Humans Beat Computers at Mapping Neurons
Learn how human gamers are helping scientists and computers draw more accurate maps of the human retina. How do human stack up against artificial intelligence?
Published February 2, 2016
By Diana Friedman
Academy Contributor
The average human brain is estimated to contain between 86-100 billion neurons and 10,000 times as many synapses. In the retina, the layer of tissue at the back of the eye that receives and sends visual signals to the brain, scientists don’t yet know the number and type of neurons that are connected. Even with advances in software technology for neuron reconstruction, it would take a team of 100 people working non-stop for 500,000 years to map the neurons in one human brain. However, researchers devised a novel solution to this obstacle-using a browser-based computer game, anyone can help in neural mapping and contribute to neuroscience research.
Created by Sebastian Seung’s research lab, the computer game EyeWire uses crowdsourcing to study the trillions of connections between neurons (which Seung calls “the connectome”) that are hypothesized to store memories, personality, and perhaps even intellect. The game is similar to a puzzle, in which players are assigned a high resolution picture of a partially reconstructed neuron branch from a retina; on the right side of the screen, players “color” the parts of the image that are neurons.
The goal of the game is to select the areas that artificial intelligence (AI) has missed, improving the outline of the neuron and helping neuroscientists more accurately map neurons in the retina. The researchers hope that by gaining a greater understanding of the neurons in the retina, they can then apply these techniques to assessing neurons and synapses in the human brain.
Humans vs AI
Why recruit humans to complete this task, rather than rely on AI alone? Humans are actually more proficient at spotting neuronal connections than most AI, so the data from the game is being used to help the machines learn how to better perform the task. Player data is also submitted to Seung’s lab, where a computer reassembles the colored sections and produces a 3-D image of the connected neurons. Anywhere from 5 to 25 game participants trace the same set of connected neurons before it’s determined to be valid.
The game offers performance-based badges, chat, and weekly challenges, but some players received an unexpected bonus for playing the game – co-authorship credit in a published study. The research, published in Nature, sought to evaluate how mammalian retina detect motion by reconstructing Off-type starburst amacrine cells and bipolar cells (BC).
To map these neurons, the research team designed a special “Starburst Challenge” and recruited over 2,000 of the top-performing users to complete the activity. In the analysis of the mapping, the researchers found evidence that one bipolar cell type prefers to wire with an Off-type SAC dendrite near the SAC soma, while another BC type prefers to wire far from the soma and with faster visual response time. All players were included as co-authors on the paper.
Soon EyeWire will be expanding beyond the computer browser, as the company plans to develop a mobile version of the game. EyeWire will also be debuting EyeWire VR, a virtual-reality version of the game with hand motion-tracking sensors at this year’s Tribeca Film Festival in New York City. Regardless of the platform, the essence of EyeWire is simple: anyone can help neuroscience research in a fun and interactive way.
Also read: High Temps Call for High-Tech Edutainment
