Q & A
Sometimes nature knows best — and that's particularly true for search engines. Nothing can beat the brain for its search engine, not even Google. And scientists are now designing search engines of the future with the brain in mind — the fly brain in fact.
The types of search engines the scientists are interested in are those that do similarity searches. Those are the ones that generate suggestions based on your past behaviour or online interests. They are often spectacularly wrong – like you just bought a "John Mayer relaxing at home" kinda song — how about this "thumpa thumpa" dance hall techno remix of Lil Kim?
There is room for improvement in computer generated similarity searches — from recommended songs to apps to recipes. The current way to find recommendations also takes a lot of computing power.
But flies do this type of search all the time, especially in terms of smells. A fly will encounter a lot of smells and some mean "food" and others mean "flee." For example, a rotting banana means food. They then have to decide if a new smell, like a rotting kiwi, is similar enough to represent food. So they have to have a program in the brain to do a similarity search. The researchers found out what that was and were able to mimic it in computer software.
The flies have a way of sensing smells in their brain that is a hierarchy of neurons. There are 50 smell-sensing brain cells called olfactory neurons. They first detect the chemical stimulus of a smell, then they distribute those signals across 50 projection neurons in a particular pattern and finally to 200 Kenyon cells. So the signal is expanded and spread out across the neurons until there is a unique pattern for a smell.
The fact that they basically spread the information over a large number of cells allows a refinement of the information. Here's how Saket Navlakha from the Salk Institute for Integrative Biology explains it: "Let's say you had a hundred people and you wanted to find some grouping of these people into clusters. And you take these hundred people and you put them into a very small crowded room. What the fly is doing is it's taking these hundred people and spreading them out over a football field. You can imagine then that it's really easy to identify groups and delineate the boundaries between groups in the really large space compared to this really compressed space."
And of course, it works really well for identification of similarities, in this case in terms of smell.
Computers do it a totally different way. Instead of expanding the search area, or expanding the room, so to speak, computers try to narrow the scope of the search by putting similar items into buckets, effectively shrinking the room and filling it with similar stuff. And it works decently well, especially because we are rarely looking for exact versions.
This is how Saket Navlakha explains it: "If you're looking for images, then as long as they are close enough to what you're looking for, then the user is going to be happy. And so what this motivated was a different way of trying to solve this problem instead of comparing one at a time."
So computers start with one image, like a rotting banana, and try to slowly narrow down the buckets where they would find things similar to that image, like the bucket labelled "bananas."
So we see two complete opposite strategies to accomplish the same thing: The fly tries to separate out the things to compare as far from one another as possible to really find the most accurate comparison.The computer tries to reduce down the number of possible things to compare the original to.
That's the coolest part. The group that published in Science found that if they make a search algorithm that mimics the fly brain's way of searching through data and compared it to a traditional search engine, the fly was more accurate given the same amount of computing power.
They haven't tested speed or efficiency yet, but there's all the indications that the fly brain way of searching data will be better at those too.
It's more likely that the Fly Brain Search Engine will be bought by Google. The team is already in talks to expand the prototype or proof-of-concept to a real application. And it's just a matter of time before computer science can begin to mimic the power of the biological computer – the brain.