Google DeepMind Forms to Rival OpenAI's 100M User ChatGPT

In a major shake-up aimed at rival OpenAI, Google has unleashed a new weapon in the AI arms race: Google DeepMind. By combining the forces of its two elite AI research teams, Google Brain and DeepMind, the tech giant is making a bold statement about its commitment to winning the battle for AI dominance. This merger could be a game-changer, bringing together some of the brightest minds in the field to create the next generation of powerful AI models.

Google vs. OpenAI 

In November 2022, OpenAI released ChatGPT. Since then, ChatGPT has found its way into all facets of society. By many accounts, Google was caught off guard by how performant and popular ChatGPT has been. 

Google announced the waitlist for Google Bard in March 2023. Bard has been in the conversation for this next generation of chatbots, though at least in my conversations, ChatGPT seems to be much more popular. There are no publicly available user metrics for Bard that I could find, while ChatGPT has reached over 100 million users.

Google has been feeling the pressure since November. The Information mentions that Bard internally and externally has had some problems. Sundar Pichai, the CEO of Alphabet & Google, asked that teams move with the utmost urgency in generating new ways to use large language models. Onlookers saw the rushed Bard release on display publicly with the hiccups during the initial release. However, Google is staying persistent and a reorganization effort into a more cohesive team might be just what they need.

Announcing Google DeepMind: The new brainchild

On April 20th, Google officially announced on their blog that they are combining DeepMind and Google Brain. This means that there is no longer an internal battle of DeepMind vs. GoogleBrain but rather Google Deepmind vs. OpenAI and other competitors. 

In the announcement, Pichai noted the achievements of Search, Youtube, and Gmail that have utilized AI to reach new heights. He also states that “the pace of progress is now faster than ever before.” In order to develop AI systems rapidly and safely, a new team is needed.

The tech CEO hopes that by combining all of this talent into a singular focused team, Google will be able to accelerate the progress of their AI development. Importantly, Pichai noted that Google Research will continue along its current path, though without Google Brain.

Demis Hassabis, former CEO of DeepMind, will lead Google DeepMind in this new merger. Jeff Dean, the former lead of Google Brain, will be the chief scientist of Google DeepMind. This a nuanced and fascinating leadership change. The expertise that Hassabis brings on creating demo-ready products through DeepMind previously will be critical for the release of future AI products. 

A statement released by the new Google DeepMind CEO captured many of these same points as Pichai’s announcement did. However, one additional point that Hassabis mentioned in his release was that with Google DeepMind there will be a new “Scientific Board.” This board will “oversee research progress and direction of the unit.” 

To get an idea of how smooth this merger will be for these two entities, we need to dive into their completely separate roots and history.

The story behind Google Brain

Google Brain was a subdivision of Google Research. Started in 2011, Google Brain was a product of Google X that spun up with AI greats Jeff Dean and Andrew Ng among others. After spinning out from X, Google Brain became a part of the much broader Google Research operation. Some of the most notable accomplishments of Google Brain were to develop the widely-used Tensorflow package and the Google Cloud AutoML platform.

Google Brain states on their website that they “believe that openly disseminating research is critical to a healthy exchange of ideas.” This point may be put into jeopardy as part of the new Google DeepMind. There is simply too much money on the line to win the chatbot race and I imagine Google will hold off on sharing their latest cutting edge research. In a few years maybe they’ll share some of the details of what they did.The Brain team is made up of academic researchers. This is important to note since, spending time completing a PhD myself, the best academic researchers can sometimes be oblivious to the difficulty of implementation for real software products. Another snippet that Google Brain mentions about their team is that “members are encouraged to set their own research goals.” I’m certain this will not be achievable when there is a hard deadline for the next chatbot release. Research often has a “it’ll take the time that it takes” mentality, which again, is not conducive for real-world releases.

DeepMind: a startup that dominated

Deepmind is a different beast from Google Brain entirely. They started as an independent company by Demis Hassabis and Shane Legg. As entrepreneurs, Hassabis and Legg must have developed the skillset to think in terms of product and profitability.

This is in sharp contrast with Google Brain where there likely was not much pressure to always be profitable. Instead, they aimed to generate research that could be made profitable by adding it to software products sometime down the line.

DeepMind releases demos of their innovations, rather than the source code like Google Brain’s TensorFlow. Perhaps the most incredible innovation from Deepmind was AlphaGo. The Netflix documentary, AlphaGo, details DeepMind’s journey of developing and showing off AlphaGo to the world. Equally impressive is Alphastar, which learned to play Starcraft 2 at an elite level.

Research plays a major role at DeepMind, though they are most focused on implementation.

How interconnected have the teams at DeepMind and Google Brain been previously?

Google Brain and DeepMind, even though they’ve been operated by the same parent company Alphabet, were completely separate not only in name but also in terms of collaboration. The Information’s report states that the two business units “have seldom collaborated or shared computer code with one another.” Additionally, they reported that there were “years of intense rivalry” between the two groups. These two divisions are now being forced to work together to take on one of their biggest challenges yet in building the next Google chatbot. 

Google Gemini

In Sundar Pichai’s announcement of the merger, he described a “series of powerful, multimodal AI models” to be released in the near future that will keep Google competitive. It has been leaked that internally the new model will be known as “Gemini” which will be separate from Google Bard. 

So far, we can only guess what exactly will be going into these new models. It’s common for technology companies to iterate on versions of their models until they’re performant. OpenAI has iterated on versions of GPT until they had a language model that could be released as ChatGPT. Similarly, Google has numerous products that they can pull from. Among them is DeepMind’s Sparrow that was released in September 2022. In the release, they put a particularly strong focus on how easily the chatbot could be tricked by users. They published figures that stated 78% of the time, Sparrow would return a factual answer supported with evidence whereas on the other hand, 8% of the time, the chatbot could be tricked. It is reassuring that they are paying attention to this reliability metric within their model.

As the new organizational structure gets put into place, we will see how the focused effort on Gemini turns out. Will it be released with better reception than the bungled Google Bard? With two giant AI businesses coming together, it will be a make-or-break moment for Google in its race to stay in the AI conversation.