Hi, I'm Tim Tyler, and this is a video about The Matching Pennies Prize.
The Matching Pennies Prize is a prize for automating machine forecasting.
The matchingpennies.com web site will
regularly hold tournaments based on the matching pennies game, and the winners
will be awarded prizes.
Matching Pennies
The Matching Pennies game is a standard game used to illustrate strategies
within game theory. It is one of the simplest demonstrations of a mixed strategy Nash
equilibrium. The game represents an abstract model of serial forecasting in a simple
environment partly consisting of agents with conflicting goals.
Agents succeed in the environment by modeling the other agents and predicting their
actions.
The basic idea of the tournament is to arrange things so that the most successful agents
will necessarily have to have made good progress on the problem of successfully
performing general inductive inference - by building a powerful, general-purpose stream
compressor.
This will then enable them to implement Occam's razor, develop plausible models of the
workings of other nearby agents - and so make successful predictions of their actions.
I've created other material which explains the significance of automating predictions
and forecasting in more detail. Here, I will just note that an automatic forecasting
component is a key ingredient in many machine intelligence systems - and that if you put
a forecasting component together with a tree-pruning algorithm and an
evaluation function, then the result is a complete machine intelligence.
Prize function
The basic idea of the prize is to allow those who are interested in funding machine
intelligence to contribute to the efforts of those who are interested in building it.
Computer science benefits
Next, a few words about the potential benefits of such systems.
Intelligence is one of humanity's key assets, and augmenting human intelligence with
computers has led to many of the more recent improvements in the world. Computer science
has led to an increase in humanity's networking potential, helping to socioeconomically
unite the world - and has helped us to compensate for our poor memories, poor
communication capabilities, inferior thinking tools and other mental weaknesses.
Broader-spectrum machine intelligence looks set to be one of the most important
inventions mankind will make in the future. It potentially offers many benefits
to humans. Intelligent machines could eliminate drudge labour, advance science,
technology and medicine and help humanity to solve many of its problems.
Potential problems
However, in unscrupulous hands, it looks as though machine intelligence could
potentially create enormous wealth inequality - with most of the benefits accruing
to a tiny minority.
It seems pausible that such scenarios are most likely if intelligent machines are
initially developed in secret by a powerful and unscrupulous elite.
Openness and Transparency
Open development suffers from the drawback that unscrupulous agents will have access to
the results. However, more benevolent agents having access to the code is a
major plus point - since that allows them to defend themselves and others.
Development being conducted transparently and in public - with the eyes of the world
watching - should result in the best chances of the project being managed successfully -
with the proper safety precautions being taken.
Having the maximum possible number of eyes on the code should help to identify any bugs
and problems. Also, having direct access to the source code means that ethical
individuals are in the best possible position to defend themselves and society if
problems with unethical corporations or organisations do arise.
Open source software is a well established software development model. It leads to
popular products with low defect rates - and it acts as a foundation stone in many areas
of IT today.
Trade secrets
Secret development in private allows greater possibilities for fuelling development via
exchanges with customers. This is a model that drives much of the economic activity
associated with information technology today.
However, society should probably deliberately take steps, in an effort to
ensure that such organisations are beaten during the development of machine
intelligence.
We have an example of what can happen when important information technology systems gets
into the wrong hands - in the form of Microsoft. Their operating system monopoly led to
the most appalling period of stagnation and consumer exploitation and oppression in the
history of computer science.
It seems likely that machine intelligence will also have abuse potential. Like
operating systems, machine intelligence seems likely to be a powerful technology - but
one which is morally neutral, and thus capable of being used for positive or
negative purposes.
What society probably doesn't want is for many of its members to get locked
into a patented, commerical machine intelligence system - and then have it start to milk
them once they are dependent on it, for the benefit of an elite club of shareholders
somewhere.
The technology of machine intelligence seems too important to allow a self-interested
minority to monopolise it - either via software patents or trade secrets.
Open source
Open source solutions are probably the best available defense for those who want to
avoid being exploited by the manufacturers of proprietary software. It is about the only
type of remedy that has a track record of being successful in this area.
Closed-shop organisations will have difficulty in getting anyone to trust them. I
propose that those with the relevant skills capitalise on this, and cooperate to help to
promote an open solution.
Take action
To help prevent machine intelligence being monopolised, I invite interested parties to
help to promote an open-source solution that is less likely to promote the interests of
an elite minority, and more likely to benefit all humans.
Those who can write code, can contribute by developing software. Others can help by
offering them incentives to do so.
For more comprehensive details about The Matching Pennies Prize, the prize fund,
the award schedule, or details about how to compete or donate please visit matchingpennies.com.