Long term planning + teamwork = chatbots
Posted: 4 October 2017 | By Hazel Tang
Often when we encounter a chatbot and manage to engage with it meaningfully, we will walk away impressed with its conversational capability.
Therefore, it is not surprising for businesses to start thinking about the benefits these chatbots may bring when they launch one. But the truth is that it can take rather a long time, depending on how you do it.
“A chatbot may learn a few texts (i.e. your name and demographics) that are local to your conversation but it will not learn anything from users and share with others,” said Dr. Richard Wallace, inventor of Artificial Intelligence Markup Language (AIML) and creator of chatbot ALICE (three times winner of the Loebner Prize contest).
Thus, it is the chatbot developer’s duty to spend years manually writing and growing the content knowledge base of these bots, to arrive at the kind of conversational ability which mesmerizes audiences.
“And it’s a very slow and painstaking effort,” Dr. Wallace told Access AI.
A typical chatbot developer will wake up every morning to go through the conversation box which came into the AI the night before, and refine the content accordingly, so that a specific response is created based on a specific input. A process known as continuous data testing.
As this is an ongoing effort without shortcut, most of the time businesses will have to hire a team of full time staff to oversee adding or deleting knowledge to the bot daily. Especially if they are keen to adopt the technology quickly.
Moreover, there is a need for developers to engage with others who have specific domain knowledge if the chatbot is to be utilized in a specific area (i.e. real estate).
There is a likelihood that the chatbot will start giving nonsensical replies if you are trying to confine something in a story form
On top of the challenges stemming from content creation, current chatbots are also unable to respond to more than one sentence input at a time. There is a likelihood that the chatbot will start giving nonsensical replies if you are trying to confine something in a story form.
Take an example of a typical customer phone call: “Hi, I just bought a new phone from the store yesterday. Then I took it home, plugged it in, and charged it for the first time. But after that I lost it and re-plugged it in again, it didn’t charge the second time, could you tell me what to do?”
According to Dr. Wallace, this kind of story, stringing several sentences together, is where humans really distinguish themselves from machines, as they are able to build up a knowledge representation based on all the inputs.
However, our current chatbots, which function in accordance to grammar and sentence structure, are unable to process this information coherently as a whole.
Slow and steady will win the race – and save your business
Although interesting research is underway, Dr. Wallace believes that we are still at the preliminary stage. Furthermore, for chatbots to be fully commercialized will mean an added issue of liability.
Enterprises will never want their bots to go around saying inappropriate or insulting things, because they are brand ambassadors.
For chatbots to be fully commercialized will mean an added issue of liability
Because of this unpredictability and the need for a relatively large investment, most businesses, particularly smaller firms, remained conservative when it comes to a full-fledged employment of AI.
Hence, Dr. Wallace’s advice is to think of the long term and budget at a sustainable level.
Also, Dr. Wallace does not advocate the practice of letting a chatbot loose on the internet, for users to supply it with content knowledge. This is due to the high tendency of users introducing abusive information or tricking the chatbot into saying scandalous comments. As in the case of “Tay” created by Microsoft.
“The person who is in charge of the bot content will have to spend a significant amount of their time, deleting [inappropriate] responses from the bot’s knowledge base, making the supposed creative writing process into an editorial one” Dr. Wallace said.
The journey of 1000 miles begins with a single step (and teamwork)
Nevertheless, Dr. Wallace keeps an open mind by making parts of AIML and Alice software open source, as he believes it will contribute to the wide spread market adoption and larger market share.
In fact, that was how Dr. Wallace’s collaboration with Steve Worswick started. Steve Worswisk is the maker of Mitsuku, the chatbot which clinched first place in the recent Loebner Prize.
“I think there is a lot more we can accomplish with chatbots, if there is team work. Developing chatbots is not so much of an individual endeavour, it depends on people working together,” Dr Wallace said.