Generative AI is taking the world by storm and already the applications are numerous and popping up across industries. But what do we, the data scientists at Prophia, mean when we talk about generative AI? Read on to find out.
It feels like the discussion about AI has filtered into every corner of the internet, turning us all into AI enthusiasts. In fact, there hasn’t been quite a shift in the culture around a tech solution since Uber made it acceptable to climb into a stranger’s vehicle. The truth is, we are not all AI experts but we all have a sense that AI is here to stay and change our lives.
One form of AI in particular, generative AI, has rapidly expanded its influence across various industries, even leading to a strike in the entertainment sector, where writers demanded protection from its disruptive effects. At the same time, the Future of Life policy workers initiated a letter with a six-month moratorium on AI advancement, gaining support from industry leaders with almost 30,000 signatures.
Despite the concerns, AI also brings significant benefits, such as advanced machine learning techniques in cybersecurity to detect fraud and the potential to revolutionize healthcare by providing personalized care to patients. However, the commercial real estate industry, which heavily relies on human interaction and physical assets, has historically struggled with technology adoption.
As we all become more used to interacting with generative AI, even industries that have held off on its adoption will eventually see the light in the competitive advantage it will offer early adopters. But what exactly do we mean when we talk about generative AI in CRE? Read on to find out.
Jump To A Section in This Article
What Is Generative AI?
Before Generative AI, There Was NLP
The First Generative AI Models
A Peek Behind the Computational Curtain
Generative AI in the Era of Prophia
Of course, first things first, since we are not all AI experts, what exactly is generative AI? In the world of artificial intelligence, a generative AI model is one that can produce original content. The term “AI” on the other hand, simply refers to the broader field of teaching machines to mimic human capabilities like sound, sight, etc.
What’s more, generative AI itself comes in different forms. When an automated marketing campaign sends a customer a coupon code offering them an option to opt out of future texts with “yes” or “no”, this is an example of discriminative AI. When a user asks an algorithm to produce an output based on a prompt, that AI is generative.
Whether we’re asking AI a yes or no question or we’re using it to produce a new response, the history of generative AI dates back far prior to the 21st Century.
The generative AI models we know today are based on decades of breakthroughs in artificial intelligence. One such subset of the field, and an important component of Prophia’s generative capabilities, is Natural Language Processing, or NLP.
NLP is a field of artificial intelligence and computational linguistics that focuses on the interaction between computers and human language. It aims to enable computers to understand, interpret, and generate human language in a way that is both meaningful and useful—I’m sure you can already see the connection between an algorithm’s ability to create original content and generative AI.
NLP techniques often involve a combination of statistical models, machine learning algorithms, and linguistic rules to process and analyze text data and Prophia’s development by the first data scientists on our team was no exception. Many of Prophia’s generative models today stem from NLP and the ability to recognize and process natural language. And, perhaps, the foundation reveals why we are so willing to interact with a machine.
Artificial intelligence feels like a uniquely 2023 conversation but its origins really date back to the 1960s when Joseph Weizenbaum created the ELIZA chatbot. Weizenbaum modeled this early chatbot’s behavior after the Rogerian style of psychotherapy in which responses mirror the recipient’s language in the form of a question. For example if you were to query ELIZA about a frustrating conversation you just had with a family member, the technology’s response might be, “What about that frustrated you?”
Initially, Weizenbaum’s experiment was meant to demonstrate that human-computer interaction could not be enriching for a human recipient. The result, however, entirely contradicted this thesis. Those who interacted with ELIZA were enthralled by its responses and quickly fell into deep, meaningful conversations with the AI.
Today, generative AI has come a long way since Weizenbaum’s ELIZA, but the public still seems to have the most profound reaction to AI when the machine learning functionality resembles that of a conversation, i.e. we ask it a question and get a response, like speaking to a person. This seemingly familiar and human aspect of generative AI is actually the result of highly complex computational mechanisms at work.
When ELIZA would respond, “Why did that frustrate you,” to a recipient in the 1960s, or Siri reminds us of an upcoming family member’s birthday, those manufactured responses are the result of machine learning and AI. The two are slightly different but key to understanding what we mean when we refer to the “generative AI” at the heart of Prophia’s proprietary machine learning.
As we mentioned, artificial intelligence refers to the broader field of teaching machines to mimic human intelligence. Machine learning, on the other hand, teaches machines how to perform a certain task or generate a response by identifying patterns in a dataset. Our own CTO, Chris Hsu, provides this example to remember the different yet symbiotic relationship between AI and machine learning:
“Imagine a race car driver as an AI system, and the car as the machine learning model. The race car driver's goal is to navigate the race track as efficiently and quickly as possible, just like how an AI system aims to solve a specific task or problem. In order to drive the car effectively, the race car driver needs to learn and adapt to different situations on the track. Similarly, a machine learning model needs to learn from data in order to make accurate predictions or decisions.”
Chris Hsu, Prophia CTO and Co-Founder
Where artificial intelligence is able to mimic some of the same capabilities as a human (sight, speech, etc.) and machine learning makes predictions based on data synthesis, generative AI is a particularly fascinating breakthrough because it is an AI model that is able to create based on its perceptions; build something entirely new. For instance, if you prompt generative AI to create an image of a black dog, it will compile thousands of images of black dogs to generate and create that image. Hence the “G” in “ChatGPT”, Generative Pre-trained Transformer.
While Prophia is just one iteration of generative AI, the output may or may not resemble that of ChatGPT or ELIZA and it can’t help users work through a difficult conversation or create an original image of a black dog. It can, however, help answer difficult questions about a building and its leasing, getting information to users quickly and make recommendations based on underlying data.
This is an example of generative AI that provides an output in the form of data. The AI works from common, yet proprietary, real estate building documents, such as lease agreements, to create annotations with data. That data is then synthesized by our platform and generated into a powerful lease summary with hyperlinks connecting our AI-generated lease abstract with the original source document.
As a result, this connection with the original lease is almost like a conversation, allowing individuals like asset managers, property managers, brokers, and financiers to interpret and parse data with proper context. With Prophia, users can easily search through thousands of AI-layered contracts and concepts to find the insight that is key to decision-making.
This type of AI-generated response is both accurate and immediate (like ChatGPT) and illustrates the strides AI is making since the first models were trained on human behavior in the 1960s. What will the next iteration of a generative AI application for CRE look like? Prophia data scientists have already begun investigating the platform’s ability to generate conversational outputs and support ChatGPT-style inquiries from users.
But NLP takes time and copious amounts of training data. Luckily, Prophia continues to make strides everyday in the development of its AI and NLP, as a first-ever CRE solution capable of supporting conversational inquiries and generating an accurate response. If you would like to be included in the generative AI conversation in CRE, please reach out to a member of the team to learn more about the ways in which Prophia can transform your team’s portfolio data.