A Wharton professor gave A.I. tools 30 minutes to work on a business project. The results were ‘superhuman’
Artificial intelligence is presenting new possibilities on how to do work, and leaving many observers nervous about what will become of white-collar jobs.
Ethan Mollick, a management professor at the Wharton School of the University of Pennsylvania, has been closely following developments in generative A.I. tools, which can create essays, images, voices, code, and much else based on a user’s text prompts.
He recently decided to see how much such tools could accomplish in only 30 minutes, and described the results this weekend on his blog One Useful Thing. The results were, he writes, “superhuman.”
In that short amount of time, he writes, the tools managed to do market research, create a positioning document, write an email campaign, create a website, create a logo and “hero shot” graphic, make a social media campaign for multiple platforms, and script and create a video.
The project involved marketing the launch a new educational game, and he wanted A.I. tools to do all the work, while he only gave directions. He chose a game that he himself authored so that he could gauge the quality of work. The game, Wharton Interactive’s Saturn Parable, is designed to teach leadership and team skills on a fictional mission to Saturn.
First, Mollick turned to the version of Bing powered by GPT-4. Bing, of course, is Microsoft’s search engine—long a distant to second to Google—while GPT-4 is the successor to ChatGPT, the A.I. chatbot from OpenAI that took the world by storm after its release in late November. Microsoft has invested billions in OpenAI.
Mollick instructed Bing to teach itself about the game and the business simulation market of which it’s a part. He then instructed it to “pretend you are a marketing genius” and produce a document that “outlines an email marketing campaign and a single webpage to promote the game.”
In under 3 minutes it generated four emails totaling 1,757 words.
He then asked Bing to outline the webpage, including text and graphics, and then used GPT-4 to build the site.
He asked MidJourney, a generative A.I. tool that produces images from text prompts, to produce the “hero image” (the large image visitors encounter first when visiting a website).
Next, he asked Bing to start the social media campaign, and it produced posts for five platforms, including Facebook and Twitter.
Then he asked Bing to write a script for a video, an A.I. tool called ElevenLabs to create a realistic voice, and another called D-id to turn it into a video.
At that point, Mollick ran out of time. But, he notes, if he’d had the plugins that OpenAI announced this week, his A.I. chatbot, connected to email automation software, could have actually run the email campaign for him.
According to OpenAI, plugins for Slack, Expedia, and Instacart are among the first to be created, with many more to come. The problem with A.I. chatbots, the company notes, is that “the only information they can learn from is their training data.” Plugins can be their “eyes and ears,” giving them access to more recent or specific data.
Mollick writes that he would have needed a team and “maybe days of work” to do all the work the A.I. tools did in 30 minutes.
Bill Gates wrote on his blog this week that ChatGPT and similar tools “will increasingly be like having a white-collar worker available to help you with various tasks.”
Actual white-collar workers might be forgiven for feeling some anxiety.
