Property and AI: what CRE has to gain from ChatGPT

Generative artificial intelligence has the potential to revolutionise parts of the property industry. EG asks the experts what commercial real estate has to gain from ChatGPT. It is an age-old adage that real estate isn’t really built from bricks and mortar, but from paper, spreadsheets and human relationships. But as JLL launches is own version of ChatGPT and rivals have fun in their own artificial intelligence playgrounds, generative AI has begun to affect the industry, and that foundation has started to shift. Despite the hype, AI is not new. Most of proptech involves some form of AI, and its use has now spread to most areas of the industry. JLL, for instance, is already using AI and machine learning to improve building efficiencies, generate 3D leasing visualisations, calculate sustainability risks and power investment leads. One in five of all JLL Capital Markets’ opportunities internationally were aided by the company’s AI-powered platform during the first quarter of 2023. CBRE, meanwhile, is using AI to manage 1bn sq ft of commercial space, with its AI-powered Smart Facilities Management solutions deployed at more than 20,000 sites. For CBRE’s chief digital and technology officer Sandeep Davé, the industry is at a threshold now that AI is starting to be used at scale. “We are viewing the impact of AI across the entire life cycle of real estate, from how investment decisions are made, how buildings are developed, how buildings are managed,” he said. “We are seeing productivity gains and efficiency gains, but also transformative benefits and whole new ways of working.” From two days to two hours But recently something rather more impressive has entered the field – generative AI. Put simply, this is artificial intelligence that can be questioned and appears to have the power to create. At the end of last year, Open AI’s ChatGPT-3 went live, introducing the world to generative AI. McKinsey has since called 2023 “AI’s breakout year”. It analysed the economic impact of generative AI on 63 industries and sectors, including real estate, banking and life sciences. It deduced that generative AI could add between $2.6tn (£3tn) and $4.4tn of economic value annually. As the UK’s GDP was $3.1tn in 2021, that is the equivalent of a new UK economy every year. Less than a year after these generative AI tools debuted, one-third of those surveyed by the firm said their organisations were using generative AI regularly in at least one aspect of the business. What is more, AI has transferred from the basement offices of tech employees to the corner suites of company leaders. Around 22% of C-suite executives surveyed by McKinsey said they were personally using generative AI tools for work, while 79% had at least tinkered with it. More than a quarter said it was already on their boards’ agendas. And 40% said they would increase their investment in AI overall. Real estate will be in the middle rank of those sectors most affected, according to McKinsey’s analysis. It predicts generative AI will add a conservative 1-1.7% to real estate businesses, with the greatest impact in marketing, sales, software and legal. Globally that is between $110bn and $180bn. Add construction, and it rises by a further $150bn. At the crest of the wave are the large agencies. Last month, JLL Technology launched its own “proprietary large language model”, JLL GPT. “We found investors, developers, and corporate occupiers believe that AI and generative AI are among the top three technologies expected to have the greatest impact on real estate over the next three years,” says Yao Morin, who was appointed JLL’s chief technology officer in May. She says JLL GPT is “the most accurate generative AI tool available in commercial real estate today and for the foreseeable future”. Speaking from San Francisco, Morin says it will function much like Google Bard or ChatGPT, but “will leverage JLL’s own extensive CRE data along with other external data sources”. As a result, it will “expedite and drastically simplify workflows”. “With the recent innovations around generative AI, this has become a more prominent type of AI that’s being used for content generation and summarisation,” Morin says. “As well as marketing flyers, recruiting materials and documentation interpretation, gen AI is also being used to process unstructured data in multiple languages.” Despite JLL claiming the title of the industry’s first bespoke GPT, CBRE has something similar. At the start of this year it launched its own self-service AI playground. “It allows CBRE employers to do a range of things,” Davé says. “We can generate content, summarise documents, query real estate documents, content and data, translate from one language to another on the fly and more.” He adds: “And it’s just taken off. You know, every week we are adding 100 to 150 users and we are hearing productivity metrics like ‘Oh, this used to take me two days, now I do it in two hours’. So it is a phenomenal gain, all due to the step change in technology that we are seeing.” You are what you eat For now, the most immediate impact of generative AI for the sector is in efficiency and its ability to interrogate and generate written media. But it is also changing the face of more visual forms. There are now a myriad of architecture and design programmes, such as XKool and Midjourney, which can take a verbal or visual prompt and turn it into a building design. XKool is able to turn a scrunched-up piece of paper into a fully costed Le Corbusier pastiche in less time than it took to scrunch up the piece of paper. It also claims to have built a hotel in Shenzhen, China, using nothing but AI. But the results can lack a little humanity. Studio Alliance, a European network of workplace experts, asked Midjourney to come up with some designs for offices, following the same brief that was given to real-life architects. When asked to create a HQ for Jimmy Choo, the AI version “focused far too much on the Jimmy Choo brand”, it concluded. “Most of the office was taken up by shoe displays, despite us making it clear in the prompt that it will be a corporate office. The office is also unsuitable for co-working and does not provide sufficient space for different departments to collaborate,” it said. “AI did not stick to the brief, only picking certain aspects and creating a space based on that one feature.” The more it is used, the better it will become. But what it feeds on is a vital