DEV Community

Cover image for While You're Busy Chuckling at Google Gemini, Guess What? Google's Cooking Up Something Even Bigger!
Dmitry K
Dmitry K

Posted on • Edited on

While You're Busy Chuckling at Google Gemini, Guess What? Google's Cooking Up Something Even Bigger!

Unlocking superhuman knowledge that will transform us into Superhumans

Google's visionary, Demis Hassabis (CEO of DeepMind Technologies), has long dreamed of a goldmine in AI application, and it's now inching closer to reality. Google DeepMind's new study, "Bridging the Human–AI Knowledge Gap" [1], is a monumental stride towards fulfilling Hassabis's sacramental aspirations. This dream envisions elevating humanity to superhuman status by equipping us with:

  • Access to the superhuman knowledge of a supreme machine intelligence
  • The ability to sift through this vast ocean of knowledge and identify what we can comprehend and digest
  • Educating humans to transfer this super-intelligence's wisdom to us

Artificial Superintelligence: Not Just a Fantasy Anymore

We're not just spinning tales here.

First off, artificial superintelligence isn't just a concept—it's a reality, and there's more than one example out there. We're already convinced of the superhuman abilities of AI systems like AlphaZero in chess and Go, and AlphaFold's prowess in predicting protein structures. However, the extent of superhuman knowledge in AI systems like ChatGPT remains a subject of intrigue and suspicion.

To put it in perspective, AI's superhuman capabilities manifest in three ways:

  • The sheer computational power of machines;
  • A novel approach to reasoning with existing knowledge;
  • Knowledge that humans have yet to acquire.

The latter two are what the authors term as superhuman knowledge. This exploration into AI's uncharted territories of knowledge is not just groundbreaking – it's reshaping our understanding of intelligence itself.

The Rapid Rise of Diverse Artificial Superintelligences

The landscape of artificial superintelligence is doomed for a rapid expansion, fueled by the ongoing process of fine-tuning large language models with specialized training datasets. This evolution makes Hassabis's tripartite dream increasingly relevant.

Interestingly, from a business standpoint, this development, rather than the creation of AI-based chatbot assistants, could be the real goldmine in AI application.

  • While pragmatists like Sam Altman focus on the immediate returns from popular AI dialog assistants (currently attracting 62% of AI development funding), visionaries like Demis Hassabis see beyond;
  • Hassabis envisions a future where humans are educated by specialized machine superintelligences, imparting a vast array of superhuman knowledge, thus paving the way for a superhuman era.

The Achievements So Far

Progress has been remarkable in utilizing AlphaZero AI. A framework based on this AI has been developed to:

  • Identify Concepts Known to Both AI and Humans:
    • Shared by AI and humans (M ∩ H);
    • Known only to humans (H − M);
    • Known only to machines (M − H), representing superhuman knowledge;
  • Spot Special Concepts (M − H) that a human can learn:
    • These are initially challenging for humans but can be understood and learned (like AlphaGo's famous 37th move against Lee Sedol);
  • Train Humans on Advanced Concepts (M − H):
    • By observing the superintelligence's actions, this process metaphorically transforms humans into superhumans, enhancing their understanding of complex concepts.

Advanced Concepts (M − H)

Framework's Impact on Chess Grandmasters

This innovative framework was experimentally tested on the world's leading chess grandmasters (ratings around 2700 and 2800). The results were telling. Post-training, these grandmasters showed a marked improvement in identifying conceptual moves from the (M − H) area, compared to their performance prior to observing AlphaZero's moves. This signifies a clear enhancement in their ability to grasp and apply advanced, AI-derived chess strategies.

The Beginning of a Superhuman Era

  1. This is just the beginning;
  2. Adapting the framework from chess to language models is challenging but feasible;
  3. If Hassabis's vision materializes, the acquisition of superhuman knowledge could revolutionize science, technology, and business.

Google's latest breakthrough isn't just a tech marvel. It's your invitation to the future! Imagine enhancing your brainpower with AI's superhuman knowledge. But here's the twist: it's not just about smarter machines; it's about YOU becoming smarter.

Think it's too good to be true? Share your thoughts, challenge the possibilities, and let's debate! Are we on the brink of a superhuman era, or is this just science fiction? Comment below and stay tuned for more updates!

[1]: Original Post – While You're Busy Chuckling at Google Gemini, Guess What? Google's Cooking Up Something Even Bigger!

Top comments (0)