Every year attention of thousands of nerds and tech enthusiasts is getting focused on the one peculiar event. It is related to the most intelligent chatbot phenomenon. Since 2014 it has taken place in England (Bletchley Park) and it’s ruled by four top technology minds, who are there to judge.
As you may have understood, it’s a competition. The main purpose of taking part in it is craving to win Loebner Prize and to become noticed, of course. However, this contest differs a lot from any other related tech contests you may have heard of. What makes it different? The character of competitors. They are the most intelligent chatbots!
If you wonder why somebody would want to make chatbots compete, or why those smart technology guys may want to waste their time, then you have missed a lot about this venue. First and foremost purpose is built-in Artificial Intelligence that makes chatbots so interesting and this competition so important for the tech industry.
What do you know about Alan Turing and his test? Even if you know nothing about it from the tech side, you may be able to recognize familiar words by the famous movie that is called “Imitation Game”. Turing was a famous mind who was able to build a powerful computer for cryptanalysis (decoding) of nazism’ Enigma. However, he wasn’t just prominent computer scientist who contributed a lot in developing the world the way it’s but created a pass for way further development.
One of his most well-known developments was Turing test. It has been broadly used for years since 1950 to detect human-like intelligence in artificial agents. Later on, in 1990 it was used as a base for creating the contest for Loebner Prize.
The idea and the application of the Loebner competition remind the original Turing test a lot. Four judges are there to distinguish chatbots from human beings by means of conversation. When the main purpose is reached and, therefore, the prize of $100 000 is taken, the competition will be over.
The day when intelligent chatbot comes up totally indistinguishable from humans so it can successfully mislead all the judges, the inventor gains the sum of total Loebner Prize. And the presence of Artificial Intelligence will be officially acknowledged. However, until now each year differs from other ones by the sum the prizes (for gold, silver and bronze places) represent and the level of true intelligence the “winner” (chatbot) possesses.
Every time the contest is held society falls apart into the two cohorts. The first one is looking forward to seeing Loebner Prize being completely taken and, thus, true Artificial Intelligence being born. Another one hopes for some delay. So that the competition remains without the main winner and those tech geeks have some time to invent it.
Even though chatbots still are quite distinguishable from human “competitors”, it’s obvious that every year we’re closer to the high purpose of developing the most intelligent chatbot. However, the tech world is too sparse and the data scope is too massive to be able to trace its advancements only by means of media tools. To handle this and to stay on the pulse of the last developments in this venue Loebner competition exists. That’s why it’s so valuable.
In 2016 the first place was won by Mitsuku, the chatbot created by Steve Worswick with the help of AIML. It was its second time of winning Loebner prize (it also gained the gold medal in 2013). Earlier on, in 2015 and 2014, the first place was taken by chatbot “Rose” developed by Bruce Wilcox.
What the most intelligent chatbot should be like
I ask you from now on to think about chatbot in terms of conversational partner – you can imagine language exchange buddy or customer service assistant or anyone else you’ve got to chat with but don’t think about it as if it’s just a piece of software. Because it’s not anymore.
- Image, speech, emotion recognition
First, let’s think about why you would need your the most intelligent chatbot to be able to recognize images and natural language. Just imagine talking to someone sharing photos, recording voice messages and getting no response or just an inadequate one. What would you feel like? Frustrated, irritated or simply like wasting your time? Quite normal to think and feel like that. You don’t need to be patient with software.
But what if your buddy you’re chatting with appreciates what you’re sharing and finally you feel active involvement from his (her) side? Would you feel good? I bet you would. Then, at some point, you may find out that it’s an artificial one. You would laugh but you wouldn’t get irritated.
Image and speech recognition is a must for any intelligent chatbot if it’s meant to appear for users as something more than just automated response. If you’re to create a chatbot you need to consider the services of making your bot understand a variety of responses at the first place. And if image and speech recognition is a base for conversational user interface then emotion recognition is something that still comes up an impressive feature.
To apply image and speech recognition in your application try such services like IBM’s Watson, Wit.ai, Microsoft Cognitive Services, Amazon’s Alexa, TensorFlow. In fact, there’re plenty of them emerging right now, and a slight research can expose a lot of options more or less innovative, convenient, easy to use and up-to-date. However, those one mentioned above are probably the most famous due to the loud names of the companies.
