Editor’s note: This is a guest commentary from Oren Etzioni, former CEO and current technical director at the Allen Institute of Intelligence, a Seattle-based organization at the forefront of natural language processing research.
The integration of ChatGPT into Microsoft’s Bing search engine heralds a tectonic shift for consumers, publishers, and advertisers on the web.
In recent weeks, Bing has gotten a lot of buzz, and Google has lost $100 billion-plus in market capitalization. But changes over the next five years will be far more profound.
What seemed like a staid and steady web search industry, largely monopolized by Google, has been thrown into disarray by the fact that content generation is now instantaneous, fully automated, and its cost is rapidly dropping toward zero.
This shift means that authoritative sources and genuine experts will be more important than ever.
For consumers, more efficient search has a long history which includes Google’s knowledge panels (2012), featured snippets at the top of the search result page (2014), voice assistants such as Alexa (2014), and now ChatGPT.
Over time, a concierge experience will emerge where a consumer can ask a question, receive an answer from a chatbot, and engage in a dialogue to further refine the response.
This could be good news for consumers, but it also raises a thorny question: who is the concierge working for?
As a consumer, I hope for an objective and informative answer but the chatbot will not necessarily oblige.
- A chatbot’s answers depend on its training text which contains myriad biases.
- The chatbot might be manipulated through its training process (similar to the practice of Google bombing).
- Different chatbots will emerge that represent particular perspectives (the GOP bot), commercial interests (the ExxonMobil bot), and specific individuals (would you like to converse with Biden bot? The Kim Kardashian bot?).
- And of course, chatbots could be influenced by advertisers.
Historically, search engines distinguished between search results and sponsored or “featured” results (i.e., ads) but product placement may surface inside chatbot responses, undermining their credibility.
In response, consumer advocate chatbots will emerge, charging a subscription fee instead of being ad-supported. As chatbots proliferate, search engines will emerge that help the consumer find the “right” bot for a conversation. Meta-bots could collate multiple responses to a question, each originating from a different chatbot.
Consumers will be inundated with an unprecedented amount of automatically generated “noise” in the form of websites and messages — emails, posts, responses in social media, and more.
The minimal cost of generating seemingly authentic text (along with pictures, audio, and even video) will result in unprecedented information pollution and even AI-based forgery.
In response, I have argued for a stronger role for digital authentication of identity (who actually wrote that message?) and for rules that require bots to identify themselves. Consumers have a right to know if we are interacting with a person or a bot.
As the volume of content increases, publishers will face unprecedented pressure to remain relevant, discoverable, and valuable. Certainly, clickbait websites with titles like “top 10 things to do in Seattle” will be replaced by more personalized and up-to-date chatbot responses. Collections of reviews found at Amazon or on Google Maps will remain informative only if the reviews are appropriately authenticated — otherwise, it will be all-too-easy to create volumes of fake reviews.
Likewise, the information on social media (popular posts, for example) will only be meaningful if popularity isn’t manipulated by bots. Authoritative sources will become even more essential as people clamor for reliable facts in a maelstrom of misinformation. Brands and reputations will be built on providing genuine, authentic answers.
In a world where “what” is said is so easily manipulated, “who” said it becomes increasingly important.