🔒 Beware OpenAI’s engagement-based GPT Store model – Parmy Olson

OpenAI’s latest move to monetise its language model technology through the “GPT Store” raises concerns as it adopts an engagement-based revenue structure. Developers will be paid based on user interaction with their GPTs, echoing the controversial models of tech giants like Facebook and YouTube. Despite OpenAI’s assurances against prioritising engagement, the store’s structure may incentivise addictive and potentially harmful AI services. As the GPT Store opens a new chapter, vigilance is crucial to prevent the negative consequences associated with attention-driven business models.

Sign up for your early morning brew of the BizNews Insider to keep you up to speed with the content that matters. The newsletter will land in your inbox at 5:30am weekdays. Register here.

By Parmy Olson

OpenAI is expanding beyond nifty products like ChatGPT and has announced that it is following through with its pledge to distribute AI’s benefits to all of humanity.

Just kidding!


It’s actually launched a “store” to monetize its language model technology and share that wealth with other businesses. And it’s using an incentive structure that has a history of unpleasant side effects on — you guessed it — humanity.

The “GPT Store” is available to OpenAI’s enterprise customers and anyone who pays $20 a month to use ChatGPT Plus, offering a selection of specialized versions of the tool in areas like research, education and design that are created by third-party developers. Think of them like apps that you talk to via chatbot. For instance, I asked the “AllTrails” GPT to recommend running routes in my town, and it provided a list of ideas with links to maps. “Coloring Book Hero” generated some pictures of copyrighted characters for my kids to scribble all over, and the “Books” GPT churned out a list of wilderness survival novels after I told it that I’d re-watched The Revenant.

OpenAI says developers have already created more than 3 million of these GPTs, which led The Atlantic to point out that this was ChatGPT’s “FarmVille Moment.” Indeed, this harks back to when Facebook allowed other software engineers to create apps for the site back in 2007, helping it become a sprawling platform. But OpenAI’s store is similar to Facebook in a more disturbing way too: It has an engagement-based revenue structure.

Builders “will be paid based on user engagement with their GPTs,” OpenAI has said, which means that the more popular, engaging and potentially addictive a GPT service is, the more money its developers can make. (The company said it will give further details on how those payments will work at some point in the first quarter.) OpenAI might be the world’s leading AI firm, but it is harnessing one of the most established business models on the internet.

Companies from Facebook to YouTube to e-commerce sites and news organizations use engagement as their primary metric for paying, or charging, third parties. That has made attention the currency that drives our online economy. Facebook’s business model, for instance, revolves around maximizing user engagement to drive ad revenue. The more time people spend on the platform, the more ads they can be shown. YouTube pays creators based on the number of views their videos receive, while TikTok pays influencers based on their followers’ engagement with their posts. Many news sites like Forbes.com have also paid writers based on page views. 

The great benefit of this model has been to make most of the content on the internet free. But there are plenty of downsides too. By prioritizing content that keeps people on Facebook for longer, its algorithms often show posts that spark outrage. That has created political echo chambers and lured people into groups that spread conspiracy theories.

YouTube’s model has led creators to focus on sensational or controversial content to drive views, motivating influencers to post provocative videos, while news sites have gravitated over the last decade toward sensationalist, click-bait headlines, many of them focused on trending topics instead of in-depth journalism. The engagement-based model has even affected TV. David Chase, the creator of The Sopranos, recently complained that the golden age of television had come to an end after he was warned by executives to not make shows that “require an audience to focus.”

Sam Altman has previously denied that OpenAI used engagement as part of its business model. When he testified before Congress last year, the OpenAI CEO concurred with a lawmaker who warned the company against repeating the mistakes of social media networks. “We try to design systems that do not maximize for engagement,” Altman replied. “We’re not an advertising based model. We’re not trying to get people to use it more and more.” But other developers likely will, and the GPT Store doesn’t have to be ad-based to maximize for engagement. Subscription models like Netflix Inc. do that too, and are leading members of today’s attention economy

One thing that seems to be drawing people’s attention to AI is relationships. Based on a cursory search of what’s currently available, dozens of developers have already submitted “girlfriend” bots to OpenAI’s store. Users can only seek relationship advice or engage in friendly chat with these bots, since OpenAI’s forbids developers from launching ones that foster “romantic companionship.” Yet with more than three-million-and-counting GPTs to oversee and likely more to come, it’ll get harder to ensure all of them follow those rules. 

OpenAI is cracking open a Pandora’s Box of potential side effects with its adoption of a business model that has already reshaped human behavior in repugnant ways. Social media companies whose algorithms subtly encouraged toxic activity often failed to enforce their guidelines banning that behavior. OpenAI can set itself apart from its predecessors if it can enforce those guidelines properly. That would be a welcome change, but it will be tough to pull off. 

Read also:

© 2024 Bloomberg L.P.