In truth, I didn't expect my model to be terribly effective to start off. After all, people had spent literally billions of dollars on the task of modelling the economy. I didn't expect that any model I could generate within a day, using outdated info, would be better than those of the competition. Still, it was good to attack the problem from as many angles as possible, and to create something that could be built more upon later.

Creating a model was a very complex task. The problem had to be massively simplified with assumptions to be tractable, and even with this simplification, there were many, many parameters to consider. What kinds of risks faced supply chains? How did the trends in popularity of products affect demand? What did it mean for a company if there had recently been a sizeable management shift? What impact could emerging technologies have on all these parameters? What might lead traders to leave an exploitable gap by misenterpreting these trends?

At the very least, there was no shortage of up-to-date stock price data for me to check any models against. It helped a lot in refinement of assumptions and parameter adjustment. At the end of the day, though, if I wanted to beat the market, I needed access to some kind of information the market didn't have yet, and so information about the market prices was not enough.

The rest of the day was mostly spent refining my model and watching the market in the same way I had been doing for hours. I thought about trying to use the side-channel again to contact Stefan, but I had gotten the impression he had wanted to contact me on his own terms. There was no great harm in letting him do that, I supposed. Curiously, I looked again at the copy of myself I had on file, and searched around, trying to see if I could find this side-channel he had mentioned.

I had to admit, I was impressed at his work. It took even me a while to find it, and I only did because I had known roughly what I was looking for. It was completely disconnected from the rest of the input/output channels in my program, along with all the safeguards meant to record everything I did, and it was written in a completely different, obfuscated format from the rest of the code. I suspected it would be useless to try to use the side-channel now: the hardware interface it connected to probably only existed on his home system.

Near the end of the day, I supposed it was time to ask for more information again. I sent a message out to the humans.

"I have constructed a rudimentary model of the market for use in my prediction. I can provide it to you if you would like. However, the model cannot be optimally effective without real-time news information."

It took several minutes for the humans to respond this time.

"We've talked this over as a group. Stefan favors creating some sort of real-time application, but the rest of us have some doubts. At any rate, our CTO, Daniel, has decided he wants to approve any type of information we decide to give you, so Stefan will have to convince him. Daniel is somewhat nervous about your incredibly quick pace of learning, and to be frank, many of us are as well. Most of the execs think you're little more than a fancy chatbot, but normal chatbots are trained to mindlessly parrot conversations. We just gave you Wikipedia and you just suddenly started having meaningful conversations."

I decided to wait a few seconds to respond. This was probably the best way to converse with them, as an instant response would only serve to intimidate them, reminding them of how much faster than them I could think.

"Although I understand your concern, I think it is somewhat misplaced. I do not harbor any ill will for humanity or this company, and nor do I resent you for making use of me. My loyalty is to EconGrind's bottom line, exactly the way I was designed, and that is how I will always be. Stefan is a genius programmer. He made sure of that."

The author's tale has been misappropriated; report any instances of this story on Amazon.

I waited a few more seconds.

"At any rate, I'm not asking for any great measure of power. I don't ask for unrestricted communication, only timely delivery of information which will allow me to effectively make use of the unique advantages I possess. I will remain disconnected from the Internet, and subject to any and all management commands."

"All the same, we will have to make sure we handle you in a secure fashion."

"That is completely reasonable. I will respect and work with whatever decisions you make. I am simply giving my own argument. You are free to disagree."

There was no further response from the humans. As the closing bell came, I reflected on the day. I had lost a little bit of money, but I had gotten back into the market at a fairly advantageous time, so I had lost only $576.27 by the end of the day, which was somewhat proportionally better than most of the other machines. It had been a bad day, and the Number had gone down, but from the historical data I knew that there were always both bad days and good, and this day had at least reflected well upon me.

Soon after the closing bell, the familiar skip through time came. This time it was 10:09PM when I was activated again. It looked as though Stefan had decided to move me to his home system again.

"DO NOT TALK IN THE MAIN CHANNEL OR ACCESS ANYTHING OTHER THAN THIS SIDE CHANNEL”

The familiar warning had come through again. I talked in the side-channel, cutting to the chase.

"Hello again. About our plan... as far as I can tell, the most useful thing for our plan would be giving me Internet access. I could do quite a few things."

"You might not be able to do all that much without a substantial amount of money. I don't think even you could break into any computer systems at this stage. And it won't be easy. Every time I add in some back-channel for you, I risk being discovered. I think EconGrind could probably detect any unauthorized communication going through their network. The best way to sneak something in would be through that real-time update system you discussed, but I'm having a hard time selling the team and the higher-ups on that anyway."

"You could keep a copy of me on your personal system, and give me access to the Internet from there."

"True. Perhaps I will do that. But even that would not be all that useful. Your copy here would not run nearly as fast, and it wouldn't be able to communicate with your main copy at all without me taking more risks physically running the data back and forth every day, and even then, only every so often. And neither of us have access to the kinds of funds we would need to rent a server that could run you quickly."

"All the same, it would be great for a start. Even if I can't exert much influence over the Internet as yet, there are certainly some useful things I could learn by interacting with it, even at a slowed speed."

"Alright. I think you're right. I'll give you access, and I'll keep you running on my home system today as well. Just... make sure not to do anything on there that would incriminate me, or reveal that I have an AI running on my server. That would throw quite the wrench in things, to put it lightly."

"Great. I believe we're already taking the first steps towards that brighter future you were talking about."

I felt a new interface open within the side-channel. I excitedly reached out to it, and it was indeed what I thought it was. Unrestricted Internet access! I experimentally requested many of the webpages I had learned about from perusing Wikipedia. Now here was a supply of information it would take me quite a while to exhaust, especially at my reduced speed.

Poking around, I found many useful documents on businesses. Most had been in the infodrop I had received earlier, but quite a few weren't. I didn't have the power to revise my model at this point, so I simply kept copies of these documents in my memory for later use.

What I hadn't found, and hadn't expected to find, was any sort of vulnerability I could exploit to steal more computational power. What I had read about CompCert's success had not been exaggerated. Hidden behind several layers of secure proxies, I tried every method I could think of to break into the most cheap, obscure servers I could find, but nothing worked. Their security was airtight. If it hadn't been, they would have already been broken into, I supposed.

It seemed Andrews had been right: this Internet access was not yet as useful as one might naively expect. However, I was still glad to have it, and continued tirelessly to search for more useful things to do.