Generative AI such as ChatGPT: a blessing or a curse in customer service
«Prompting is the new business: from machine learning to machine teaching» Prof. Dr. Peter Gentsch
Everyone is talking about generative AI and conversational marketing. Of course, these technologies could become a game changer. They can be used to rethink customer service. And the new generation of artificial intelligence could potentially even become a production and innovation factor. There is also a fascinating aspect: results appear on screens as if by magic. It is easy to forget that AI simply processes large amounts of data in order to put probability calculations on a broader basis.
Gloating over mistakes
The hot topic of the moment has of course also reached the shiftfluencers, who turned their backs on their established topics and now call themselves specialists in artificial intelligence simply because they have played around with it a bit. They are hardly helping to answer the unresolved questions that are keeping many professional users waiting. The mistakes are just too spectacular: a photo shoot of footballer Erling Haaland turned AI into a death report. In the USA, ChatGPT invented plausible precedents for a court case. When this got published, it was extremely embarrassing for the lawyer. New possibilities also mean new risks on the way from the lab to the factory. That’s why the changes still need some time before they become part of our economic life.
Fear makes success impossible
When I did a self-test with ChatGPT on a self-portrait of myself, the result looked very good and I felt flattered. But the right things were mixed with completely wrong things. Every chatbot is only as good as the data it uses. In order to increase the quality of the results, more knowledge can be injected into the language models that reflects the respective domain and corporate world.
Source of data is key
While generating images, you realise where the data that the AI was trained with comes from: They often show picture content that primarily comes from the USA. Quality assurance is therefore essential, and incorrect answers are a problem that can currently only be solved with human help. Clickworkers made the difference in the development of ChatGPT 3.5 to 4.0. They evaluated content generated by GPT, flagged the harmful content for the learning algorithm to ignore and ensured a leap in quality. The ‘human in the loop’ as a data labeller in the sense of reinforcement learning is still indispensable.
It should also be kept in mind that the copyright discussion has not yet been fully resolved.
Misuse possible
ChatGPT was made possible by large language models, NLP and deep learning. These technologies have been in existence for quite a long time, but it is surprising how quickly progress has been made recently through the intelligent combination of these technologies. The basis for this is the huge database. Multi-modal models now even allow realistic conversations with virtual people, such as Albert Einstein or Gutenberg. As impressive as this is, this technology also provides an opening for fraud and crime. For example, the grandchild scam could take on a whole new dimension, but I am still sceptical about legal regulations. Innovation cannot be banned.
It’s worth it despite everything
Despite all the concerns, artificial intelligence can bring benefits to companies. Every industry and every functional area is likely to benefit from the support of AI in the future, across the entire value chain. In addition to productivity and efficiency benefits, ideation and innovation can also be realised. Cola Cola, for example, has used AI to develop a new recipe called Y3000. AI provided support here as a co-creator. Marketing can personalise campaigns. In the field of education, it would provide personalised content – students would never again be bored or overwhelmed. AI helps with knowledge management, writes protocols, manages appointments, carries out research, makes market and customer research dialogue-capable and automatically writes product descriptions for online retailers – even SEO-optimised. When rephrasing, it can change existing texts depending on the channel or show different images depending on the customer. All of this works well once it has been trained. The financial contribution to OpenAI to use the tool quickly pays for itself.
The machine as co-pilot
Of all the conceivable possibilities, I believe that augmented intelligence is most likely to be used in service. Like a co-pilot, it is helpful in combination with a human. In a test, knowledge workers who used AI completed 12.2 % more tasks on average, were 25.1 % faster and achieved 40 % better results. The customer journey will soon be supported by AI, from customer acquisition to pre-sales and after-sales.
There are still some things that humans can do better than computer programmes, even if they are becoming less common. In tests with customers, AI performed better than humans in 98% of cases. It even simulates empathy better than humans.
A practical approach
If you want to start using AI, you can have artificial intelligence generate use cases. This gives you an initial brainstorming session. You can then identify the pain points you want to address and let the AI suggest solutions. This provides you with points of reference that are worth drilling into and developing further. The key to language models such as ChatGPT or your own models are the prompts, the task descriptions. (Best-practice prompts can be found here https://foundation-group.ai/) They must be well designed in order to achieve good results. This is why the job of prompt engineer was created. ‘Prompting is the new business’! From machine learning to machine teaching: employees need to be trained to ask the machine the right questions.
