Some technology waves change the world — notably the world wide web, social media, and smartphones. Others don’t — such as virtual reality and blockchain.
It may be too soon to decide where AI chatbots fit. But despite the massive hype and enormous investment — which contributed to a near-doubling of ChatGPT provider OpenAI’s valuation to $157 billion, according to the New York Times — generative AI appears unlikely to change the world.
What makes the difference? Waves that matter pass three tests:
- They make life much better for people.
- Many people eagerly pay a high price for the products springing from the new technology.
- The resulting profits more than offset the investment required to build and continue to improve those products.
Sadly, generative AI does not pass any of these tests.
Before getting into the reasons why, proponents of generative AI argue the best is yet to come. And while that could happen, I have observed ChatGPT doing interesting things — but so far nothing significant enough to change the world.
For instance, in January, one of my Babson College students uploaded a book I assigned for a course into ChatGPT. He told me he spoke his questions to ChatGPT and the book responded in “a very high quality” voice. In September, another student said she had trained ChatGPT to perform various tasks for her startup – she referred to these AI-performed tasks as her employees.
Generative AI Has Not Yet Made Peoples’ Lives Much Better
To change the world, a new technology must make peoples’ lives much better than they were before – or in tech-speak, the product must be a killer app.
A case in point is Apple’s iTunes Store – which resulted in a near-quadrupling of iPod sales. Introduced in 2003, two years after Steve Jobs launched the iPod, the iTunes Store enabled people to pay 99 cents per song that could be played on a PC, burned onto a CD, or transferred to an iPod. The quarter before the iTunes Store launched, Apple sold 78,000 iPods — the quarter after, iPod sales soared to 304,000, according to the Harvard Business School case Apple Inc. In 2015.
The iTunes store was a killer app because it made the iPod so much more useful for consumers. For example, joggers who formerly listened to music on a Sony Walkman flocked to the iPod. Why? It was smaller, lighter, and enabled them to customize playlists.
Generative AI so far lacks such a killer app. The small number of students who are using ChatGPT is anecdotal evidence that the technology is not making their lives much better.
For instance, in that September Babson class. only two other students used ChatGPT occasionally for research. Due to hallucinations, the students had to double-check the results – which probably means a Google search is a more efficient way to get answers.
Hallucinations are a feature — rather than a bug — of large language models. That’s because they are trained on data — not all of which is accurate — to make predictions about what the next word in a sentence might be. Sometimes the LLM guesses right, sometimes not.
Generative AI’s flaws help explain why a relatively modest number of people use the technology. In September 2024, ChatGPT had 250 million weekly active users, OpenAI CFO Sarah Friar told CNBC — a drop in the bucket compared to Facebook’s 3.1 billion MAUs and Instagram’s more than 2 billion MAUs, BackLinko notes.
Meanwhile, the average ChatGPT visit lasts a bit more than three minutes – small compared to Facebook and Instagram’s more than 11 and 8.5 minutes, respectively reported SimilarWeb.
To be sure, OpenAI expects revenue to nearly triple to $11.6 billion in 2025, according to reports from anonymous sources cited by CNBC.
Users Do Not Pay Enough To Cover ChatGPT’s Costs
While ChatGPT and Microsoft Copilot have been heavily hyped, neither are generating enough revenue to cover their costs. For instance, in 2024 OpenAI expects to generate about $3.7 billion in annual revenue while spending $8.7 billion — producing a $5 billion loss, noted the Times.
While Microsoft declined to quantify its Copilot revenue in the most recent quarter, the AI-powered assistant’s costs are significantly higher than its revenues. For example, GitHub Copilot, a service that helps programmers create, fix and translate code, costs between $20 and $80 per month to run — significantly exceeding the service’s $10 per month subscription fee, according to the Wall Street Journal.
People are not willing to pay enough for ChatGPT and Copilot because they fail the critical test of a killer app — they do not relieve customer pain more effectively than current products. Performance and cost issues with Copilot are causing customers to pause their $30-per-month-per-user Copilot contracts for Office 365, according to The Information.
For instance, I wish I could get back the time I lost trying to use Copilot to help me write an article. I prompted Microsoft’s AI assistant to answer four questions about a BusinessInsider report of a pharmaceutical company that canceled its contract for Microsoft’s AI chatbot. Copilot ignored my first three questions and provided vague answers to the last one.
In August, I asked ChatGPT4o to read through my recently published book Brain Rush and return a story that potential readers would find compelling. Sadly, that story was completely made up. When I prompted ChatGPT to try again to find a story from the book, it confidently presented me with another bogus tale.
Profits Could Fall Short Of The Investment Required To Build Generative AI
Companies will spend over $1 trillion on generative AI in the years ahead, according to Goldman Sachs. Meanwhile, data centers that train and operate LLMs are scrambling to obtain enough electricity to keep them running.
Indeed, in September 2024, Microsoft spent an estimated $16 billion on a 10-year contract with Constellation Energy — the company that operates Three Mile Island — one of whose reactors famously melted down in 1979, according to my September 2024 Forbes post.
Other data centers must wait until the 2030s to get access to sufficient electricity to meet their generative AI needs while the electricity requirements of a ChatGPT query are 10 times that of a Google search, reported the Wall Street Journal.
As I noted in my Value Pyramid case study, most generative AI use cases help people overcome creator’s block — such as the anxiety about writing an email. Fewer generative AI applications help improve the productivity of business functions such as customer service or coding. And few, if any, applications of AI chatbots enable companies to add new sources of revenue.
It will take at least one history-making killer app that drives such expectations-beating growth to justify that gigantic investment. In the absence of such a revenue-driving application, with a value per active user of $628, OpenAI appears very expensive compared to Meta Platforms’ $285.
Until leaders design and deploy AI chatbots to make life much better for many people, the promise of generative AI could fall short.