Tuesday, 30 October 2007

The Artificial Child

I've been thinking about Stephen Spielberg's movie Artificial Intelligence. In that movie, a mother and family buy an artificial child and then desert the child when they realize it is defective.

Like many people, I have a desire for children. I believe this desire is a natural instinct preserved through many generations of human evolution. However, children are very expensive. You have to buy a bigger house, bigger car, pay school fees, university fees, and then pay for costs like clothes, food, medical care, etc. The child can distract you from your job and thereby cost you a promotion or lower your salary because you work less to spend more time with the child.

How fantastic would it be then to have access to an artificial child, a child whose features are accurate enough to satiate evolutionary instincts for paternalism or maternalism while still being cheap and disposible? An artificial child is an area of consumer demand that is so far untapped by business. I suggest building an artificial child.

What will this child look like? The child will be just like a human. It will have a head, two arms, and two legs. Whether this child will look like a human child or whether it can be made to look like cute non-humans like teddy bears is yet to be decided. One thing that I think is good is the ability to grow. The artificial child can easily have limbs that grow over time. Of course, it is probably impossible for a small baby-sized machine to grow to the size of an average adult, which is why growth will need to be capped off at a certain point. The artificial child will not be able to grow to adulthood but will remain a child.

There are three inputs to the artificial child: (1) vision, (2) audio, and (3) movement. Vision is what the artificial child can see. The child has a camera where its eyes are and can see its environment. The images picked up will be sent to a computer. Audio acts the same way as vision except the microphone picks up sound. There will be voice recognition software so that the child can understand human speech. Movement comes from its limbs. Parents like to touch their baby and move the baby around. When the customer touches and moves around this artificial child, the movements have to be received, digitalized, and then transmitted to the computer. For example, if the child is picked up, the baby will have to recognize this.

There are two outputs: (1) audio and (2) movement. The child can talk through speakers. The child can move its limbs. It can walk around, point, etc.

How do you coordinate output and input? What comes to most people's minds is artificial intelligence. However, this isn't necessary. No AI is needed for this artificial child to be lifelike. My solution is to take a real child and then monitor its life. Whatever the real child experiences it sotred into a database. Careful attention is paid to the three inputs and two outputs and especially how output reacts to certain inputs. The aim is to make the artificial child mimic the behavior of the real child. The actions or output of the artificial child will depend on probability and the probabilities will come from the real-life data collected from monitoring of the real child.

Suppose you pick up an actual baby. You do this many times. You notice that when you do this the baby receives no resitance from its feet and arms but it does sense touch on its back and bottom. After picking up a real baby hundreds of times you also notice that, say, 70% of the time the baby laughs and, say, 80% of the time it flails its arms and legs around. All this data is put into a database and stored in the memory of the artificial child. When the artificial child realizes that it is being picked up through its three inputs, i.e. there is no resistence to its feet, someone is touching its back and bottom, etc, then the artificial child will recognize this and then with 70% probability will laugh (by playing the audio) and with 80% probability will flail its arms and legs around.

More than one real baby can be monitored and the data can all be averaged out. The more real babies you monitor the better your data and the more life-like your artificial child will be.

It is important for the artificial children to act their age, so obviously if the artificial child is one year old it will not mimic the behavior of a real 10-year-old child. That would ruin the illusion.

No comments: