Adobe Stock

It has been a little over six months since generative artificial intelligence burst into the mainstream consciousness with the public release of ChatGPT. The mass extinction of humanity by AI hasn’t yet occurred, and the job market is as tight as ever. Was all the hype completely overblown?

Having AI write a three-act play starring characters from your company was a great party trick, but it hasn’t tossed the whole world upside down. Yet, there is a lot more clarity about the immediate path ahead, and how you and your company can take advantage of its benefits.

Generative AI Is Predicting, Not Thinking

When you first experience the output from these pieces of software, it feels like magic. The wonder, surprise, and a hint of fear all mix together to create an intoxicating emotional cocktail that compels us to assume there is real thought occurring behind the black mirror. There isn’t. All it takes to deceive us that software is conscious, it turns out, is being really good at prediction.

Think of generative AI as a trained chef making soup. The input prompting what you give AI is similar to choosing the base ingredients for a soup—water, vegetables, and chicken. The chef then thinks with an instinct formed by experience, “Garlic and onions go well with chicken, so in they go!” The AI uses its training to predict what words or phrases logically follow the prompt, similar to a chef adding ingredients based on their cooking experience.

Just like a chef tastes and adjusts the soup’s flavor, the AI constantly refines its output. Both the chef and the AI finish their processes when they predict any more additions won’t improve the end result. Both can make mistakes, but with more experience or training, the predictions improve. As with making a pot of soup, the goal of generative AI is to create a satisfying and coherent output.

Some critical human ingredients are missing from AI output for now: values, empathy, judgment, and wisdom. Those who attempt to leverage AI as a replacement for critical thinking rather than an augmentation will be easily outmaneuvered by the competition.

English Is Now The Most Powerful Programming Language

Javascript and C++ were programming languages taught to me when I was in college. Until very recently, it was necessary to learn entirely new ways of speaking and writing in order to interact with bits and bytes of a computer. English is now likely to be the only programming language my four children will ever need to learn. When you input a prompt for an AI system, you are programming it with each character that you type.

This isn’t a new concept, but it does have infinitely more significance than before. Every family has someone who is the go-to Googler. “Kevin, can you please help me find more information on the best kind of shin guards for an 8-year-old soccer player?” my wife will ask in a frustrated tone after she’s spent 10 minutes looking for the answer and is hitting dead ends or poorly filtered results. Unconsciously, I select more relevant and curated search terms into Google and uncover the answer in 20 seconds. In order to get the best and desired output, you need to think about the words you enter as exactly what they are—programming code.

For example, if I put in the prompt of “Sally takes the cake,” there is potential confusion on if I am inferring that Sally literally is taking a cake or a common cultural idiom. If I add a single word to my line of “code” and instead say “Sally really takes the cake,” the intent becomes unmistakable to the AI.

Whomever in your organization is the best “programmer” for AI can easily compile and share their most useful prompts and give others a recipe to follow as the base for a delicious bowl of soup. In some cases, particular sets of prompts—just like valuable software—will become a highly guarded asset by companies.

Five Current Areas Of Excellence

For many, it will be helpful to consider the five main areas where AI excels in today’s world in order to extract the most usefulness: generation, extraction, summarization, rewriting, classification, and question/answer.

  1. Generation: The most well-understood use today is generating words and images based on prompting. Remember that more thoughtful prompts with multiple examples will create higher quality output.
  2. Extraction: Pulling out data from one context or form for use in another (i.e., creating text from audio, text from images, or useful raw data).
  3. Summarization: Simplifying large piles of information into succinct thoughts (i.e., reading every customer comment or review of your company and distilling them into statements about customer sentiment that your executive team will actually read).
  4. Rewriting: This isn’t another word for plagiarism. Instead, it refers to changing from one version to another—often for tone, feeling, or a different medium.
  5. Sorting: Labeling or grouping large amounts of data into subsets (i.e., labeling the content of every photo in your builder’s image library, or identifying options from your blueprints).

These five areas are tasks that AI does really well today—especially when trained on the right kind of data. As I consider the list, it feels like a lot of the same tasks we might assign an intern or less experienced team member, which really drives home that AI is likely first to substitute for tasks and not experienced employees. However, experienced team members who integrate AI for these five tasks can multiply their impact and value.

Two Big Remaining Issues

There are two large, complex issues that six months later are not much closer to being remedied: copyright and bias.

In 2018, the case of Naruto v. Slater set the precedent that non-humans could not hold a copyright. It’s quite an interesting case involving a monkey that took a selfie, but we don’t have the time to go into full detail here. The important takeaway is that content generated by an AI system is currently not protected by copyright. As expected, there are a lot of gray areas, and those areas will invite lawsuits and disputes over ownership. Prompters beware.

The second issue of bias is sadly not unknown to the real estate industry, and it is important that as we implement AI into our workflows that we are aware that it still exists. It isn’t so much that it exists in the code, but in the data on which the code is trained. Whether intentional or not, making sure data bias does not impact the output of an AI system will be critical to engineering a better built world. The success of our continued journey into the future hinges on balancing innovation with caution and empathy with excitement.