On the way to conversational super AI?

The classic, rule-based voice assistants on smart speakers and smartphones or chatbots on websites are not perfect. Sometimes they don't understand me and I have to try out different variations until I am understood. But why is it so difficult for a computer to process spoken language correctly? Nicolas Konnerth, Head of Voice at ERGO, and Sebastian Groth, Product Owner for Voice Assistants and Conversational Services at ERGO, reflected on this on LinkedIn.

Sebastian Groth, Product Owner Voice Assistants and Conversational Services

“I guess every bot developer secretly dreams of an artificial intelligence like J.A.R.V.I.S. or C3PO that can easily understand a human's spoken instructions and derive correct actions from them – without any misunderstandings”, writes Sebastian Groth in his new essay and wonders: “But why is it so difficult from a programmers' point of view to interpret the spoken word correctly?” In a stroll through the history of speech recognition, he looks back at the beginnings in the 1980s (“grammar as a key”), considers the 1990s (first machine learning approaches) and finally lands on current methods (Deep Neural Networks and Large Language Models). 

So how does the development of language models continue? Are we on the way to a super AI like J.A.R.V.I.S. or C3PO that understands our everything just as well as a human? If you want to read more, you will find Sebastian Groth's essay here on LinkedIn (in German only):

Nicolas Konnerth, Head of Conversational AI

A detailed look at what Large Language Models (LLMs) already exist, what they can do and what pitfalls they bring with them is available in the latest blog post by Nicolas Konnerth, Head of Voice at ERGO. LLMs use the experience of huge amounts of data to calculate what answer a user might expect to a question. The answer is therefore a statistical probability.

“Large Language Models such as GPT3, BERT or LaMDA always have a seemingly perfect answer ready and come so close to real intelligence that some even claim that there is a real sentience behind them,” writes Nicolas Konnerth. But the technology has its difficulties. Discover more and see some vivid examples in his blog post on LinkedIn (in German only):