AI summary by TwinMind
Are Insect Brains the Secret to Great AI? | Frances S. Chance | TED
TED
9m 32s
Introduction: The Challenge of Artificial Intelligence
- Creating intelligence on computers = holy grail of AI for quite some time
- Logical approach: study human brains (substrate of cognition) to create AI
- Human brain complexity: 86 billion neurons, each communicating with thousands of others
- Current technology/knowledge may be insufficient for human brain replication
- Solution: start smaller with insect brains
Insect Brain Specialists as Model Systems
- Three fascinating insect brains chosen as examples:
- African dung beetles: rolling large balls in straight lines
- Analogy: like making snowman when ball is as big as you, standing on head
- Sahara desert ants: navigation specialists
- Wander considerable distances to forage
- Know how to return home after finding sustenance
- Dragonflies: hunting specialists
- Capture ~95% of prey they choose to pursue
- African dung beetles: rolling large balls in straight lines
- Neuroscientists study these as model systems to understand animal nervous system problem-solving
- Research goal: bring biology's best solutions to computers
Dragonfly Brain Focus
- Dragonfly brain: ~1 million neurons (vs 86 billion human neurons)
- Still complex but more manageable than human brain
- Study focus: what insects do well, not human intelligence
- Dragonfly specialty: interception
Dragonfly Interception Mechanics
- Hunting behavior: fly to intercept prey (not straight at it)
- Aim for where prey will be (like soccer player intercepting pass)
- Requires coordinate transformation:
- From eye's frame of reference (what dragonfly sees)
- To body's frame of reference (how to turn body to intercept)
- Coordinate transformations = basic calculation animals perform to interact with world
- Humans do instinctively when reaching for objects
- Same visual image, different arm trajectories based on head position
Speed and Computational Efficiency
- Dragonfly response latency: ~50 milliseconds after seeing prey turn
- Remarkable speed metrics:
- Half the time of human eye blink
- Suggests few computational steps for interception
- Brain computational step = single neuron or layer of neurons working in parallel
- Single neuron processing time: ~10 milliseconds to add inputs and respond
- 50ms response time = maximum 4 computational steps/layers possible
- Neural circuit for coordinate transformations: at most 4 layers of neurons
- Circuit small enough to identify and study with current tools
Research Model and Testing
- Built model of neural circuit calculating dragonfly turning
- Key finding: coordinate transformations in only 1 computational step (1 layer of neurons)
- Computer simulation capabilities:
- Predict individual neuron activities during hunting
- Predict action potentials/spikes when dragonfly sees prey move
- Testing approach with collaborators:
- Compare predicted neural responses with recorded living dragonfly brain responses
- Ongoing experiments: living dragonflies in virtual reality
- Method: show movies of moving targets while electrode records neuron activity
- Goal: identify neurons responsible for coordinate transformations
- Next step: understand how neurons work together for calculations
Building Blocks for Complex Functions
- Basic/primitive calculations = building blocks for:
- More complex functions beyond interception
- Cognition itself
- Neuron computation methods may differ from existing computer approaches
- Goal: build computer chip that replicates biological brain function AND method
Practical Applications
- Drone technology:
- Computer same size as dragonfly brain
- Capture some targets, avoid others
- Personal hope: small army defending backyard from mosquitos
- Navigation devices:
- Replace GPS with dung beetle/ant-based navigation
- Guide to straight/easy path home
- Power efficiency:
- Human brain power requirement: equivalent to 20-watt light bulb
- Brain-inspired computers could have extremely low power requirements
- Smartphone/smartwatch: daily charging needed
- New brain-inspired device: charging every few months or years
Conclusion and Future Vision
- Richard Feynman quote: "What I cannot create I do not understand"
- Insect nervous systems = opportunity to understand brains through computer creation
- Computer creation goals:
- Not just knowledge acquisition
- Real impact on devices, vehicles, artificial intelligences
- Potential: tiny insect brains can lead to remarkable computers
- Future possibilities through insect brain study
