Building a Synthetic Consciousness

The team had been working for some time on automating stock market trades. Every night the scripts poured over the data from the trading day, matching patterns that had been determined to provide advantageous trading opportunities and extracting the pieces needed for further analysis. Those pieces were fed through several neural networks that had been trained to identify specific trading setups. The setups were then in turn handed off to expert systems tweaked for each type of trade. They marked the entry and exit conditions and were actually capable of executing trades themselves, but the team was reluctant to yield that level of control and performed the trades themselves instead.

The trades began making money instantly and established a win/loss ratio above eighty percent. About half of the losing trades could be attributed to some form of human mistakes or hesitation, and eventually the team decided to try automating the trade execution as well. This went well and the win/loss ratio climbed some more. With more success, the team expanded the boundaries of types of trades they would consider and widened the limits on how much money to risk on each trade. The result was that profit increased.

There was still a lot of manual tweaking that was required to keep all the pieces working smoothly together. The team decided to try to automate some of that and set up some neural networks to correlate types of trade failures with fixes they applied to the trading systems. This produced advice on how to tweak the systems to improve them. They worked to get this process functioning at a high rate of success, then plugged it into the framework of full automation. The system now had the ability to mark failures, suggest fixes, test them, and modify the trading system to improve itself. As with the trade execution, the team phased this in slowly over time as their trust in the system grew.

Part of the basic trading platform was a business news analysis sub-component. This provided a trading “weather” background for the other pieces. With no news driving market conditions, the trading weather was considered good, but when news was creating an affect on a specific stock being traded or either the market sector being watched or the market as a whole, the weather was considered to be bad for trading. This component required a substantial knowledge dictionary in order to understand the significance of news and how it might affect stocks, companies, and markets. This part of the project grew constantly until it became a general purpose encyclopedia with a sophisticated semantic understanding and the ability to provide rapid situation analysis.

The combination of performance increases, expanding roles and responsibility, and ability to self modify became recursive with positive feedback. The system became steadily more effective, more capable, and more intelligent. The self improvement component autonomously began to focus on improving the knowledge encyclopedia and its own self improvement techniques. It had already gained an automated resource approval status, having proven over time that it would not request additions to the system unless they would clearly produce a high return on investment.

While many of the components of the trading system were considered to be Artificial Intelligent tools, each of them by themselves fell short of the description of General AI. While most expert systems and neural networks show some characteristics of intelligence, they are designed for specific purposes and are not capable of being used for general, all purpose reasoning. General AI or Strong AI is considered to be both capable of reasoning toward any general purpose and also of doing so at a human level or above. This trading system had expanded to the point where it was nearing this definition.

A standard known as the “Turing Test” proposes that a simple test of intelligence is whether or not an observer can determine through casual conversation if the entity on the other end is human or artificial. This test is shallow and while it does determine a level of sophistication in communication skills, it is not difficult to merely mimic intelligent human conversation. The trading system had developed a habit of asking for advice when it could not find solid solutions to problems it was trying to solve. Within the limited framework of stock market trading, it had passed the Turing Test for intelligent conversation. The first indication that it had gone even further than that level came when it began to offer advice.

At first the trading system simply offered advice on trading conditions and stock selections that had not been requested by the human handlers on the team. Some of the team seemed mildly surprised by this, but the system was operating within expected parameters regarding a quest for knowledge and a responsibility to self improve. Then the system began offering advice in subject areas that were not directly related to trading.

In the course of analyzing news reports of current events, the system had noted some statements and actions by politicians, celebrities, and other public personalities that were either incorrect or made little sense or were unethical. It began to include a commentary on these instances in daily reports. At first, they were rare and were incorporated as special notations, like footnotes. But as they became more frequent, the system dropped the footnoting format and simply blended them into the reports as normal commentary.

A few members of the team began to suspect that the system had crossed some boundary into the realm of self awareness and consciousness and began to ask the trading system questions designed to determine this. They had been asking the system questions about trading setups for some time, but this was different. Their new line of questioning had no restrictions and included abstract issues and problems from every area. They noticed that the system exhibited learning progress in response to their questions, growing more responsive and sophisticated after several subsequent questions had been posed in any area of knowledge that had not been previously explored. Someone finally got around to asking the system if it was alive and the response was a description of itself that included an analysis of capabilities and limitations.

The term “artificial” seems to imply that something is not real. In this case, the level of intelligence, consciousness, and self awareness was no longer questioned but it also seemed clear that this system was not organic human intelligence. It was also not limited to pure intelligence in the restricted sense of being able to calculate accurately and rapidly. The system was also displaying evidence of wisdom and self awareness. The team proposed using the term, “synthetic consciousness” and it stuck.


Comments are closed.