However, emotion recognition isn’t anymore something that comes from the futuristic perspective of our geek fellows. Yes, it’s undergoing continuous advancements but it already works and it applicable in every chatbot application. If you’re still in doubt about how real it is, I recommend you to try it by yourself (it’s always more helpful than just trusting what others say): Affective, Visage Technologies, nViso
- Self-learning ability
This feature is different from the image, speech and emotion recognition. The latter implies is basically relies upon learning ability (Machine Learning). However, usually, it means that you feed your chatbot with those inputs training it and bringing to the point where it can adequately respond to subsequent inputs from users. However, when it comes to self-learning it becomes more difficult as the machine (chatbot) is actively learning from inputs by itself. You don’t need to do anything with it by yourself, however to certain extent control is needed. I will give you a few examples.
Consider Xiaoice that is the artificial software system that lives in WeChat (the most popular Chinese messenger) and created in collaboration with Microsoft. Thousands of Chinese consider it as more than virtual agent but as conversational agent worth spending your time by chatting with. She synthesizes the responses in real time depending on the inputs from users (messages). Just like a real person! Who knows, maybe one-day Xiaoice will jump at the next stage of artificial intelligence regarding that it has already entered self-learning level.
However, not all the most intelligent chatbot examples look so positive like Xiaoice. There’s another popular one that tech society is used to come up with arguing possible risks of developing such self-learning system – Tay. It was also built by Microsoft and run on Twitter. After less than 24 hours she turned racist and sexist and started to publicly offend people. It’s considered that reason for this lies in multiple attacks of the trolls. Therefore, her self-learning ability has turned out to be defective as she wasn’t able to filter what she learned from.
We all love when the conversation flows naturally. Therefore, we want it to be consistent. It’s quite a complicated issue that becomes possible with all the featured above issues – speech, image and emotion recognition, various learning abilities.
Nevertheless, sometimes even plain chatbots based on the hard-coded logic can maintain consistent conversation flow. However, usually, it’s possible only within a narrow range of topics and, therefore, user inputs. When the conversation steps out this range, chatbot grows incapable of adequate response.
Undoubtedly, being able to get back to the previous conversations is a must for a chatbot. Just imagine how awkward it can appear when you find out that your conversation buddy has no idea on what your preceding conversations were about. Isn’t it something extremely disappointing? Don’t think that with chatbots it may be different. I hope you remember that we agreed on treating them in terms of sterling conversation partners.
- Ability to express emotions
Well, the emotional response is something that is quite complicated. Even though emotion recognition is already quite a real stuff, the expression is a black corner. Definitely, it’s the area for further advancements as it’s not clear enough how the machine can really feel anything in order not to imitate this process but to make the machine naturally express those emotions humans are capable of. Mentioning it here has two purposes: to point at the priorities in further developments and review current alternatives.
The latter ones are about making conversations natural by picking the right language. It shouldn’t either sound machine-like, synthetic or draw attention to weird phrases. Instead, it has to give a feeling of lightness and naturalness. That’s why it’s so important for the most intelligent chatbot to possess the language that sounds familiar to the audience it gets in touch with. At least until chatbots become capable of experiencing emotions and expressing them.
The most notable chatbot examples and CUIs
- Amazon Alexa
The first place in this list I had to give to Azuma Hikari. It is inevitable as it embodies the complicated system that contains functions of conversational interfaces like Alexa and chatbot systems like Xiaoice. It’s totally innovative and can become a huge breakthrough in the world of intelligent conversational interfaces.
Cortana, Siri, and Alexa are rather conversational interfaces while Xiaoice, Mitsuku, and Zo are chatbots. Each of them has various features that make interactions with them extremely natural though in different ways. I highly recommend you to check each of them in order to to get the clearer understanding of how their work and what they serve And perhaps to get inspiration for building your own virtual agent.
The current problem of chatbots venue users face today is about expectations that go far beyond what one or another chatbot is really meant to serve for. You need to remember that not each of the most intelligent chatbot you come across on Facebook or KIK was invented to conduct a human-like conversation (even though it’s desirable). Some of them exist only to make a search for needed product easier and faster.
So there’s no need to expect from them something highly natural and exquisite. They are just handling the narrow range of the tasks they’re meant to do. At the same time, there’s completely different cohort of chatbots whose primary job is to sound natural and, therefore, create comfortable conditions for users. So don’t be confused by the word chatbot and stay aware of that they’re different. Just as we, humans, are.
The better we understand the world we live in – the higher becomes the possibility that our contributions to its development will yield better world for the next populations.
Today most of the CUIs are based on the hard-coded logic with the elements of NLP, machine learning, computer vision and speech recognition
Let’s trace together with the way from the simple device (where we all started) to a smart virtual assistant.