Google and AI have become so intertwined that it's hard to imagine one without the other. But this deep integration wasn't always the case. It was a gradual evolution, starting with a seemingly simple application that would revolutionize how people around the world access information.
Google's initial foray into AI, though not explicitly labeled as such in the early days, was fundamental to its core product: Google Search. Early on, Google utilized techniques like PageRank, which analyzed the link structure of the web to determine the importance and relevance of pages. This was enhanced with machine learning to help the algorithms learn, identify patterns and use them to improve search results.
Over time, Google's search algorithms became increasingly sophisticated, incorporating more advanced machine learning techniques to better understand natural language, identify user intent, and personalize results. This continuous refinement was the bedrock upon which Google's AI ambitions would be built.
The true acceleration of Google's AI adoption came with strategic acquisitions and internal breakthroughs. A pivotal moment was the acquisition of DeepMind in 2014. This London-based AI company had already made waves with its ability to teach computers to play Atari games at a superhuman level using deep learning.
Demis Hassabis, CEO of DeepMind, has seen a lot of things during his career, and if AGI is what everyone is after, then data is the key towards opening the gates.
"So SIMA is deciding what actions to take and what. You can give it a goal, like go and find the key in the room. And then it will send out commands as if it's playing a normal computer game. But actually, at the other end is Genie 3 generating the world on the fly. So there's one AI generating the world, and there's another AI inside of it."
"So it could be really useful for, of course, creating unlimited training data."
In an episode of the Release Notes podcast by Google Developer, Hassabis discussed a variety of topics related to recent AI releases and the future of the technology.
The DeepMind executive emphasizes that the company in injecting Google with a new level of expertise and capability towards building more and more powerful large language models (LLMs), in a way that benefits Google in its rivalry against others, notably following the launch of OpenAI's ChatGPT.
Since launching Gemini, for example, Google started enhancing the experience and the capabilities of pretty much of all of its products and services.
But from where Gemini came from, was kind of subtle: it was creating AIs to excel in one particular task, one at a time.
AlphaGo is one of the prime examples of this.
Hassabis explains that DeepMind was always focused on "agent-based systems", which can be described as "systems that can complete a whole task" rather than just doing one small thing. For instance, a whole task might be playing a game very well. "This," Hassabis concludes, "is obviously the way to get to Artificial General Intelligence (AGI).
Before LLMs boomed, Google was focusing on creating LLMs, but without putting focus on perfecting these agent-based systems. And when Gemini was launched and the war raged, Google was prioritizing in building more and more LLMs.
But then, with Google Gemini's Deep Think, the company started seeing a way to make LLMs think deep, reason more, and conduct parallel thinking.
This, according to Hassabis, is shifting things back towards agent-based systems with gaming as a benchmark for the AI's 'intelligence.' The advent of the thinking models is a little bit of a hark back to DeepMind's original gaming work on things like the aforementioned AlphaGo, as well as AlphaZero.
In this case, by putting SIMA, or Simulated Agent that can take controls and play an existing computer game, into Genie 3, the "groundbreaking world model that creates interactive, playable environments from a single text prompt" can create a world for SIMA to interact with, which results in unlimited training data.
This in turn can be useful for a lot of things, from AGI, robots to future interactive entertainment.
In other words, Hassabis is saying that the AI industry is finally realizing the benefits of gaming for both AI use and AI training, just as how DeepMind used to recognize it before AI was the headlines of tech.
It's like running in around in a circle. But in this particular lap, Google is more ready than ever.
Further reading: AI Terms Like 'AGI' And 'Super Intelligence' Are 'Designed To Activate People's Dopamine'