Deepmind’s Sparrow chatbot could launch in beta this year. Is this a way for Google to get out of the “innovator’s dilemma”?
OpenAI’s ChatGPT is the biggest AI hype of all time – and would make Google fear for its own core business because it can, among other things, give more or less reliable answers to questions. It’s supposed to be the domain of the search engine giant.
Google has answers to ChatGPT
Google might respond to ChatGPT. It showed great dialog-optimized language models, such as LaMDA and Flamingo, even before OpenAI’s successful product. With Meena, Google already had a chatbot ready that could have believable conversations with humans by the end of 2020.
So far, the company has not made a product of its research. According to his own statements, this is mainly due to security issues, although other reasons may also play a role – more on that later.
But ChatGPT and especially the strong involvement of Microsoft in the dissemination of OpenAI put pressure on Google. The company must at least prove that it can control OpenAI or even surpass it.
Deepmind Founder and CEO Demis Hassabis now discusses Time’s perspective that Deepmind’s Sparrow chatbot will go into “private beta” later this year. This is notable because Deepmind has so far operated primarily as an AI research institute, developing technologies that Google then incorporates into consumer products.
Deepmind introduced Sparrow in September 2022. Like ChatGPT, the chatbot is trained with human feedback, which Deepmind says makes it more useful, accurate, and harmless. Additionally, Sparrow will have access to the Internet through Google, which will allow it to incorporate up-to-date information into its responses. According to Time, the planned beta chatbot will produce additional sources that match AI responses.
Sparrow is based on Deepmind’s Chinchilla language model, which has few parameters but was trained with lots of data. The language model, which was introduced in April 2022, outperformed GPT-3 in common language benchmarks. However, ChatGPT is based on the more advanced version 3.5 of GPT.
In any case, there are good reasons to believe that Sparrow would work as well or better than ChatGPT – and Google would also have more powerful language models like PaLM up its sleeve.
Will Deepmind help Google get out of its “innovator’s dilemma”?
Why hasn’t Google yet responded to ChatGPT or even introduced a similar model before OpenAI, even if it could? The reasons officially given by Google – shortcomings in the reliability and security of large language models – are credible and probably play a role.
At the scale of Google, a chatbot spreading lies or hate speech would pose a significant reputational risk. Or worse, a chatbot so believable that people give it a conscience. Who wants to provide the hotline for something like this?
OpenAI’s ChatGPT, though growing rapidly, still only has a fraction of Google’s users.
The most important reason, however, could be that Google is stuck in the “innovator’s dilemma”. The term was coined by Clayton Christensen in 1997 to describe a situation in which established companies find it difficult to adopt new technologies or business models that disrupt their traditional markets.
These companies often have significant resources and capabilities (Google), but their existing customer base and internal processes prevent them from taking advantage of disruptive innovations. This can ultimately lead to the demise of the business because smaller, more nimble competitors are entering and taking over the market (OpenAI).
Google’s core business, search, is growing and very profitable. The company prints money with every page of search results it displays.
A search chatbot would need a new monetization strategy, and it’s unclear if it could be as profitable as Google’s current search. Google could launch a successful chatbot, make it potentially profitable – but if the profits are lower than Google’s current search, the company would still lose. This is Google’s dilemma.
A possible outcome for Google to at least achieve a smoother transition: the company is launching a business-oriented chatbot through its sister company Deepmind and builds its own competing product of Google Search – and OpenAI.
This would allow Google to reassure its shareholders and react quickly if chatbots emerge as a serious search alternative. This is where Deepmind’s Sparrow beta, announced for 2023, could come into play.