Autonomous Electronic Markets
Markets have traditionally operated in a vacuum of information leading to myriad inefficiencies. For example, sellers exploiting buyer ignorance and distance between retailers have set their own fixed prices and imposed them on customers. And manufacturers at the mercy of inaccurate demand predictions are often stuck with either too much inventory or not enough. But with the advent of e-commerce, some of these inefficiencies are starting to shrink. Geography matters less and with the touch of a few keys consumers have ample access to pricing information and a product's "true" value. With distributed online information, markets are coming into a new era of interactive pricing involving the buyer and seller alike. Some suggest the Internet could be an instrument that drives capitalism towards its purest form -- a seamless, efficient, and self-correcting system of supply and demand.
The Environmental Marketplace
Today, environmental commodities can be bought or sold like any other. For example, under the Clean Air Act Amendments (CAAA), companies can buy and sell the right to pollute based on a system of trading emissions "allowances" on the free market. Under the law, companies emitting less than their assigned caps on emissions can sell allowance credits to other companies that exceed their own caps. This enables the polluter to choose the most cost-effective option for achieving compliance: either buying pollution credits or installing stack scrubbers and other expensive equipment to clean their emissions. The only stipulation is that regional air quality continues to meet federal standards. Companies bid for the allowances on an open market in which prices vary according to market conditions.
Computer-assisted environmental markets for clean air allowances are now in place in a number of areas. For example, the Regional Clean Air Incentives Market (RECLAIM) of the South Coast Air Quality Management District in Los Angeles, California, is using a system called Automated Credit Exchange (ACE) for trades in sulfur dioxide and oxides of nitrogen. The multiple land zones, applicable years of emissions, and overlapping regulatory cycles have so many elaborate combinatorial options there's no way a company could price the credits, nor realize financial benefit from the system, without computer software. On the ACE market, companies can buy or sell credits for different years and regulatory cycles at one point in time. The package calculates the best way to sell the credits by evaluating all the possible trades and calculating market prices for each part of the credit package. John Ledyard, a professor at the California Institute of Technology and one of the designers of the ACE system used in Los Angeles, says the success of the program shows computationally assisted markets not only create financial liquidity in these complex situations, they also enable market-based regulatory policies to succeed.
Because pollution credits in Los Angeles are sold infrequently, once a week at most, there's little need for rapid Web-based communications. But in a more active system with frequent trading, real-time communications on the Internet could be indispensable. One framework in which the Internet could play a vital role is the Kyoto Protocol. This multinational treaty has a target of reducing worldwide greenhouse gas emissions to 5 percent below 1990 levels by the 2008-2012 time frame. One way it aims to achieve this is by establishing a mechanism for trading credits (designated "assigned amounts units" or AAUs) for carbon dioxide, a gas suspected of contributing to global warming. As with the CAAA, under the Kyoto Protocol it doesn't matter who cuts their emissions, just so long as there is a net global reduction. Assuming the treaty were implemented, companies throughout the world would be bidding for AAUs on a global market -- across time zones, currencies, and more. According to leading e-commerce researcher and Caltech professor Charles Plott, real-time transactions involving all the various players will be crucial in order to optimize the efficiency of the system.
Plott and his colleagues at Caltech's Laboratory for Experimental Economics and Political Science have recently created a model through which companies can bid for AAUs on the Internet. The model is called the International Greenhouse Gas Emissions Trading Marketspace, and it was developed in collaboration with the Paris-based International Energy Agency. Participants using the model can project emissions trends into the future, simulate sudden variations in their inventories, and calculate the effects of domestic emissions policies on AAU pricing. Empowered with this information, they can compare the cost of buying the credits versus taking action to reduce emissions.
Eventually, future e-markets could become so complex, with so many participants and competing market variables, that human oversight would be a hindrance. Researchers and software companies have set high hopes on so-called software agents, which learn about their users' interests and act independently on their behalf. IBM researcher Jeff Kephardt describes the "autonomous" e-markets of the future as "a seething milieu in which billions of economically motivated software agents find and process information and disseminate it to humans and, increasingly, to other agents." According to Kephardt, agents will evolve from facilitators to decisionmakers, while their degree of autonomy and responsibility increases with time. Ultimately, he predicts, transactions among software agents will become an essential, and perhaps even dominant, portion of the world economy.
To computer scientists, particularly specialists in artificial intelligence, agents are nothing new. Researchers have been trying to build prototype agents for at least 20 years. A number of fairly primitive versions are in commercial use today -- doing things like sorting e-mail and customizing web pages. But these agents, while useful, still rely on the initiative and programming ability of the user. A class of more advanced commercial agents called "bots" (e-slang for robot) are now being designed to reside on the Web and do things like compare prices for online retail items or disable viruses before they infect your computer. However, even these provide only a glimpse of the more intelligent versions yet to come. Patty Maes, an artificial intelligence guru and professor at the Massachusetts Institute of Technology, predicts agents will eventually become "robust and adaptive, capable of learning from experience and responding to unforeseen situations with a repertoire of different methods." An ideal agent, she says, should sense the current state of its environment and act independently towards its goal.
Maes is a pioneer in an emerging field called "artificial life," that seeks to apply biological principles to networked devices. Artificial-life agents are truly autonomous, in that they program themselves. According to Maes, these autonomous agents will even undergo "artificial evolution," resulting in an electronic ecosystem housed in the next century's computer networks. True to Darwinian form, successful agents will survive and reproduce, while those that fail are purged.
Just how these autonomous e-markets would function in the environmental sector remains to be seen. Should a worldwide framework for emissions trading be established under the Kyoto Protocol, or some other analogous agreement, it's possible that agent-based systems could handle huge volumes of international transactions among industrial facilities. Caltech's Ledyard suggests autonomous e-markets might emerge in a hypothetical emissions trading program for mobile sources, for example, cars in an urban area. Furthermore, networked sensors could be employed to police the system and make sure reductions targets were achieved.
Trust and Security
Those who see the emergence of autonomous e-markets as just another example of technology run amuck have some legitimate concerns. After all, what's to stop a highly evolved agent from turning into a self-replicating virus? Furthermore, agents in an e-market are delegated with buying and selling commodities -- which means they have access to money. An agent could be subject to attempts to manipulate its buying decisions in favor of another, malicious user that in effect converts it to an online "e-zombie." This may sound like science fiction but the issues are real -- security concerns are critical and may never be completely resolved.
Another potential problem is trust: How can scientists ensure that agents will perform as instructed? Maes acknowledges that delegating responsibility always involves some loss of control -- whether the responsibility is passed on to a human or a machine. Nevertheless delegating responsibility is essential if efficiency is to be achieved. Building security into these systems, as well as some measure of flexibility to accommodate error, will be an ongoing challenge for scientists in the years to come.
Next section: Trust, Technology, and Society