Twyla AI
Published in

Twyla AI

6 Common misconceptions about chatbots

Not everyone I meet in my daily life has heard about chatbots yet. That’s okay. But I’ve noticed that even those with professional experience or an understanding of AI/Virtual Assistants have some surprising misconceptions about certain points. Time to straighten some things out.

1. Chatbots can speak freely

A surprisingly widespread misconception is that chatbots can speak on their own. This would mean they can create language and are therefore creative. This is a myth about artificial intelligence. AI and machine learning are certainly systems that can generate knowledge. But this knowledge is generated based on statistical data. This leads to the conclusion that less frequent sentences or topics are not validated. You can see that in translation software such as google translate. Less common meanings of a word are being translated incorrectly just because there is fewer data to feed the system. Deriving qualitative conclusions from statistics is a dangerous field if we think of fake news and algorithmic propaganda. Until we can trust an algorithmic retweet to give thoughtful answers to our questions, we can’t rely on “the magic AI box” for content creation. The last very famous example I want to mention here is that of the Microsoft bot named Tay. It was exposed to the internet and all its knowledge came from communication with the public. As curious people — along with haters and trolls — saw an opportunity to shitstorm anonymously, Tay turned from an innocent chatbot into one spouting hate speech within 24 hours. Tay had become limited because the frequent hate comments and their repetition reinforced these areas in its knowledge architecture. So remember not to let chatbots collect phrases or process statistical data unsupervised, as this says nothing about the quality of the speech.

-> Chatbots are not creative with language. Everything is scripted, even the phatic “mhm”.

2. Rebuilding a chatbot in another language is as simple as copy-pasting the content into a translator

Of course, you can have a Spanish, English, German and Chinese Chatbot in the same way that you can build your website in these languages to address your customers. Keep in mind that even when it comes to website translations you need to bear with cultural differences. In game design and in the translation industry, we would speak of transcreation (or localisation) rather than translation. A Chinese website would not only look different because they have a different alphabet. It would also it contain location-specific information and above all, different conventions and standards. A job in translating a website cannot be done just by pasting the original English text in translating software. Especially for Chatbots, this is not how it works. Some dialogues would have a completely different structure due to varying rules for politeness in the different languages or different words and colloquial phrases are said. While in some cultures it might be good to ask direct questions, it could be a no-go in others. And these two don’t necessarily need to be languages that are as distant as English and Japanese. Languages from the same language family can vary a lot in these aspects, such as German and English.

-> A bot in another language has to be transcreated with all its dialogue structure.

3. Chatbots are bad

Another misconception is: chatbots are bad. People think they’re taking jobs away from humans and providing bad service. They think companies just want to save money to the detriment of customers and jobs. This is only part of the truth. Yes, money can be saved but that doesn’t mean the service is automatically worse for it. Deploying a chatbot can be a great improvement to customer service because the routine tasks are handled by the bot, leaving only more complex cases to agents. Agents who have more time to dedicate to those cases, now that they’re no longer bound by the solving of standard problems.

-> Chatbots are a means to achieve goals.

4. All you need to do is upload your FAQs and the bot will be able to converse.

Many people think that thanks to AI you can easily upload an FAQ file and the magic of intelligent software will enable Google duplex-like dialogues.

Far from it! FAQs are a good knowledge base for designing interactions, but they don’t replace a well-founded awareness of problematic use cases and the observation of user behaviour.

One should also not underestimate users when they ask a chatbot for advice. They are quite capable of using a search engine to find an article. The chatbot doesn’t need to mimic this function, after all, something else is expected of him: conversation.

-> FAQs are a good raw material among others that you will need to build your chatbot architecture.

5. Once a chatbot is published it can improve itself.

This is a particularly widespread misconception, but it is not too surprising. Since many people believe that chatbots build themselves thanks to AI, they also think this AI would improve chatbots and make them learn from every conversation.

It is true that neural networks can learn on their own. But there is no bot software that allows independent learning from past conversations without some degree of human intervention.

Over time, the bot will better understand its users as it receives more and more word combinations. Then — thanks to human intervention — it connects sentences from users in the chat to the correct answers.

Of course, these answers are usually improved and refined by human intervention.

-> A chatbot can be only improved by humans adding and refining his knowledge.

6. Chatbot developers should design conversations

Who has the skills to create a chatbot conversation? Is it really the developers? First of all, it makes sense that the people who know most about their own business should do it, preferably those who have strong communicative and writing skills.

That’s not necessarily a UX designer, who is closest to a graphic designer and hence, not always skilled in writing.

Copywriters aren’t graphic designers either, so still, there is a difference between a Graphic Interface UX Designer and a Conversational UX Designer. But at least the shift from developers designing conversations to graphic designers working in UX is a constructive step. But if we want to get the best out of the technology, we should seek out the most eloquent people in the content creation team, the ones that have experience in writing for a living or at least — experts at communicating with customers. I would rather trust a customer service agent to build interactions than a developer. You might wonder now — what do customer service agents know about chatbots and UI design? Actually, more than it seems at first glance. They spend their entire day with the customers that will use the chatbot, they are the ones that know their needs and every possible question, their whole workday is filled by thinking only about the customer. Unless you choose to work with tools where the copy and NLP (Natural Language Processing) has to be created with JSON or NODE.JS, you should not need to rely on a developer team solely to create your conversational interactions. This will cost you more as you will build the technology from scratch for each bot and maintenance will bind you to more expenses.

-> Who should do Conversation Design in first place

  • Conversation Designers
  • Copywriters
  • Business owners
  • Customer Service Agents
  • Marketing departments

There may be more misconceptions about AI and virtual assistants out there. Of course, we can’t address them all at once. However, we can expect some to resolve themselves when collective knowledge and experience reach common awareness. No matter what, one thing remains true:

The more we learn, the less we realise we know.

--

--

--

Twyla is pushing forward the thinking of Conversation Design by challenging established concepts and setting new ones. We don’t build Chatbots, We design Conversations.

Recommended from Medium

Case for Standardisation in Radiology Second Opinion

This stem picture depicts how nature sequences its production process, this helps in us going to standardised delivery in any service sector

Your Bot Strategy Is Dead In The Water Without Humans In The Loop

Impact of Advanced Technologies and the Pandemic on the Telehealth Industry

How to Teach AI to Kids…In 30 Minutes

Artificial Intelligence (AI) in Video Marketing

AI vs ML vs DL vs Data Science

Representations for Content Creation, Manipulation, and Animation

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Maggie Jabczynski

Maggie Jabczynski

I am a Linguist and Copywriter with a background in Anthropology & Ethics, working as a Conversation Designer in telecommunications

More from Medium

I Tried ANOTHER Artificial Intelligence Software

Google serendipitously have solved a string towards artificial general intelligence

Simple SpaCy 3. x end to end Custom NER pipeline

Using AI to Identify Automobiles in Hollywood Cinema