On November 30, 2022, the renowned generative model ChatGPT was released by Open AI. With its “chat” style graphical interface, one can ask something, (with the so-called “prompt”) and receive an answer in a few seconds.
Today, ChatGPT has expanded its services to include not only “text only” results but also images (via “DALL-E”) and videos (via “Sora”). But over time, competitors have come to the fore including Copilot (by Microsoft), Google Bard (now Google Gemini, by Google), Claude (by Anthropic), Firefly (by Adobe), not to mention all the other services (more or less on a paid-for basis) that can be found with a quick and simple search on the internet, some of which rely in turn on the above-mentioned services.
Thanks to their data analysis capabilities, these types of artificial intelligence systems are able to produce results in a few seconds or minutes with information and processes that would, depending on the case, require several minutes (or even hours) of analysis with human intelligence.
About a year and a half after the first release of ChatGPT to the general public, and a few weeks after the release of GPT-4, we can, indeed we must, ask how generative AI can be applied in the workplace. Listed below are some instances from the author’s direct experience, more or less.
“Textual” generative intelligence for creating an automated telephone switchboard
A company I came into contact with decided to make a telephone switchboard that, thanks to the synergies of a series of software, is able to ask callers who they want to talk to, check the availability of the desired person, and if they are not available, make an appointment or send a reminder via email.
It was possible to implement such a switchboard by integrating a number of technologies.
The first was speech recognition software to transform the caller’s speech into text. The text becomes the “prompt” with which callers “tell” the generative intelligence system what they want (e.g. “talk to Gino about the tax bill”) so that the AI system can process the text, understand if Gino is available (by looking at the appointments in his electronic calendar) and forward the call to the internal telephone number. In the event that Gino is not available, the AI system is able to find a free slot and propose an appointment, or get a message dictated to be sent to him via email.
To provide feedback to the caller, the AI system’s “output” text is in turn translated into audio with a speech synthesizer.
From this application of generative AI, one wonders if the figure of secretary/receptionist will eventually disappear.
It is difficult to discuss this in absolute terms. Undoubtedly such decisions will depend on the entrepreneur in question. The entrepreneur may decide to assign that worker additional tasks or roles that cannot be fulfilled by the artificial intelligence systems available (at least not now). Or the worker could be assigned the role of monitoring the system to ensure it functions correctly, thereby guaranteeing “human intervention”, as required by the GDPR, and “human oversight” as required by the EU AI Act.
Entrepreneurs who wish to automate “as much as possible” so as to reduce staff to a minimum should in fact bear in mind the business requirement of being able to (and having to) guarantee operational continuity. Operations are threatened at the precise moment in which there is a failure (for whatever reason) in the functioning of the IT infrastructures and related services that allow the “digital secretary” to perform well. And the penalty could be interruption of this part of the company’s activity.
However, it is equally plausible that not all switchboard operators or “customer care” workers can become “chatbot supervisors”. At least in some cases — depending on the degree to which AI systems can manage “first-level” assistance autonomously — it will be necessary for these workers to reinvent their technical and soft work skills.
Generative artificial intelligence for “impromptu” modification of emails or reports
From discussions with people involved in human resources and B2B marketing, it emerged that people are resorting to generative AI as they carry out their work.
It turns out that artificial intelligence can be used to “filter” the tone of direct communication with employees. The content of the text can be revised “in a flash” to make it more suitable and appropriate for the business context.
In the event that a report is needed urgently, with generative AI it is possible to produce a basic semi-finished product. Finalising that report takes less time than writing a report from scratch or on the basis of existing templates.
Such practices should make us all reflect on the importance of using these tools particularly during the phase of checking and revising the final document. It is important to remove from the prompt any parts that allow persons to be identified (starting with their names, surnames and the company they belong to). Because once we enter these inputs into the AI chat, we don’t know what happens to the data, especially if we rely on a free service that uses input data to improve the model.
Generative artificial intelligence to make complex topics more accessible
Today, social networks are one of the most important mass media for circulating information.
Whether the goal is dissemination or promotion, the possibility of incorporating multimedia content into posts makes it possible for those who publish to “hook” their audience with videos or images that help their audience assimilate the content. It is an useful way to draw attention to the “sea of feeds” and address complex themes.
The ability to generate automatic images and text can significantly speed up the creation of posts, newsletters and other communication campaign material. If created from scratch, it would take longer to produce similar results.
Having content that is more or less self-generated can be extremely useful, but even in these cases, it is always advisable to tread carefully and carry out a sort of peer review of what the generative AI system produced. The risk of causing harm to readers and organisations is just around the corner. Such dangers are exemplified by the Avianca case where the plaintiff’s lawyer cited judicial precedents that did not actually exist — they were the result of hallucinations by ChatGPT.
* The author is contributing in a personal capacity and not on behalf of the organisations he is involved with.
Image: Photo by Nathan Dumlao on Unsplash
