Google will join the AI wars, pitting LaMDA against ChatGPT

Of the world’s large tech firms, Alphabet Inc. may be caught most deeply in the innovator’s dilemma. The classic theory from Harvard Business School Professor Clayton Christensen says large companies struggle to innovate because they fear hurting an established business. Alphabet has been in that bind for the last few months, coming under tremendous pressure to respond to ChatGPT, the OpenAI tool that could reinvent internet searching with its remarkable conversational answers to any question. But Google has to be cautious: its $150 billion search business makes money every time we click on ads and links; single, synthesized answers to queries could draw those clicks away.

Now Alphabet CEO Sundar Pichai has decided he has little choice but to take that risk.

On Thursday, as Pichai announced fourth-quarter results that narrowly missed analysts’ estimates, he said that Google would make its AI-based large language model known as LaMDA available “in the coming weeks and months,” and that people could use it as “a companion to search.”

A pledge to “focus on AI” is par for the course in Big Tech right now. Mark Zuckerberg said much the same thing this week when he described how AI would help improve Facebook’s products. But Google has more street cred than perhaps anyone else when it comes to artificial intelligence. It owns DeepMind, one of the pioneers of reinforcement learning and deep learning, a cutting-edge approach to AI, and Google researchers invented the “transformer” technology that powers ChatGPT.

Google also has LaMDA, a large-language model trained on billions of words on the public internet, just like ChatGPT’s model. But there’s a good chance that LaMDA is even better. It benefits from a broader array of research talent at Google and huge amounts of computing power, not to mention feedback from millions of users for constant fine-tuning. One of Google’s own engineers even believed LaMDA was sentient after chatting with it.

But the innovator’s dilemma has forced Google to keep LaMDA hidden away, fearful it could cannibalize its own search results or make offensive remarks and wild mistakes. ChatGPT has become well-known for its frequent inaccuracies, and OpenAI can get away with that as a smaller company. Google, with its 3.5 billion searches per day, has no such luxury.

The critical question now is how Google will incorporate a chatbot into search, since people aren’t using ChatGPT in the same way they use Google. Consider that Google’s top 10 search terms are all brand names like Facebook, YouTube and Amazon. This is because people frequently use Google to navigate to other sites. Other popular terms are often transactional, like “restaurants near me” or “Samsung Galaxy phone.” These aren’t terms that you would use for ChatGPT, and they are lucrative for Google.

Google will probably create an additional search category for something like “conversational answers,” to sit alongside images, maps and news, and this “companion” will serve the long tail of informational search queries that don’t make Google much money, such as finding a recipe or looking up a historic event.

Where it has to be extremely careful is in making sure its own version of ChatGPT doesn’t become liable for bad advice. It’s become second nature for people with an ailment to Google their symptoms, but if they go to the company’s new search companion and get bad medical advice, that could turn into another legal headache for Google and a danger for users.

Like it or not, Google must adjust to a world where content is increasingly generated by AI. The news site CNET has been using AI to generate financial-advice articles (more than half of them with errors) while online influencers are encouraging people to use ChatGPT as their own automated content farm. As the web gets flooded with even more spam than before, Google’s search algorithms will have work harder to rank quality content.

The company has been scrambling to define its AI strategy, declaring an internal “code red” after ChatGPT became popular and reassigning teams to work on new AI prototypes. Pichai has even called back Google founders Sergey Brin and Larry Page to help strategize Google’s response to ChatGPT.

But while Google previously bought other firms to help it build a powerful and lucrative ad tech stack, it can’t similarly buy its way to a stronger foothold in generative AI. Pressure from antitrust regulators is growing as a lawsuit from the Department of Justice threatens to break the company up.

The painful irony for Google now is that it faces being penalized for being both too dominant and not competitive enough. To toe that line, it must now put aside its famously cautious approach to innovating and use the expertise it has in house to meet the challenge from Microsoft-funded OpenAI. The fact that Pichai has said Google could launch its new service “in weeks” shows how much he sees ChatGPT as a threat. Google normally doesn’t move this fast. It just has to be sure it doesn’t stumble as it rushes out its secret weapon.

More From Bloomberg Opinion:

• Google Faces a Serious Threat From ChatGPT: Parmy Olson
• Making Sense of Sensory Overload in the Markets: John Authers
• Beware ChatGPT Trying to Teach Your Kids Math: Parmy Olson

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

Parmy Olson is a Bloomberg Opinion columnist covering technology. A former reporter for the Wall Street Journal and Forbes, she is author of “We Are Anonymous.” Bloomberg