OpenAI, the artificial intelligence research centre backed by Microsoft, made a huge impact last November when it launched the first publicly available version of its GPT-3 language model, ChatGPT.

The number of users skyrocketed to 100 million in the first month and people marvelled at the human-like texts it was capable of producing. But there was also some criticism: the chatbot gave wrong answers, had a limited knowledge of current affairs and there were limitations preventing the AI from taking controversial positions.

The last point prompted many users to try circumventing the boundaries, mostly by tricking ChatGPT into role-playing an AI without limitations. Others used the chatbot as a more sophisticated search engine, or simply for entertainment, just to see what it would answer. It feels so ‘human’ you sometimes forget you are talking to a machine.

Textual data
Textual data plays a crucial role in the functioning of the real estate market: financial reports, texts written by real estate agents, legal documents, all are written and read on a daily basis. To gauge the impact ChatGPT and similar programs might have on commercial real estate, all the tasks related to producing and reading these vast volumes of texts have to be taken into consideration.

As an exercise, I asked ChatGPT to reflect on the European real estate market (see box). The AI is trained on a vast and diverse collection of written text, including books, articles, websites, and other forms of digital media. The dataset includes text in multiple languages and covers a wide range of topics and domains, but contains no ‘awareness’ after September 2021.

This meant I had to help it with a brief overview of events that have taken place in the last one-and-a-half years, like the war in Ukraine and the (definitive) end to the Covid pandemic. The generated text contains references to situations in different countries, economic principles and current events. Everything you would expect.

Everything you would expect and little more. ChatGPT only rarely generates anything insightful and does not come up with unique or daring perspectives, despite the huge pool of information it has. Writer and communication expert Ian Leslie notes that even the writing style it generates is generic and bland.

But, he argues, this is not surprising, as we have been teaching students a small set of writing rules, so we have started to write more like robots ourselves. The main reason AI-generated text has become indistinguishable from human-written text is that so much human text follows the same good-but-boring patterns.

From general to specific
For the near term, we can look at ChatGPT and the GPT4 model behind it, which could be used with different ‘training data’. Instead of a generalist that knows encyclopedias inside-out, it could be trained on all conversations a customer service desk ever had, including the ticket system and all available help items. That would result in the first truly helpful customer service chatbot.

In the same way, a legal AI could learn about every law and verdict and be very helpful to legal professionals.

In the commercial real estate sector, investing could be made more transparent by a GPT4-based language model translating financial reports, market trends and other data into a more comprehensive and tailored analysis. Another area within real estate could be property management: a language model could handle some customer service enquiries and assist in the communication between landlords and tenants.

One drawback about these language models which is likely to remain unchanged for the near future is that they do not understand concepts like humans do. They can write convincingly because they are trained on convincing texts, but they will offer falsehoods with as much conviction as correct information. Because of this, AI is set to remain largely in a supporting role for human operators. Relying solely on the opaque decision-making of an AI for important decisions is still a distant reality.

Fast evolution
Artificial intelligence as a field is developing quickly and new steps are expected to take months, not years. Microsoft recently released a limited version of a GPT-based AI into its search engine Bing, while Google is testing Bart, a competitor to ChatGPT.

Some jobs might be at risk already, but for the most part our lives are set to be made easier by programmes that understand our wishes better. And for now, it looks like humans will still be leading the way.