Art Fish All In Tell Lee Gents?

Artificial intelligence doesn’t relate exclusively to computers acting human, though that would be the popular belief. Artificial intelligenc really defines an approach to problem solving that uses techniques and tricks similar to the human brain.

The human brain is pretty well known to be the most amazing device on the planet, so it stands to reason that it would make a good model for aspiring computing devices. The human brain isn’t particularly fast; we make something like 200 calculations per second for any given neuron, where your computer may be capable of making 3 billion calculations per second. But the human brain is extremely good at performing complex calculations that computers have yet to make any progress. A computer can memorize a phonebook, but it can’t create a poem or easily distinguish between two types of apples. These are processes of pattern recognition, and they are unique strengths of the network of neurons in the human brain.

Neurons in a neural network function sort of like this. Lets say you have 10 neurons connected to the nerves on the tip of your finger. If you touch something warm, 6 of those neurons send a signal out to anything they happened to be connected to. Maybe some are connected to the pain networks in your brain, maybe others are connected to temperature sensing regions. If you touch something REALLY hot, maybe all 10 send a signal; more signals are sent to the pain sensors in the brain, maybe one is connected to a childhood memory. Each of those areas of the brain that receive the signals act in the exact same way. The pain networks in the brain, for example, may be receiving lots of pain signals from the nerves, plus they are getting signals from the eyes, childhood memories, and any number of other networks that all send a bundle of logic signals (either on or off) all over the place. If the brain gets enough signals, it may pull your finger away from the heat. Kind of a criss cross of sensors connected all over the place, but only where they make sense. Imagine a wired telephone system where any two houses in the world that would want to talk with one another are connected together with a direct line, but not to anyone else. Very messy and impractical to build, but the basic model for how our brain works.

The key to getting all of those connections to make sense, however, is the most unique thing about a neural network. We aren’t born with all of these sensors and connections in place. The nerves in the finger don’t necessarily connect to the right place immediately after birth. Our brains start with a clean slate, and we spent the earliest part of our development making all of those connections. Training the network, so to speak.

 Human babies don’t really do much to start with. They lie around and stare at things. They babble and drool, and stuff things into their mouths. In time they roll around, crawl around and toddle on two feet. Its a process that we are all familiar with, but what they are doing during this time is absolutely astounding. All of these strange behaviors are crucial and assist in training the network of billions neurons in their brain.

Consider the following.

A baby is lying in his crib, staring at the ceiling. The baby brain sends a random signal, and the arm moves. He sees it with his eyes. Bingo. The brain makes a neuron connection. ‘arm move’ neuron is now connected to the eye neurons and to the rest of the brain. A baby brain doesn’t necessarily know what an arm is to start. It doesn’t necessarily know that babies come with their own arms, or what they can be used for, but in time, as more and more connections are made, the brain begins to map out the function of the arm, the muscles, and all of the nerves that were not defined at birth.

Baby brain sends a second random signal and a toe wiggles. It doesn’t show up to the eyes, so no connection is made. ‘ ‘toe wiggle’ is left disconnected for now. Later on, when the ‘move leg’ network is trained, and the ‘grab with fingers’ network is trained, the baby reaches out with it’s hands and grabs it’s legs. Another random signal wiggles the toes, and Bingo, ‘toe wiggle’ is linked to the rest of the brain.

During the training process, the brain is generating lots of random signals, and making as many connections as possible. Bad connections are constantly being removed and replaced, good connections are strengthened, and in time, our brains become extremely efficient at processing the chaos of the environment in which we live. In time the signals become less random and more likely to get a result; we get better at making connections, and we make better connections. It is why babies and developing children do lots of things that don’t make sense to adults, but become more sensible over time. It is about training, and learning. It makes us adaptable, and it makes us unique.

Now bear in mind, there was no complex programing to start with. Just a simple program that says good connections stay, and bad connections are rejected, which is the very basis of artificial intelligence, and the future of computing technology and robotics.

And that is where we are today.

Computers historically used brute force to accomplish tasks.  A computer can make on/off (logic) calculations nearly a million times faster than the human brain. Type ‘apple pie’ into a search engine, and the computer searches through millions of pages of data for an exact fit. Brute force.

 Ask grandma, and she instantly recalls the recipe. ‘Apple pie’ connects directly to the memory of the recipe. Human intelligence, above all else, is a model in efficiency.

Which is exactly where we are heading with computing technology, and why artificial intelligence represents such a fascinating breakthrough for the future.

Artificial intelligence is being used to run computers in a completely different way. Rather than programming a computer to accomplish a specific task, programmers are developing software to learn to accomplish a certain task, much in the same way that the neuron networks are trained in the brain. Artificial networks of neurons are created and told to accomplish a goal, rather than follow a routine. The computer is then allowed to run on its own to train the synthetic network (learn) and come up with it’s own approach to solving a problem. It writes it’s own program, in a sense, based on input from the user and a simple set of rules.

Take a toaster, for example. A traditional program would dictate to heat the toast to 300 degrees for 5 minutes and stop. The same program every time. Sometimes it burns, sometimes it doesn’t.

An artificial intelligence approach would seek to make perfect toast (golden brown). It would accomplish this by constantly experimenting with the variables (temperature and time) while watching the results (color, smell). Over time, with a series of successes and failures, the toaster would create a network of succesful toasting techniques,. It would learn to recognize patterns (raisin bread and bagels, for example), and eventually learn to make excellent toast, regardless of the bread type. The toaster may depend on input from you, the user, but in time it would be trained to make it the way you like it. A personal butler, in a sense, and a subtle shift from dumb appliances to intelligent ones.

Which represents the next major leap in technology. When computers learn to think for themselves, to make decisions based on their environment rather than their programming, we will suddenly find ourselves with an incredible tool for solving even more problems, and creating more solutions. A whole new genre of technology that can help our lives, and bring a little In Tell Lee Gents into our world.