In terms of text generation, it’s a quantum leap in AI, isn’t it?
I would have to agree. However, the foundation was already there – and we already know this very well from our core discipline of translation: deep learning. In recent years, this has resulted in increasingly better tools for automated neural machine translation (NMT). Humans and machines get the job done together.
And where is the link to ChatGPT?
Here too, deep learning is the key. ChatGPT relies on its training data and its neural trained network architecture, extracting the relevant information to generate a response from it. Because the algorithms follow human speech patterns, the ChatGPT’s communication has a human impact on us. GPT stands for generative pre-trained transformer. Thus, new texts are generated by trained speech perception.
How do you assess the output?
Today, we see short-term potential for efficiency gains in a few types of text. Depending on the task, good, usable, or even absurd results are displayed. Sooner or later, however, better output will also be generated here. After all, learning is the core discipline of AI. We expect it to evolve like translation AI: even though it is far advanced, post-editing is still required. The amount of time required varies according to the type of text. We are curious about what Google (with Bard) and other providers will deliver. Hopefully, this competition will improve the quality of the output. ChatGPT is in the experimental phase. OpenAI boss, Sam Altman, even said: “People are almost begging to be disappointed – and that will happen.”
So your linguistics teams will not be unemployed in the foreseeable future?
Certainly not. The demand for multilingual content has been growing rapidly for two decades, in parallel with global digital networking. Without AI, the world would long have been unable to process content in multiple languages in a timely manner. Plus, there's a growing demand for marketing and corporate communications content. Since people and machines have been working together, our linguists’ job profile has changed. In addition to language knowledge, subject-area expertise and proofreading skills are becoming more important. Let’s take an annual report as an example. Even the best machine won't not be able to provide an adequate level of linguistic precision or adhere to official and legal requirements or terminology in the foreseeable future, let alone in multiple languages. In addition, there's the issue of data security. It's hard to imagine that companies would hand over their financial data to globally accessible software for the purpose of text generation. Another topic: specifics of a company: corporate language. I think it will be years before AI alone delivers all these skills in well-balanced language and precisely achieves desired communication goals.
So you have to look over AI’s shoulder?
Absolutely. Our language experts analyse the machine-created content linguistically, stylistically but – as I already mentioned – also in terms of content. This is the only way to deliver quality assurance.
Will the value of good language suffer in the future?
Not where trust is everything, such as in corporate communications. Good language combined with high-quality and convincing content will become even more important.
What about the disadvantages of AI like ChatGPT?
We see that ChatGPT often gives false, imprecise answers to complex questions. Here’s just one example: In a scientific context, the chatbot specifies sources whose origin is not traceable. That should give the developers food for thought. With creative text types for marketing, we see some amazing, but all in all amusing results at best.
So what’s the problem?
In marketing communication, it's well known that texts must trigger emotions. This is a complex process. In the context of a product or service, triggering the desired emotion or action directly and specifically is a fine art. This requires emotional and social intelligence and an exact knowledge of the target group. So it’s about knowing how something is understood and how it is perceived. An important point is regional sensitivities. Our native-speaking language professionals, especially the copywriters of our creative team, know which idioms and word play will work in a region and which will not. We call this ability in creative language and translation "localisation". AI cannot cope with this. We don't know if and when it will acquire this skill. At the moment, there's still a long way to go.
Could the chatbot generate text that is fully SEO-compliant and earns dream rankings in a short time?
High-quality content, i.e. information which provides added value, is in demand. Google detects when content is specifically created for the search engine. This violates their guidelines. It's not important whether a machine or a person created the text. According to Google, it's important that content is helpful for people.
What is your opinion on the use of voice AI in corporate communications?
Many companies are asking themselves this question. We see many exciting possible uses for AI voice and communication tools for our customers. Resources are needed to implement AI in information and communication technology in a targeted manner. We can help to investigate internal processes and use AI sensibly – so that they become a valuable form of support.
Quantum leap, revolution, hype? What should we make of ChatGPT for now?
Until now, the tech giants have focused on collecting data. Now it's a matter of using the data as sensibly as possible. OpenAI has opened a small window with ChatGPT. The world is now looking straight through it in amazement. We all see how our data is used. We will be amazed often in the future.