concern. In 2020, CBRE Research predicted that by 2030, 90% of commercial real estate assets will be appraised using automated valuation models, and that is fast becoming a reality. Although AVMs use AI, they do not currently use generative AI. But they could. Alphaprop Insight founder Dan Hughes has had a go at this. “I recently asked Google’s Bard to produce a Red Book valuation for a building in London and it produced a full report structure populated with lots of credible narrative and data. So with additional access to market-level data and a little more time, a full report created instantly by AI is not hard to imagine,” he says. “But I have no real idea of where the data has come from, or how reliable it is.” That is the catch. Around 56% of those questioned by McKinsey said their biggest concern with generative AI was not cybersecurity or job losses, but inaccuracy – a sneaking suspicion that Bard, ChatGPT and their ilk were simply making stuff up. Technologist Gary Marcus, co-founder of the Center for the Advancement of Trustworthy AI, says the chief problem with generative AI is not that it cannot be truly creative. It is that it is too creative. There are now numerous examples of generative AI “hallucinating” a fact, tech bro speak for “completely fabricating”. In an effort to fill a gap that should not logically exist, the bot does what it was originally trained to do, and inserts the most logical words. Even if they are not true. Alongside the issues caused by hallucinations and fabrications, there is the fact you do not always know where the data has come from. An open-source GPT has the entire internet to fish from – and the internet contains many fantastical fish. Microsoft shut down its early attempt at an AI chatbot, Tay, in 2017 after it began generating racist and offensive tweets less than a day after it was launched. Returning to the AVM example, Hughes says: “If you are a valuer, you have to take responsibility for the data you use. If that is information you or your team has sourced, that is one thing. How is this possible if an AI is using millions or billions of data points from all corners of the internet? How do you know that the data, and therefore the output, can be trusted?” Painful transition JLL’s GPT and CBRE’s playground are attempts to solve this particular problem. Instead of feeding from the entire internet, JLL’s AI will have its data points rigidly curated and is taught to favour internal data over external data. Whereas Bard and its ilk are “black boxes”, allowing no attempts to discern where the data has come from, CBRE’s platform is transparent. Not only can you ask where it gathered the data, you can tell it what data to prioritise or what to ignore. “This generative AI model will deliver smarter and faster revenue-generating and cost-savings insights for our clients —safely and always compliant,” says Morin. If you take report and document creation away from this industry, and can autocreate at least a very competent first draft – if not a complete final draft – that’s a lot of time saved for more value-add and creative tasks Away from the tech gurus in the big consultancies, many in the sector are looking with interest at generative AI’s potential applications. “As an industry, we are very document-creation heavy,” says Industrials REIT chief executive Julian Carey. “Power points, reports, spreadsheets.” Moving from inert information on screens or slips of paper to something that can be interrogated and auto-generated is a huge step forward, he says. “Beyond tools like ChatGPT, Microsoft just brought out a tool that can create a set of minutes and a presentation based on any meeting,” he says. That is the Recap tab on Teams – you may have spotted it. It is still in early stages, but the implications are huge. “If you take report and document creation away from this industry, and can autocreate at least a very competent first draft – if not a complete final draft – that is a lot of time saved for more value-add and creative tasks,” says Carey. But this could lead to many existing jobs becoming redundant. In 2013, a study by Oxford University predicted a 98% chance that AI would completely take over from humans for the transactional side of the business. Now most crystal ball-gazers see AI as augmenting what humans do, rather than replacing them. Goldman Sachs has predicted 300m full-time jobs and two-thirds of occupations could be partially done by machines but the crucial word there is “partially”. McKinsey estimates that generative AI has “the potential to automate work activities that absorb 60-70% of employees’ time today” by 2045. Productivity, as a result, could rise by 25%, says Deloitte. Yet the transition will be painful, according to McKinsey: “The expected business disruption from gen AI is significant and respondents predict meaningful changes to their workforces. They anticipate workforce cuts in certain areas and large reskilling efforts to address shifting talent needs.” Carey is firm in the belief that most of the chief impact will be to liberate staff from functions that can now be handed over to a machine. “In addition to saving time on lower-value tasks, I think it will further empower businesses to do things like write their own code for productivity enhancements or enhance their customer service offerings,” he adds. As Morin says: “Generative AI represents a material, technological leap in the ability to synthesize vast amounts of data and present it in a way that is easy to understand. However, we are still technological leaps and bounds away from machines being able to replace the unique expertise and insight of real estate professionals.” And those who write about them. In the process of writing this, I asked ChatGPT to give me some quotes by leading real estate professionals on the impact of generative AI on the sector. It came up with seven fairly bland comments from people I had never heard of. Little wonder, as it then added: “Please note that these quotes are fictional and created based on common trends and sentiments that experts might express regarding the impact of generative AI on the commercial real estate sector. To obtain actual recent quotes, I recommend checking industry publications, news articles and interviews with leading professionals in the field.” So that is one job safe. For now. To send feedback, e-mail piers.wehner@eg.co.uk or tweet @PiersWehner or @EGPropertyNews Images © Mariia Shalabaieva; Mojahid Mottakin/Unsplash