Does an AI really speak or is it just pretending?

Words are the most powerful weapon. We don't want to drift into bellicose vocabulary, but this phrase powerfully sums up the importance of language. One wrong word can trigger a catastrophe, the right “yes” can change an entire life. To leave the philosophical level again: In customer communication, the written and spoken word are decisive for a good customer relationship and thus also indirectly for business success. But if human communication already leaves a lot of room for misunderstandings, how can an artificial intelligence communicate without making mistakes?

Does an AI really speak or is it just pretending?

This depends largely on two factors: first, on the requirements of the queries addressed to the AI. And on the other hand, how well the engine of “One Conversation” is trained. Or to put it simply: How well it knows your language.

Technically, such language learning nowadays usually takes place with so-called neural networks. Similar to the brain, the artificial neurons exchange signals with each other via synapses. During learning, the individual synapses adjust how strongly they transmit the signal of one neuron to another. The more artificial synapses a language model has and the more text it has processed during training, the better it can, for example, guess how to complete a sentence fragment.

In the case of “Clouds are gathering, it will soon ...”, “rain” would be a good choice. In a winter context, on the other hand, “snow” would be more appropriate. In this way, AI lines up word after word, resulting in longer texts. Exactly this type of AI has seen tremendous improvements recently with so-called “foundation models”, which can contain hundreds of billions of parameters, and counting.

Does that qualify as “really speaking”? Ultimately, such an AI merely reproduces and combines what it has learnt – words and sentences that had been built by humans. The key is: The AI can also learn from its training to reproduce and combine smartly – in a way that makes sense for humans. For the One Conversation programme, this means that the conversational engine can be trained on what we want to achieve: Consistent and courteous communication that focuses on the needs of the customer.

Is there room for emotion in this? The answer to that is a very clear “it depends”. So if we know how artificial intelligence works, it quickly becomes clear that emotions don’t natively exist there. But you can certainly bring politeness and friendliness to the engine, in other words, manners in the broadest sense. Ultimately, this is no different than what education does – you teach the next generation how to deal sensibly with their fellow human beings. Call it decent behaviour – and that can create positive emotions.

If our conversational engine successfully addresses our customers’ needs that way, we are on the right path.

Stay tuned.

Yours, Jan