What is Artificial Intelligence?
Artificial Intelligence (AI) is an ambiguous and opaque word; it creates hopes and fears. Here we want to explain a little what does it mean at Spectrm and how we use it. 🔦
Keep it simple: Artificial Intelligence is a new way of programming computers. Instead of putting in the code with a lot of /if this then that/ you use techniques to let the machine learn. That’s why it’s called „machine learning“. Machine learning is the technique that is able to create “artificial intelligence”.
Our parents probably never told us that snow is cold or the taste of chocolate. We learned it and they let us learn it. The same principle applies to machine learning.
Be smart: The brain is unparalleled. Applying the same principles means not that you get somewhere close to what our brains are doing every day. Still, that new approach produces astonishing results and solves unsolved problems like self-driving cars and many more.
One of this unsolved problems is to understand what a user wants if he is just sending a text message.
Understanding the user
What Spectrm does is trying to understand what the user just wrote you have to make sense of his words and sentences. The ability of a computer to do that is called Natural Language Processing.
Be smart: Computers are able to „read“ numbers and words for decades, but not natural languages like English or German. Using just a predefined keyword like a command is something that chatbots have been already used in the 1980s.
Natural language processing tries to understand language in their full complexity. There are sometimes thousands way of saying the same. Natural language processing wants to understand all of them.
Here are some techniques how this works:
From grammar lessons, we know that a sentence has different parts like a subject, object or a verb. Entity recognition works similarly.
The computer tries to break the sentence into different parts to abstract it. The parts here are called entities. Entities can be locations, persons or organizations.
Having entities abstracted is a good start, but without the connections between them, we don’t really understand what the user what. A semantic parser is there to make that connection by analyzing the semantic structure of a sentence.
80% of communication in a conversation is non-verbal. The feelings of the other person are important to understand what he wants to say.
For chatbots that is similar, but written language is of course much more difficult. Sentiment analysis solves that problem. It tries to abstract the usage of certain words and then calculate a score to decide whether it’s positive or negative. Instead of words, you can also analyze emojis. 🙌
How we use it – Our principles
We built our own
Companies like Google, Microsoft or IBM and many more are providing solutions for natural language processing. Each built great technology, still we believe there is an incredible advantage to not just plugin there.
- First, it allows us to create something unique for every customer. Our own algorithm is designed to take not only domain but also company specifics easily into account. Monster is for Redbull something else then for Unity3D.
- Second, it is just a question of downtime and owning data. Outsourcing the brain of a bot is a risk you don’t want to take. We neither.
- It allows us to enable the magic for you. You are in charge. You can use our own fuzzy matching algorithm or play it safe with a rule-based approach. You can supervise the algorithm without being an AI scientist to create something your company has never seen before.
So exciting …
With the help and financial support of the Google DNI we started on our mission to enable our customers to speak to every user individually. A meaningful conversation that helps both to get the most out of their connection. We are thrilled to be on this mission.
If you have questions, drop us a line.