DeepMind's new journey as part of the mothership begins

Google DeepMind , one of the world's leading artificial intelligence labs, is preparing to embark on a new phase of its life as the AI race heats up. 

Founded in 2010 and headquartered in London, the Alphabet-owned entity announced last month that it's merging with longtime rival Google Brain as part of a plan to create a single AI superlab: Google DeepMind. (This happens to be what DeepMind was called when it was first acquired by Google). 

"Combining our talents and efforts will accelerate our progress towards a world in which AI helps solve the biggest challenges facing humanity, and I’m incredibly excited to be leading this unit," said DeepMind cofounder Demis Hassabis in a blog post. Alphabet Inc. and Google CEO Sundar Pichai said in another blog post that the restructure will significantly accelerate Alphabet's overall progress in AI.

Big tech firms across the board are making AI a number one priority, according to Nina Schick , the founder of AI advisory firm Tamang Ventures and the author of Deepfakes. 

"This (AI) revolution will have a bigger impact than internet, smartphone and cloud revolutions of the past," Schick said, adding that adoption of AI will be incredibly fast. 

AI systems of the future could be far more general than today's relatively narrow agents, which are typically only good at one or a small handful of tasks. AGI, or artificial general intelligence, has long been seen as the holy grail in the AI community. While there's no fixed definition of what AGI actually is, most researchers agree that it's when an AI system displays some kind of consciousness, or sentience. 

Hassabis said this month that rapid progress in the field means AGI could now be possible within a decade. Previous estimates for AGI were typically several decades, with some predicting it could take 50 to 100 years to develop.

Why join forces? 

"There is compelling business logic for Alphabet to do this [merge]," wrote Fortune journalist Jeremy Khan on LinkedIn. 

The Google DeepMind merge should prevent the two labs from duplicating research and competing with one another, venture capital investor Nathan Benaich told LinkedIn News UK. 

"By merging Brain and DeepMind, they’ve pulled DeepMind out of the 'bet' zone, and made it a unit of Google, ensuring that any DeepMind effort is, in effect, an Alphabet effort. Practically, it signals just how seriously they’re taking improving AI capabilities."

The move is being viewed by some as a direct response to the rise of OpenAI , which is backed by Microsoft ( LinkedIn 's parent). Microsoft is one of Google's biggest rivals, with the two competing on everything from search engines to cloud computing. 

"If you’re focusing on building AGI, OpenAI has proven that one way to get close is to prioritise engineering over pure research and gaining access to huge amounts of compute in order to train huge models on huge datasets. Google Brain and DeepMind merging tells me that large labs that have been trying to build AGI are starting to get antsy," Benaich said. "Alphabet has added another 'bigger is better' to the formula: if you have huge amounts of researchers working on training huge models on huge datasets using huge amounts of compute, you might have a better chance of getting to AGI first."

However, as history often teaches us, small teams, like OpenAI, focused on one large bet can move mountains, Benaich said. 

AI advisor Henry Ajder wrote on LinkedIn: "Ever since Sundar Pichai issued an internal ‘code red’ in response to OpenAI’s ChatGPT/Microsoft’s Bing Chat, it's been clear they were going to shake things up to stay competitive and relevant in the AI space." 

He added: "Unifying the two [DeepMind and Brain] into one AI powerhouse aligned with Google’s broader vision for both experimental and applied AI makes sense as the 'AI arms race' heats up."

Tough times ahead? 

DeepMind and Brain have operated quite differently ever since Google acquired DeepMind in 2014 for around $600m (£483m), beating out a rival bid from Facebook. 

It might have made sense to merge DeepMind and Brain immediately after the deal, given they were both working on AI. But as part of the acquisition, Hassabis and his co-founders – Oxford dropout Mustafa Suleyman and New Zealander Shane Legg – negotiated a significant amount of autonomy for DeepMind, meaning it could continue to operate as a separate company in London. 

At the time of the acquisition, DeepMind and Brain were also working on quite different subfields of AI. Brain was working on areas like computer vision (using AI to classify and process images) and natural language processing, while DeepMind was focused on reinforcement learning, which involves training a system to find and seek a reward, often in a gaming environment. 

In the years that followed the DeepMind acquisition, Brain was tasked with building AI for products like Translate and YouTube, while DeepMind has largely operated independently and focused on its core mission of "solving intelligence" and creating AGI. 

DeepMind's work has involved training AI systems to play complex games such as Go better than the top human players and solving grand challenges in science, such as predicting how a protein will fold. 

DeepMind and Brain were competing for access to as much computing power as possible in Google's data centres, according to Fortune. But the differentiation between the two labs has blurred in recent years as DeepMind's remit expanded beyond reinforcement learning and into areas that Brain also works on. As a result, it didn't make sense for Alphabet to keep funding two separate efforts, Fortune reports.  

Not everyone is convinced that DeepMind and Brain will be the perfect union, partly due to their different approaches to AI. 

One former DeepMind employee who asked to remain anonymous due to clauses in their contract told LinkedIn News UK that it seems like a "classic big company response" to an innovation challenge and it makes them question how much independence DeepMind will have.  

They added that "DeepMind's long-term AGI goal seems to be impossible" if the firm is a division of Google "chasing product doodads". 

Broader impact

"DeepMind have been terrified that this [merge] would happen, and resisted it at every turn for many years since the acquisition," one AI veteran with close links to DeepMind who asked to remain anonymous due to fear of repercussions told LinkedIn News UK. "It's a very depressing day both for AI in the UK and all the DeepMind folks." 

They added that they believe DeepMind is "putting a brave face on it" and that many of the company's AI researchers may now have to focus on building large language models that can compete with OpenAI's GPT models.

"Any merger of this size is always going to have teething problems," AI advisor Ajder said. 

Eze Vidra , a venture capitalist who used to work at Google, told LinkedIn News UK that he believes AI is the next big platform shift and the stakes are high for the world's largest tech firms. 

"Google has a lot of catching up to do," he said. "The merger of Brain and DeepMind will enable the company to use their vast AI talent pool and capabilities in a more concentrated, strategic way. I believe that Google is well positioned to be a leading force in the AI space, but its efforts until now have been more fragmented."

DeepMind declined to comment on the claims and pointed LinkedIn News UK to blog posts from Hassabis and Pichai. Google did not immediately respond to a request for comment.