Unless you've been completely out of the tech loop for years, you've probably heard about deep learning. It's associated with machine learning and depends on algorithms inspired by the structure and function of the human brain.
Some forward-thinking chatbot engineers are now looking at ways to tie together the power of deep learning to make chatbots more sensitive.
Chatbots Help People Overcome the Fear of Being Judged
As people who've tried online dating know, the nervousness of being judged by another person is naturally much more strong during a real-life date than the dialogues that happen on messaging platforms.
Individuals often find it easier to be themselves and stop nerve-wracking about what others might think when they're chatting online instead of face to face.
That's the concept behind the Woebot, a chatbot therapist that checks on patients daily. The Woebot deduces data about a person's mood and processes text messages or emoticons entered into the chat window.
Enthusiasts of the technology believe it'll help people overcome the stigma associated with mental illness that could cause them to resist seeking treatment.
Some Pitfalls Illuminate Areas for Improvement
If you've ever had a conversation with a chatbot, you might have come away from it feeling like you got answers to your questions, but the exchange itself was underwhelming.
Researchers at Facebook determined there are predominant downsides making that the case. First, chatbots do not show steadiness in their personalities. That quality makes them seem one-dimensional and makes it more obvious they are not humans.
Also, Facebook's team realized most chatbots couldn't remember what the humans they interact with have said in previous conversations. That is another downside.
Plus, when a chatbot encounters a question that's outside the scope of its programming, it tends to respond with canned-sounding responses, such as "I can't help you with that or I don't know.
Like chatbots, speech-recognition technology has experienced some issues that make it less than perfect. For example, it might misunderstand words, have trouble deciphering slang words and not work well for people with strong accents.
However, many people find it works faster than typing, especially after they train the technology and decide it's worth dealing with the negatives.
In Facebook's case, it is looking at ways to make improvements. For example, some developers train their chatbots with movie scripts.
You can imagine why that input does not always result in authentic conversations. In contrast, people at the social media company built a database of more than 160,000 lines of dialogue.
The material came from Amazon's Mechanical Turk service, which recruits human workers to perform simple tasks. Participants created personas used to inspire the dialogue by coming up with five biographical statements and writing conversations around those bits of information.
So far, the persona-based chatbots show more personality than those trained with movie scripts.
Building Chatbots With Emotional IQ
Another thing that often lets humans know they're chatting with bots is a lack of emotional intelligence. However, a research team at Tsinghua University in China made a chatbot that recognizes emotions and gives appropriate responses.
The researchers relied on a dataset of 23,000 phrases found on a Chinese blogging platform, then tagged each one with an emotional charge. The result was used for a deep learning algorithm that classifies sentences according to the emotions displayed within.
The project features two parts. First, an analysis of a conversation with a human occurs to detect the sentiment. Next, the chatbot uses that information to come up with a relevant answer that's also emotionally appropriate.
Researchers believe they're the only group currently doing this kind of work. However, it's easy to see why it could be so valuable as chatbots increasingly become part of the public landscape.
Many companies already use chatbots for customer service tasks, but a technology that doesn't have emotional detection capabilities might miss when humans become upset. As a result, people could become frustrated or even feel disgusted with a company.
As chatbots become emotionally in tune thanks to the deep learning studies above, people could have more fun chatting with them. They might still be able to tell they're not communicating with humans in some cases, but the overall conversations should be more engaging.
We've just taken a fascinating look at chatbot improvements in the works or being used now. How might they affect your future online interactions? Only time will tell.