The progress of artificial intelligence models over the past few years has been faster than almost anyone expected. Some advances have left society scrabbling to adapt. Teachers are struggling to stop students using chatbots to write their essays, artists claim they are losing paid work to image-creating AIs and efforts are under way in some places to replace journalists with large language models. But bigger changes are afoot.
Google, Microsoft and Baidu, which run the three most-used search engines in the world, have all announced that their services will be upgraded, or at least augmented, with AI chatbots. Rather than entering a simple query and being presented with a list of potentially relevant websites, users will instead be able to ask niche and detailed questions and get a bespoke answer.
It sounds like a simple change, but the ramifications could be far wider and more fundamental than they might seem at first, including damaging online shops, making more services subscription based, threatening endeavours like art and journalism and shaking up the online advertising industry.
Advertisement
The potential benefits to having an AI chatbot answering your search queries could be large. Want instructions on fixing a flat bicycle tyre? Check. Want them written as a Shakespearean sonnet? No problem. You can ask it to plan a holiday that considers the whims and tastes of every member of your family, or even to write an application letter for a new job. The answer doesn’t have to appear anywhere on the internet, but the AI will still craft an appropriate response.
at digital media agency says that AI-powered search engines could be transformational for users, increasing efficiency and productivity – if the problems AIs sometimes have with accuracy can be solved.
Quick, no-fuss visits that start and end at a search engine could be good news to many, but this set-up is conceivably even better news for the tech firms that run these programs. The Google and Baidu search engines, and Microsoft’s Bing, will be able to reduce how much they pass people on to other websites and instead keep them engaged on their own, where they can be shown advertisements and monetised.
This lack of movement across the internet is where one big issue emerges. The sea of varied answers and potentially opposing viewpoints we see as we go through a list of search results would be replaced by output from a single entity.
at the University of Oxford says that such centralisation risks creating an internet where “echo chambers might become smaller and more solidified”. This could mean people mainly coming across opinion and information that reinforces their own viewpoints, rather than being exposed to that from other communities.
Because this single AI would be controlled by a company reliant on profits, there is also a danger its responses might be tailored to serve up certain content, says VĂ©liz.
“It worries me that they might get income from ads publishers influencing the output of the AI,” says Véliz. “The AI could end up recommending the brands that are paying their bills.”
Instant content creation
Such outcomes aren’t unprecedented. Google has previously been by European Union regulators for giving advantages to its own shopping comparison service. And although the workings of AI models are mysterious, the firms behind them can still steer or influence output or the language used.
An AI model would also be able to deliver instantly created content, such as a short story about your pet dog to entertain your toddler. Doing this rather than diving into the diverse wealth of human-generated writing, images, video and art online risks homogenising culture itself, says at the University of Birmingham, UK.
He imagines a time when blog posts for prominent people and organisations are written automatically by the same AI, bands are fronted by neural-network generated images and artwork for a range of private and commercial settings is quickly and cheaply generated by entering a text prompt. This ersatz art will be good enough to look convincing at first glance, but will be bereft of real creativity, he says.
“If you believe punk was a spontaneous cultural happening, then it’s not likely to be repeated with AI,” says Lee. “I think one thing AI can’t do is explain either the meaning or story behind its art, so perhaps it could generate a Guernica but wouldn’t be able to contextualise it in terms of [its painter] Picasso and his reaction to the Spanish civil war.”
We are at a point at which AI chatbots are technically impressive, but not yet totally reliable factually. This lack of accuracy presents another potential problem. Language models like OpenAI’s ChatGPT, which Microsoft is integrating into Bing, essentially output a string of words based on the statistical likelihood that they would appear together in the real world. However, this isn’t the same as assessing accuracy or objectivity.
Prominent AI errors
, Sports Illustrated and have all recently dabbled with using AI to write articles for them, although the last quickly rolled back the idea after finding the output riddled with errors.
AI-powered search engines have also got off to a rocky start. Google ran an advert for its Bard AI model in which it erroneously claimed that the James Webb Space Telescope took the first images of an exoplanet. A Bing demonstration by Microsoft revealed . Google, Baidu and Microsoft didn’t respond to a request for comment for this article.
Incorrect results may be little more than an inconvenience if you are searching for a recipe, but they could be life-threatening if you are hunting for medical advice in an emergency.
Errors aside, a change to the way search operates could fundamentally alter the internet. Search engines like Google crawl the web, following links from page to page and building up a vast database of information about them: what they contain, how many other websites link to them and therefore how influential and trustworthy they are. This information is used to tailor results and bring the most relevant information to the person searching.
When people leave the search engine to visit one of those suggested sites, that site receives readers and often gets advertising revenue based on the number of visitors. Once search engines serve up more AI results, then this stream of traffic will dry up because people won’t need to leave the search engine. If Google decides to add AI recommendations for goods or services, then the number of customers on unrecommended shopping websites is likely to plummet overnight.
Online journalism, too, is likely to struggle, for example. How would newspapers or magazines fund themselves if internet traffic – like print sales before it – plunges?
at the London School of Economics says AI will allow some publishers to automate a lot of journalism, potentially putting people out of work. “People need journalism, they don’t need journalists,” he says.
There will remain a market for insightful analysis online that AI cannot provide, though, says Beckett, even if straightforward news journalism might be written by AI. But a reduction in people arriving from search engines is likely to push more firms towards subscription models with loyal readers, rather than seeking the advertising revenue that comes with clicks, he says.
“The optimist in me says that, in this new world, there’s going to be an added premium on the human, the creative, the critical, the independent, the partisan, the personalised human stuff,” says Beckett.
Shaking up advertising
A reduction in internet traffic won’t just affect sites that are visited less. Search engines will also lose advertising revenue if they are no longer sending traffic through to third-party websites, says Gandhi.
“The whole advertising model is going to be really shaken up,” says Gandhi. This will mean subscription models for search engines are probable, too, and it will alter how websites of all types make money, he says.
Shops and other companies will need to innovate to avoid being cut out of the loop, says Gandhi, with email and social media becoming even more important in driving sales.
Legislation may be needed from regulators to help with readjustments, to change the way that the owners of content – be it art, text or sound – and personal data can retain rights to it, and adjust the ability to profit from it in a world where AI uses it to train and learn, he says.
“There’s a huge, huge trademark and intellectual property set of questions that have to be answered and played out. I think we’re very behind on AI regulation and law, and I think it’s going to be our biggest challenge and blocker to progress. Because all these regulations take many years,” says Gandhi.
The recent public mistakes by AIs demonstrate that most people are unlikely to be using AI chatbots in their searches in the short term, says at New York University. But we should certainly prepare for their widespread arrival.
“The risks to truth and trust are enormous. Few people are paying enough serious attention to them. Democracy itself is very much threatened,” he says. “Chat-style search, if it worked, could hurt or kill organisations if it doesn’t link out to them.”
