably outperforms macro trends. Nowadays computing power doubles every 3.4 months since 2012, according to OpenAI analysis. Therefore, the computational power used for training the biggest AI models is increasing more than seven times the previous rate. This analysis doesn’t take into account the recent discoveries and breakthroughs, such as BERT, which was open-sourced by Google in November 2018 and is the largest leap forward in the last 5 years, redefining the way we understand the context in the sentence.
In modern AI computing era, chatbots become more mature and polished, ready to anticipate potential user needs. They have great potential and tremendous capacity to grow and deliver results we haven’t ever expected from a bot. Chatbots are gradually becoming an integral part of digital customer service, able to resolve customer queries cheaply, promptly and consistently. According to Reports and Data, the chatbot market was estimated at 1.17 billion USD in 2018 and anticipated to arrive at 10.08 billion USD by 2026, with a 30.9% compound growth annual rate.
Moreover, chatbots provide considerable advantages both for chatbot users and for chatbot owners, opening new opportunities for customer engagement and operational efficiency and creating a new form of doing business. Chatbots are a great example of how we, humans, cast away to the machines the tasks we find boring and tiresome.
Critics may say chatbots are “overhyped”, but they can’t be ignored. Chatbots are the most elaborate and progressive examples of communication between humans and machines. They are capable to do face and voice recognition, real-time translation, text to speech conversion. Chatbots are capable to save hundreds of thousands of human recruitment hours by interviewing prospective candidates, detect fraudsters and suspicious behavior on the world’s largest stock exchange or generate a 40 GB of coherent paragraphs of text while being fed with one single sentence.