Search, Training and What Self-Play Does Not Prove
RAISE GENIUS · PARENTS & SCHOOLS
Search, Training and What Self-Play Does Not Prove
AlphaGo learning discussion in this archive focuses on repeated practice. Its historical explanation needs a distinction between a learning algorithm and a child’s experience.
Contact the teamOn this page
AlphaGo learning discussion in its historical setting
The source introduces artificial intelligence and the attention surrounding AlphaGo’s Go matches. “Latest” belongs to the original publication period, not a current ranking of AI systems.
Go has a vast space of possible positions, which makes exhaustive search impractical. That does not mean no computer Go programme existed before AlphaGo or that all earlier game AI used the same method.
Search and learning can be combined
Google DeepMind describes AlphaGo as combining neural networks with search. Earlier AlphaGo used learning from human games as well as reinforcement learning, while AlphaGo Zero later began without that human-game training.
The archive’s “three million” simulated games is not a universal specification for every version. Self-play and learned evaluation should not be simplified into a system remembering every possible board or merely copying a human learner.
A useful analogy has limits
The author connects trying, observing and revising with a child’s construction project. That can be an approachable analogy, but a machine-learning training process and a child’s development are not identical.
AlphaGo learning discussion can invite questions about a goal, feedback and the test setting. The article’s encouragement that a student could surpass AlphaGo is not a guaranteed result from attending a course.
Original photographs and artwork
These images are retained from the original page as examples of published learning activities and programme information. Some contain Thai text or historical dates. They do not guarantee the same result for a new learner. Confirm current programmes with the team.
View original images
Frequently asked questions
Was AlphaGo the first computer programme ever to play Go?
No. Its milestone should not erase earlier computer Go work.
Did every AlphaGo version train the same way?
No. Distinguish earlier AlphaGo from AlphaGo Zero.
Does self-play prove a machine learns exactly like a child?
No. The classroom comparison is an analogy.
Contact and related pages
Tell the team about the learner or organisation, relevant experience and the intended learning purpose. Confirm the current programme, equipment, availability, schedule and fees before booking.
Contact Raise Genius School service overview
- Instructions, Shared Services and Familiar Devices
- Guidance, Support, Power and Different Transport Concepts
- From a Checkout Belt to an Airport Baggage System
- A Wire-Walking Video and the Hardware–Software Relationship
- An Apparent Endless Mechanism Needs an Energy Explanation
- Recognising Containers, Recording a Return and Sorting
- Changing Direction, Sharing a Load and Pulling More Rope
- Hadrian, Wall Construction and the Rest of a Building
- Pretend Roles, Stories, Shared Building and Personal Designs
Course-selection guide · Robotics portfolio projects · Competition preparation · Team advice and coaching · Course guide · General FAQ
Related services: School robotics camps · STEM activity stations
Related learning: Raise Maker Arduino · micro:bit robotics · IOTBot & AI · Portfolio preparation · Camp proposal · Accompanying teachers
Related reference: Google DeepMind: AlphaGo · Google DeepMind: AlphaGo Zero
