Chatbots are a super hot topic in our industry right now. In Berlin, one can visit several meetups every week with this topic. We all heard or used one by now and probably some of you already have build one. Lots of our clients at Disruptive Elements also are looking into them and want to create one in order to make communication with user and customers easier.
Using AI technology to improve a 1-to-1 conversation with your customer base can be incredibly uplifting for the brand awareness and brand experience. While this is, without a doubt a great idea, I think it is necessary to also understand why it might not. Hence, I thought I briefly explain why chatbots are still failing and then give my favorite chatbot fails of all time.
Why Chatbots are still failing!
The main problem is that Chatbots are marketed as intelligent using Artificial Intelligence Techniques. While this is true, that doesn’t make them smart. For the most commercial versions of chatbots the underlying logic is a so-called decision tree. This tree works with an if-this-then-that logic.
Ravi Gupta Decision Tree Example
This means the chatbot can only go down the branches of the tree. Hence, it can only transform possible attributes as inputs into reasonable outputs a long one branch. Both of them need to be defined and thought of by developers. As soon as customer enters something outside the lines of this tree. The chatbot doesn’t know what to do and fails.
Bots that use natural language processing (NLP) or complexes neural networks and hence can “understand" human language and learn new answers are more complex to create and are still not perfect either.
This above explained reason is mainly from a technical perspective. There are many other areas like lack of context understanding, lack of transparency, and others that are limiting the “intelligence” of chatbots. However, this would be too much for this article. The future will also bring new variations that probably are way more precise.
All time favorite Bot Fails.
#3: 4 pounds of cookies and a dollhouse please
Amazon’s Alexa is a voice-based chatbot if you will. Even though, the differentiation needs to be more certain, the underlying mechanism is the same. Specific action words trigger an action to be executed.
Earlier this year, a 6-year old girl ordered a 170 dollar dollhouse and 4 pounds of cookies simply by asking Alexa for the products. The mother realized this quite quickly but couldn’t cancel the products and donated the dollhouse.
This shows how easy chatbots can be tricked and that they are very simple to use. It shows also that they are not really understanding context and can’t recognize different people yet.
#2: Tay, Microsoft’s racist bot
Not really a chatbot, but still a good example for learning bots. The AI-powered bot was users on GroupMe and KiK as well as twitter sources and became a cursing, racist chat bot with in a wink. After 24 hours Microsoft was forced to turn it off. You can read it more about Tay here.
It is an interesting social experiment in my opinion and shows how the general tone of our social streams is.
#1: Facebooks chatbot develops it own language.
As a very recent example which I also mentioned in my last post, the Facebooks bots started to develop a their own language. While it doesn’t seem to be much of a deal because they are just a broken English. At one point, the developers at FB couldn’t understand anymore what the bots were talking about and more importantly, what functions they were executing.
Cool example of the problem by Business-Punk.com
Ultimately, the engineers had to reset the bots to English. You can read more about it here.
I hope you liked this article. I would love to hear your opinion about this topic. Hence,...
Question: Did you ever experience a funny chatbot fail that you can share with us?
My series of posts is about making you think a little deeper about every day concepts. I look forward to having you follow along and reading what you throw at me.
Peace!
Twitter: @tkronsbein
Instagram: @tizian_kronsbein
Website: www.tiziankronsbein.com