Google, Beware: Ads and chatbots are a dangerous mix

The ad model has made the internet free and accessible, it also has created a kind of legalised addiction where people are constantly being sucked into distracting content funded by banner ads and pop-ups. (Source: Bloomberg)

Google and Microsoft Corp are both exploring ways to integrate ads into their versions of ChatGPT. We should all brace for the consequences.

In order to ensure that people see as much advertising as possible, there is a good chance both companies will become incentivised to make their chatbot tools engaging, and even addictive. It’s the same impetus that pushed social media platforms to keep people rapt at all times, and fostered the spread of misinformation as well as mental health problems.

According to the New York Times, Alphabet Inc’s Google is scrambling to develop an entirely new search engine that offers a “more personalised experience” using generative AI, the hot new technology underpinning ChatGPT. Both ChatGPT and Google’s rival chatbot Bard use so-called large language models that predict sequences of words to give humanlike answers to queries.

Meta Platforms Inc’s Facebook and Amazon.com Inc are also racing to add these AI tools to their own online services. But while tech firms haven’t spelled out how they will make money from the new technology, it’s clear that advertising will be in the mix on some platforms.

For instance, Google’s Project Magi, an umbrella for the generative AI features it’s bringing to search, will keep ads in the mix of search results, according to the Times.  Any transactional search queries about buying goods, such as shoes or flights, will feature ads on the results page.

Microsoft also has confirmed its AI-powered Bing chatbot will incorporate ads as sponsored links or in shopping results. Here’s one example where ads responding to a question about the cheapest Honda car were signposted with small citations.

Social media companies have been around the block with this problem. The misinformation and mental health issues that correlate with time spent on Facebook or Instagram, for instance, come in part from those apps being designed for maximum engagement.

One of the great failings of the interne has been an inability to find a model for online businesses other than advertising. While the ad model has made the internet free and accessible, it also has created a kind of legalised addiction where people are constantly being sucked into distracting content funded by banner ads and pop-ups. It’s why doom scrolling can be so lucrative.

What makes the arrival of tools like ChatGPT more worrisome is that they could supercharge advertising because of their incredible potential to persuade and manipulate. Large language models are trained to “make up believable things in a human-like way,” former Google AI researcher Margaret Mitchell recently told me. These models often make factual errors, but the bigger and more powerful they get, the better they are at sounding persuasive.

And while ads in Google search or on websites are generally clearly marked, a conversational response from a chatbot might make it more likely that people accept what they see in front of them as “the truth,” according to Sridhar Ramaswamy, the former head of Google’s advertising business and founder of rival search engine Neeva. “Of course, this can lead to unfortunate consequences.”

Remember a few months ago when the new Bing chatbot told a newspaper columnist that it was in love with him and suggested he leave his wife? That bizarre encounter illustrated how compelling a chatbot can be when its creators allow it to show more personality. Kevin Roose, the New York Times writer who had the conversation with Bing, found himself talking to the bot for hours, mesmerised by the comments it was generating. Microsoft ended up having to recalibrate Bing to a more boring, neutral set of responses.

If Microsoft finds itself in the midst of a bitter rivalry with Google over the market for ad-driven search, it may want to bring some of that old personality back. A more fun and interesting version of Bing or Google’s Bard would attract more users and keep them using the tools for longer, seeing more ads.

Facebook has done this brilliantly — and at great social cost — by designing algorithms to show content that provokes emotions like outrage to keep people scrolling and clicking. The company’s engineers have monitored people’s likes and comments to see which models powering the newsfeed maximise engagement. The models they use have often
promoted controversy, misinformation or extremism.

For now, Bing and Bard aren’t designed to maximise engagement. But neither was Facebook when it first exploded in popularity among college students in the mid-2000s. Back then, it was just a utility for connecting with friends and family, like ChatGPT, Bing and Bard are currently  practical tools for writing or researching.

But Google and Microsoft have ambitious plans for these tools to generate new revenue streams, and Google in particular looks likely to rely on its tried-and-tested ad model to do so. Google will have to ensure it isn’t inadvertently sending its users down rabbit holes of emotional manipulation or conspiracy theories, something its YouTube division has had plenty of experience with.

And if you don’t think chatbots can be alluring, consider that many people have already developed friendships and even romantic relationships with AI companions on Replika, an app that has been downloaded more than 5 million times. The potential for chatbots to act as emotional outlets, therapists or entertainers could be even greater than their role as search engines, as their tenuous relationship with facts continues to be a problem.

Microsoft and Google should tread carefully as they move to capitalise on Bing and Bard. Advertising might be the obvious way to keep their services free for all, but history has shown that model also has its costs.

Parmy Olson is a Bloomberg Opinion columnist covering technology. Views are personal and do not represent the stand of this publication.

Credit: Bloomberg