Will Computers Ever Think Like Human Beings? - with Vint Cerf

The Royal Institution・2 minutes read

The speaker discusses neural circuits, AI systems, and the importance of viewing AI as tools to enhance human capabilities rather than replacements, emphasizing the potential brittleness of neural networks and the impact of social media on polarization. Additionally, the text highlights the scientific method, challenges in scientific research, critical thinking in the digital age, and the need for international cooperation to address online harm and misinformation.

Insights

  • The speaker emphasizes the importance of viewing AI systems as tools to enhance human capabilities rather than replacements, cautioning against overestimating their cognitive abilities and highlighting their narrow learning focus.
  • Neural networks, while powerful, can exhibit brittleness due to their classification mechanism in high-dimensional spaces, leading to unexpected behavior if solely relied upon for resilience, showcasing the need for careful consideration and understanding of their limitations.

Get key ideas from YouTube videos. It’s free

Recent questions

  • What books are recommended for reading?

    Scientific American, "Thinking, Fast and Slow," "Behave"

  • How are neural circuits illustrated?

    Through a three-neuron circuit example

  • What is the evolution of neural networks?

    From single-layer perceptrons to multi-layer networks

  • How are AI systems cautioned against?

    Overestimating cognitive capabilities and narrow learning focus

  • What is the focus of Douglas Engelbart's invention?

    Collaboration and problem-solving among people

Related videos

Summary

00:00

"Neural Circuits, AI, and Human Augmentation"

  • The speaker expresses gratitude for the audience's presence on a dark and stormy night, particularly acknowledging his wife Sigrid and Dame Wendy Hall.
  • The speaker wears a transmitter to communicate with his wife, who has cochlear implants, allowing her to hear through a radio-based transmission.
  • The speaker mentions Marconi's historical talk in the same location and hints at exploring risky topics rather than demonstrating inventions.
  • Reading assignments are recommended, including Scientific American's issue on truth, Daniel Kahneman's book "Thinking, Fast and Slow," and Robert Sapolsky's book "Behave."
  • Sapolsky's book delves into brain function, emphasizing the complexity of neural circuits and the importance of connectivity over sheer transistor count.
  • The speaker compares the structure of the internet, decentralized yet unified by common protocols, to the intricate connectivity of the brain's neural circuits.
  • A simple three-neuron circuit example is used to illustrate how neural circuits drive behaviors through interactions, emphasizing the complexity of the brain's functions.
  • The evolution of neural networks from single-layer perceptrons to multi-layer networks capable of complex decision-making is discussed, with examples like AlphaGo and AlphaZero.
  • The speaker cautions against overestimating the cognitive capabilities of AI systems, highlighting their narrow learning focus and potential as tools to enhance human capabilities.
  • The importance of viewing AI systems as tools to augment human abilities, rather than replacements, is emphasized through the example of Douglas Engelbart's invention of the mouse for collaborative text editing.

16:51

"Technology, Society, and Polarization in 1968"

  • In 1968, Douglas Engelbart demonstrated the cadential Bart system in San Francisco, inventing the mouse to interact with the computer screen.
  • Engelbart's focus was on collaboration and problem-solving among people using computers to improve society.
  • Neural networks, while powerful, can be brittle, leading to unexpected behavior if relied upon for resilience.
  • Neural networks are trained through back propagation with random weights to recognize images like cats and dogs.
  • Generative adversarial networks can trick trained systems into misidentifying images by altering pixels.
  • The brittleness of neural networks arises from their classification mechanism in high-dimensional spaces.
  • Social media platforms like YouTube, Facebook, and Twitter reward extreme content for attention, potentially fueling polarization.
  • Trolling behavior on social media seeks reactions and attention through incitement.
  • Social polarization online mirrors tribal behaviors, where individuals align with a group and reject opposing views.
  • People tend to minimize uncertainty, potentially leading to the rejection of information that introduces uncertainty, conflicting with the scientific method.

33:56

Scientific Method, Hypotheses, and Nobel Prizes

  • The scientific method involves forming hypotheses and testing them through experiments to determine their validity.
  • A good hypothesis is one that can be falsified through experimentation.
  • Confirmation bias can lead scientists to only accept data that supports their theories, ignoring contradictory evidence.
  • Scientists must be willing to accept the possibility that their theories could be wrong based on experimental results.
  • Nobel Prize winners often did not anticipate receiving the award while conducting their research.
  • Nobel Prizes are often awarded to scientists who are willing to challenge cherished theories in the face of new data.
  • The Laser Interferometer Gravitational-Wave Observatory (LIGO) experiment aims to detect gravity waves using laser interferometry.
  • LIGO successfully detected gravity waves from two black holes colliding and later from two neutron stars colliding.
  • Patience and persistence are crucial in scientific research to reach a point where theories can be proven or disproven.
  • Critical thinking is essential in evaluating information from various sources, especially in the age of the Internet and misinformation.

50:23

Challenges of Online Misinformation and Technology Security

  • Bad actors online can spread misinformation, like the hoax about email charges, which can be debunked on websites like Snopes.
  • The global nature of the internet poses challenges as harm initiated in one country can affect another, highlighting the need for international cooperation and agreements.
  • Norms like not attacking internet infrastructure are proposed to protect the network, similar to how hospitals are protected in war.
  • Self-driving cars, like those in Arizona, showcase the need for reliability, safety, and security in technology, with GPS aiding in navigating obstacles like speed bumps.
  • Companies must have ethical and legal obligations to support software throughout its lifetime, ensuring security updates and access to source code for continued maintenance, especially in devices like IoT managing critical systems.
Channel avatarChannel avatarChannel avatarChannel avatarChannel avatar

Try it yourself — It’s free.