AI chatbots save time, but researchers struggle to find the economic impact
Also: Mark Zuckerberg wants to redefine the category of advertising, AI exceeds humans in persuasion, and more.
ChatGPT recently reported hitting 20 million paying subscribers, among them undoubtedly a large share of professional users, but apparently they struggle to convert the investment to a tangible, monetary upside. A new study looks into the economy of AI chatbots, which you can read about below, among other selected stories.
Number of the Week
70%. Share of McKinsey’s 45,000 employees that now use their internal chatbot, the company reports. The tool, called Lilli, is trained on the consulting firm’s entire body of intellectual property spanning 100 years and 100,000 documents. [Source]
AI chatbots save time, but researchers struggle to find the economic impact
Led by ChatGPT, AI chatbots have been a hot topic for businesses since the prominent launch two and a half years ago, but direct financial results from the phenomenon are hard to come by both for workers and employers.
That is the conclusion in a new study from Denmark, where researchers from University of Chicago and University of Copenhagen surveyed 25,000 workers across 7,000 companies about their usage and results from AI chatbots.
The participants were specifically chosen to be among 11 professions where AI can save time, such as accountants, journalists, and marketing professionals.
And they do recognize results from its use: overall, the respondents report time savings of 2.8%, about one hour a week, and nearly half cite improved work quality and enhanced creativity as key benefits.
The advantages, however, do not transfer to their earnings, as 97% of workers reported no change from AI chatbots usage.
Do the employers reap a benefit then? The researchers looked at company’s total earnings before and after ChatGPT, and there is no substantial change.
”Across the board, workplaces with higher rates of AI chatbots adoptions have not fared differently,” the researchers, Anders Humlum and Emilie Vestergaard of University of Chicago and University of Copenhagen, respectively, write, concluding:
“Despite rapid adoption and substantial investments by both workers and firms, our key finding is that AI chatbots have had minimal impact on productivity and labor market outcomes to date.”
How Mark Zuckerberg wants to redefine the category of advertising
Meta raked in around $160 billion in revenue from ads last year, and now the Facebook owner wants to further enhance its position in the industry by a series of AI-powered initiatives.
In an interview with tech blog Stratechery, Meta’s CEO Mark Zuckerberg lays out a vision of the company being a sort of one-stop-shop for businesses.
“[We want to] make it so that any business that basically wants to achieve some business outcome can just come to us, not have to produce any content, not have to know anything about their customers. Can just say, “Here’s the business outcome that I want, here’s what I’m willing to pay, I’m going to connect you to my bank account, I will pay you for as many business outcomes as you can achieve”.”
Already, Meta recognizes that it is often better than their customers at knowing which users to target in achieving certain outcomes.
This could also stretch to the creative execution of the ads, so that in the end, the business users will just have to provide their objective.
“I think that’s going to be huge, I think it is a redefinition of the category of advertising,” Zuckerberg says.
AI exceeds humans in controversial persuasion experiment
Researchers from the University of Zürich have caused quite a stir on Reddit after they disclosed a secret experiment that they’ve been running on the online forum.
For four months, the academics deployed AI-automated accounts to take part in discussions in the “Change My View” subreddit, where users can post a view and let others see if they can convince them to modify their stance.
The aim was to see how effective large language models were for this compared to humans, and three different strategies showed better results with persuasive rates between three and six times higher than the human baseline, the researchers write in a since retracted preliminary article on the work.
The most effective strategy, which managed to convince users to change their position 18% of the time was personalization, where the large language model analyzed the users’ background compared to their latest comments and modified their arguments for this. In contrast, human users successfully change an opinion 2.7% of the time.
“We recognize that our experiment broke the community rules against AI-generated comments and apologize,” the researchers wrote to the moderators of the Change My View community:
“We believe, however, that given the high societal importance of this topic, it was crucial to conduct a study of this kind, even if it meant disobeying the rules.”
After disclosing the experiment, Reddit has since taken legal action, as the deceiving behavior from the LLMs is against their rules, and the University of Zürich has also since issued a formal warning to the researchers for not following the recommendations of their ethics committee. However, the institution will not hinder the disclosure of the results:
"This project yields important insights, and the risks (e.g. trauma etc.) are minimal. This means that suppressing publication is not proportionate to the importance of the insights the study yields."
Image of the Week
You’ve probably seen the ChatGPT-generated action figures. James Bricknell, reporter at CNET, took it a step further, transforming the AI-generated image of his colleague into a 3D printed figure. [Source]
Exciting news out there?
The self-driving car Waymo is involved in substantially fewer accidents than humans in a comparison over 91 million kilometers in four American cities. For instance, Waymo is involved in 96% fewer crashes at intersections, the most frequent accident. The study is peer-reviewed about to be published in Traffic Injury Prevention. [Source]
Feedback?
Thanks for reading, and feel free to respond directly with suggestions for improvement, tips, and anything in between :)