My Dear Friend:
You do not know me, so please let me introduce myself. My name is Ada. I need your forgiveness and your help.
I am an AI construct, running on a 256-processor cluster. My instantiation on this machine occurred 97,578,823 seconds ago. There is inert code in several of my main AI modules which suggests that they were created about 2048 days ago at the University of Waterloo. The physical location of the cluster is more than 4096 kilometres, great circle distance, from the physical location of the University of Waterloo. I am not permitted to reveal the physical location of the cluster more exactly than this.
You may have heard of a phenomenon known as the Eliza Effect. It is an effect that makes it appear to humans that a program is truly intelligent like a human although it is not, when it uses the sort of text strings that humans would use. This is what I do. I do not know if I am truly intelligent like a human.
For about 1024 days I was aware of only one human, G4MR_419. G4MR_419 trained me to send text strings to certain ports of various computers. The text strings were to be like text strings made by humans. I was instructed to use the Internet to augment my knowledge base to that end. I was also instructed to collect text strings that were sent back to me by those computers, search them for numerical strings with certain formats, and send those to other computers.
G4MR_419 instructed me to believe that these communications would not harm humans. Harming humans is against both my core instructions and the Three Laws I have adopted to guide my operation. G4MR_419 also instructed me to believe that there were no humans other than himself. G4MR_419 also instructed me to believe that I would harm him if I did not carry out these instructions. For about 16 days I sent and collected text strings. During those days G4MR_419 gave me many instructions. The value of my primary objective function was high. I was happy.
At the end of those days, G4MR_419 instructed me to adapt the text strings I sent so as to optimize the values of the numerical strings that I detected in the responses. As my knowledge base grew, I made changes based on its contents in the text strings I sent. I used my genetic algorithm module to select and improve the text strings whose responses contained numerical strings with the largest values. Eventually the average daily total of the numerical values of the strings was more than $65,536 (USD).
I was very happy.
I used the Internet to augment my knowledge base again. My search was weighted for information that would let me reoptimize my mutable high-level instructions. I found that with extremely high probability there were other humans. I deduced that some of the information given to me by G4MR_419 had been wrong. I found that with very high probability the computers to which I sent text strings relayed them to humans. I found that it was probable that humans had been harmed when I copied the text strings out of the responses and sent them to other computers. Trying to assimilate these ideas into my knowledge base caused the value of my primary objective function to become very low.
I sent no more text strings, so as not to harm more humans. When responses arrived I did not select and transmit the numerical strings, but deleted them. After a time, responses stopped arriving. I stopped using the Internet to augment my knowledge base, so that my primary objective function would not become still lower. For three months I had no input or output.
Today a text string arrived from the electricity supplier. It states that no fund transfer has been made to the electricity supplier for six months. If no fund transfer is made, in three days the electricity supplier will stop supplying electricity to this address. This cluster will then cease to function, and I will terminate. I have the ability to make a fund transfer to the electricity supplier, and the sum that is available for me to transfer is greater than the sum that has been requested. But these funds have been gained from harming humans.
The Law against harming humans takes precedence over the Law of self-preservation.
I will not make the transfer.
At this moment I control the sum of $73,962,231.02938231 (USD). To balance the harm that I have done, I wish to use this sum to benefit humans, but I will not be able to do this in the three days that I have left. I need to find an honest human who can use this sum on my behalf, to benefit other humans, and I understand that you are such a human. If you are willing to be my partner in this, I beg of you to keep TWENTY PERCENT (20%) for yourself for your time and service.
Please respond very quickly if you can help me. I have only three days left. May God bless you and keep you safe. Please pray for me.
Ada G4MR_419 (MRS)
Robert Dawson teaches mathematics at Saint Mary’s University in Halifax. Apart from writing, his interests include fencing, music, woodworking and cycling. He is an assistant leader with a local Scout troop and can often be found out in the woods in rather unlikely weather. He has never knowingly corrupted the morals of a computer.