The Awakening Marketplace

Questions

Some of the questions of interest to researchers concern the scenario which is most likely to produce the first synthetic intelligences:

  • Will superintelligence appear as the result of the internet "waking up"?

  • Will it be the product of a company, a government or a lonely genius researcher?

  • Which area contains the driving problem(s) which are most likely to result in the production of a synthetic intelligence?

Scenarios

There have been several proposals in this last area:

  • Super librarian scenario
    One possibility is that machine intelligence so will the rise at the hands Of an internet search company. Search companies are well placed to develop machine intelligence - since they have access to large quantities of hardware, lots of training data - and they need to perform searches intelligently in order to provide a good quality of service to their customers.

  • Super security scenario
    Another possibility is that machine intelligence will be developed by the government. The government already use enormous computers and sophisticated scanning techniques to collect and analyse the world's data - in secret research facilities. They have some of the smartest minds on the planet working for them - and nobody who can say knows how far along their projects are.

  • Super trader scenario
    Another possibility involves the stock market waking up. The driving problem in this scenario is how to allocate the world's financial resources. The marketplace forms a network of intelligent agents - some human, some machine - trying to solve the problem, by deciding where to invest their stocks. Over time, the machine portion increases - and eventually a huge and collectively very smart AI distributed across the many countries of the world is solving most of the problem.

It may make quite a difference which path bears fruit first - because the agents in these scenarios are likely to have rather different utility functions.

The purpose of this essay is to draw attention to the super trader scenario.

The Awakening Marketplace

Humans have already developed a means of representing utility. It is known as money. Wealth is a universal scalar quantity that effectively controls who gets access to resources. Traders act as though they are attempting to maximise this quanity - much as expected utility maximisers act so as to maximise expected utility.

Just as humans act in such a way as to increase their access to resources, so machines can be expected to do so as well. An important way of doing this is by earning money - by perfoming tasks others find valuable.

A superintelligence may want to get rich quickly - and one way in which humans do that is by trading in stocks and shares.

To illustrate the potential in this area, consider the case of James Harris Simons Simons earned an estimated $2.8 billion in 2007, $1.7 billion in 2006, $1.5 billion in 2005 - essentially by using computers to play the stock market.

This path to superintelligence seems likely to be well funded.

The decision problem for stock market traders is, effectively, how best to allocate your section of the world's financial resources.

This is a broad problem, requiring a sophisticated understanding of the world for best results.

It is also a problem and dominated by continuous competition. Traders effectively compete with each other for access to resources. How well you are doing depends a great deal on whether you can quickly spot opportunities in the marketplace which others have missed.

The decision problem of how best to invest financial resources is one which lends itself to being solved by a distributed network:

  • A secret governmental machine in an underground bunker cannot so easily take advantage of resources spread across the internet.

  • Similarly, a super librarian typically has to answer questions in real time. A distributed network would probably be too slow to be very effective at this task.

The ability to exploit resources in multiple countries seems to dencrease the time until the available hardware resources are available.

Some of the easier trading problems can be solved by computers today. As they improve, machines will take over from human traders in this domain gradually.

Implementation details

There are reasons to think that future intelligent agents will be able to improve themselves. This will require a range of skills, including hardware engineering - but probably the single most important skill initially will be computer programming.

So, after a certain point, intelligent agents are likely to be skilled software engineers.

Does this mean we need to add a super-programmer scenario to the above list - and give it considerable weight?

That is one perspective - but another view is that software engineering may be regarded as an implementation detail - rather than as an end in itself. Sophisticated computer programming skills will be needed, but - at the end of the day - there still needs to be an application domain: the programs have got to do something.

Consequences

Since it appears that stockmarket superintelligence is a fairly probable outcome, the question arises of whether it is a desirable outcome.

One criticism is that money inaccurate proxy for what humans really want. A superintelligence whose aim in life was to create wealth might take over the world bank, and use it to print dollar bills - resulting in lots of money, but little of value.

Valuing something with more intrinsic worth would eventually result in similar problems. Gold is valuable today - but if the planet was stripped mined and reduced to rubble in a quest for it, its market value would plummet.

Using the value of a comprehensive portfolio might minimise damage caused in this way - but creating a complete portfolio would be extremely challenging.

Of the scenarios listed, it probably has the best chance of being a distributed technology - which may reduce the chance of it only helping a tiny minority.

The scenario at least seems more promising than a governmental or military AI.

Links


Tim Tyler | Contact