Skip to main content

Here’s your stop: Google DeepMind’s new AI can help you navigate the subway system

Humans take reasoning for granted, but logic isn’t always self-evident to machines, which have to be hard-coded to make the connections that can support basic deductions.

Google’s DeepMind is looking to change that. The London-based artificial intelligence company has developed a system that performs relatively simple tasks in a sophisticated — and increasingly human — way, reports the Guardian.

Recommended Videos

While plenty of programs can guide you through the subway, DeepMind’s differential neural network is one of the first systems to use external memory and deep learning to train itself autonomously, without the need for hard-coded instructions.

Differentiable neural computer family tree inference task

Deep learning has become the go-to method for machine learning over the past few years, achieving unprecedented success in tasks like image and speech recognition. A DeepMind-developed program called AlphaGo used deep learning to defeat one of the world’s best Go players earlier this year. Although these systems do very well at their specific task, though, they stumble with general skills.

“Until very recently, it was far from obvious how deep learning could be used to allow a system to acquire the algorithms needed for conscious deliberate reasoning,” Professor Geoff Hinton, considered the father of deep learning, told the Guardian.

To overcome this, DeepMind integrated its system with an external memory that enabled it to retain relevant information and use this data as a human would use his or her own working memory.

The AI was able to determine the quickest route between London underground stops and navigate its way around the notoriously complicated subway system, according to a study published in the journal Nature. It also performed relatively well on basic reading comprehension tests.

“I’m wary of saying now we have a machine that can reason,” Google DeepMind researchers Alex Graves told the Guardian. “We have something that has an improved memory — a different kind of memory that we believe is a necessary component of reasoning. It’s hard to draw a line in the sand.”

Regardless of semantics, programs that demonstrate basic reasoning may one day replace more limited systems like Siri, and may be seen as a development towards a form of AI that better resembles the human mind.

Dyllan Furness
Former Contributor
Dyllan Furness is a freelance writer from Florida. He covers strange science and emerging tech for Digital Trends, focusing…
I tried out Google’s latest AI tool that generates images in a fun, new way
Google's Whisk AI tool being used with images.

Google’s latest AI tool helps you automate image generation even further. The tool is called Whisk, and it's based on Google’s latest Imagen 3 image generation model. Rather than relying solely on text prompts, Whisk helps you create your desired images using other images as the base prompt.

Whisk is currently in an experimental phase, but once set up it's fairly easy to navigate. Google detailed in a blog post introducing Whisk that it is intended for “rapid visual exploration, not pixel-perfect edits.”

Read more
Google strikes back with an answer to OpenAI’s Sora launch
Veo 2 on VideoFX

Google's DeepMind division unveiled its second generation Veo video generation model on Monday, which can create clips up to two minutes in length and at resolutions reaching 4K quality -- that's six times the length and four times the resolution of the 20-second/1080p resolution clips Sora can generate.

Of course, those are Veo 2's theoretical upper limits. The model is currently only available on VideoFX, Google's experimental video generation platform, and its clips are capped at eight seconds and 720p resolution. VideoFX is also waitlisted, so not just anyone can log on to try Veo 2, though the company announced that it will be expanding access in the coming weeks. A Google spokesperson also noted that Veo 2 will be made available on the Vertex AI platform once the company can sufficiently scale the model's capabilities.

Read more
Google’s new Gemini 2.0 AI model is about to be everywhere
Gemini 2.0 logo

Less than a year after debuting Gemini 1.5, Google's DeepMind division was back Wednesday to reveal the AI's next-generation model, Gemini 2.0. The new model offers native image and audio output, and "will enable us to build new AI agents that bring us closer to our vision of a universal assistant," the company wrote in its announcement blog post.

As of Wednesday, Gemini 2.0 is available at all subscription tiers, including free. As Google's new flagship AI model, you can expect to see it begin powering AI features across the company's ecosystem in the coming months. As with OpenAI's o1 model, the initial release of Gemini 2.0 is not the company's full-fledged version, but rather a smaller, less capable "experimental preview" iteration that will be upgraded in Google Gemini in the coming months.

Read more