When highly-regarded Holocaust campaigner and educator Marina Smith died at the end of June, friends and family thought that they had heard from her for the last time. But what made her funeral particularly memorable was that she appeared able to answer questions at it, thanks to new technology developed by her son's tech firm Storyfile.

Video technology meant that pre-recorded video clips of Marina talking about her own life were selected by AI in response to questions from mourners. The technology did not create or put new words in her mouth; all responses were from genuine recorded footage and enabled her to have a virtual presence at the funeral.

So-called conversational videos may become more widely adopted in the near future, providing family history with a sense of interactive posterity.

For example, videos will allow children to interact with teenage versions of their parents, bridging generational gaps.

To produce the conversation, while the person is alive, he or she must make recordings answering numerous questions. Clearly, the more comprehensive the answers, and more varied the topics, the more in-depth this conversation would be, with relevant companies offering a range of guided subject matters.

The concept of having such posterity in an individual's life is nothing new, with personal diaries having offered glimpses into people's hopes, dreams, and fears for hundreds, if not thousands of years. More recently, around a decade ago, early You-Tubers created vlogs of themselves talking, describing their 15-year-old lives, and again as 25-year-olds, creating moving and retrospective time-travel monologues, where teenage insecurities are answered by mature adult versions of themselves.

Introducing AI to make this an interactive experience with the departed, however, is a new concept. AI selects appropriate clips to play in response to questions from remembrance videos, creating the illusion of speaking to the dead. Many celebrities have taken up the trend, with individuals such as Star Trek actor William Shatner appearing on Storyfile's website as a main case study.

Such technology is also being rapidly explored in the realm of sales and customer service, where automated responses can save human time and effort, to be spent elsewhere.

Using AI to create fully synthetic versions of people living or dead, however, is still a long way off. Humans are so unique and complex in their individual history and nuances that at this stage, even the appearance of a living person will still not emulate the real thing. Human interaction is a two-way street, and, eventually, the sheer unpredictability of genuine human curiosity and interaction will cause AI to be left with few options to respond.

At this stage, using people's genuine words seems to be within the ethical boundaries of today's society. However, utilizing AI to put words in people's mouths, for example, by rearranging words and phrases to form unique things said by a person, is dangerous territory, and not without its current technical limitations.

This was clearly seen by Meta, Facebook's rebranded company, when it released its own AI chat bot. Meta's BlenderBot 3 was criticized for making offensive remarks about the company's founder, Mark Zuckerburg. By drawing on phrases and opinions heard all over the internet, this form of deep learning and data collection led to unpredictably dark comments being made by the program. Offensive comments about Jewish people in relation to Zuckerburg and his wealth were spewed by BlenderBot, and criticisms of United States politics and individuals, such as Donald Trump, were also expressed.

The sheer unpredictability of AI dialogue may become difficult to regulate and use safely. For now, using AI to guide the words of the living, unadulterated, seems to provide the living with comfort and introspection, from those who cannot be with us anymore.

Barry He is a London-based columnist for China Daily