Quiz: Did AI make this? Test your knowledge

Some work tasks pose more risks than others when it comes to generative AI. In the case of a real estate listing, in which a human inputs the details and uses the AI to create syntax, format and write the filler words, the risk is relatively low, Rahman said. That’s because the AI probably will rely on the data provided versus pulling from its repository of data, which could include misinformation and errors.

“You can rely on AI’s ability to crowdsource what works in formatting it and display,” Rahman said. “It’s going to be helpful in that area.”

Other areas in which the tech might succeed include creating verbiage for flash sales or social media posts, which might be quick, short, formulaic and easy for AI to replicate.

But humans may still want to do a quick read, even for the shortest of items, to make sure the AI didn’t misinterpret the prompt or get details wrong.

“A lot of the ways you write up a real estate listing does follow a specific format,” he said. “It’s more programmatic in nature.”

That said, humans may still be needed to ensure the listing is accurately represented and that areas a seller wants highlighted come through in the message.

While generative AI may be a great place for brainstorming ideas, prototyping or even gauging interest in a product, creative professionals may want to pause before manufacturing or selling a product developed by this technology. Two big pitfalls that could arise, Rahman said: using bad data or liability issues.

For example, if a company used social media posts to help it determine what’s hot right now, it could end up using data generated from bots versus real users and ultimately create something no one wants. And when it comes to liability, creators have to be careful they’re not creating a product that could be copyrighted or protected. So if a company wants AI to generate a funky set of generic bell bottoms, that could be okay. But if the company asks for a set of bell bottoms with an image of an avocado on them, it might be using an image, or parts of one, copyrighted by an artist.

“It does put the onus on the end user to do the homework,” Rahman said. “And if they don’t, it could have downstream consequences.”

The issue gets particularly tricky as generative AI tools don’t tell you what they used to generate the result they present. So a professional may not know whether what they’re creating could create liability issues.

Producing positive-sounding articles about specific brands or products is pretty easy with generative AI tools, Rahman said. But relying on the AI entirely could have costly consequences.

First, how well the AI does probably will depend on how much information the prompt includes. Some companies consider AI prompt-writing a job itself, given how complex it can be. But AI systems have the tendency to fill in blanks not provided in the prompt with information from their data sets. And that could mean that a sponsored article promoting a product gets some facts wrong or improperly represents a brand, product or topic.

And AI could be influenced by fake information. For example, if AI were asked to create sponsored content for a specific restaurant and the data sets included fake reviews posted on popular sites, the sponsored article could be skewed with false data. And without human review, the article may sound entirely normal, given that generative AI is generally good at syntax, Rahman said.

“Everything may sound right, but one line might be inaccurate,” he said. “That could reduce the trust in the brand.”

As with idea generation for products, professionals might find generative AI helpful in the brainstorming process for creating a logo or slogan. Generative AI tools could also help in communicating ideas.

“In being able to explain, ‘This is what I’m thinking for a logo,’ it helps bridge the expertise gap between a lay businessperson and a designer,” Rahman said. “It lowers the barrier.”

But given that a brand’s identity is so closely tied with these creative elements, a human probably will need to build on those ideas before solidifying the initial idea, Rahman said. AI might help professionals come up with a catchy phrase they want to use or even a style they want to mimic for the logo. But do those elements reflect a brand’s values and is the logo unique enough to stand out? That’s where human expertise comes in, Rahman said.

Professionals will also need to keep in mind possible issues with copyright and trademark infringement. And they may have to use at least a couple of AI tools, as was the case with this example, because image generators often fail at producing text.

Generative AI could help candidates more easily write a cover letter and even follow some of the best formats to do so, Rahman said. But it also may have a drawback if candidates want to break the mold.

“It is relying on what is expected and plausible,” he said. “If you want to stand out in a creative way, it may be less helpful.”

Because generative AI is just an advanced predictive text system that uses big swaths of data to pull answers from, it may help candidates who struggle with promoting their successes or writing in a smooth, coherent fashion. It also could help people who struggle with the English language, Rahman said. But if you want your cover letter to look different from what is traditionally expected, Rahman suggests you tweak the AI-generated letter, play with the prompts more or write it yourself.

Once again, an issue that could arise from using AI as an illustrator is infringing others’ original work without permission. Photographers and artists have filed lawsuits against some AI image generators over the issue, and if they win, any images created with those models could be considered problematic.

But even human illustrators often look to other people’s work as inspiration for things like book covers, Rahman said. So AI is somewhat simulating that, he said. Where we draw the line as to what is considered inspiration and what is stealing is still up for debate. In this case, for example, DALL-E is probably creating a children’s book cover based on what is publicly available.

“We’ve seen this play out in the music industry when people use beats from one song and mix them,” he said, adding that music cases often consider how much a song resembles the original work. “We’re going to have to have society decide … [when] we don’t have to give credit.”

Generative AI generally handles summarization well, especially if you’re feeding it with the data you want summarized, Rahman said. That could have big implications for sectors like law or medicine that rely on literature and documentation. But “the devil is in the details,” he added.

The technology may misinterpret some sentences or make inaccurate inferences, for example. It also may generalize elements that could be important to highlight for professionals who need to know a specific medical symptom or inference of a law. And when it comes to legal or medical literature, that could have a massive impact on outcomes.

AI transcriptions or summarizations could also change how professionals do their jobs. So if an AI tool is summarizing a patient interaction, a doctor might inadvertently pay less attention to the details the patient is offering, relying more on the technology to pick that up. That could play a role in a doctor’s ultimate determination.

“AI does excel in making connections humans don’t necessarily see,” Rahman said. “But you do need a human to verify and understand whether it has validity.”

Using generative AI to resemble a photograph is tricky. That’s because the creation produced by the AI is a conglomeration of images, so details that would be important in representing the reality of a photo could be distorted.

An image created by AI often has a distinct style, and avoids replicating a real photo, Rahman said. Some signs that a realistic image was generated by AI include blurring of specific areas, often in the background; distorted or incomprehensible text; and warped details like a single tooth of a person. Then there are more obvious giveaways like disfigurements of the body or items that magically disappear in half of the photo. In this case, people’s fingers blend together in the image and the man on the left’s arm appears to disappear and not fit to scale. Also, the background audience’s details seem to blur together, almost blending into their environment.

While generative AI can’t yet do the job of a photographer, it could serve as a brainstorming ground for photographers who are thinking of ways to conceptualize a photo shoot.