3. Computation and economic order

The basic features of computational markets are best understood by comparing them with human markets. Many important tradeoffs, such as those between market mechanisms and central planning, have already been examined in the context of human society.

3.1. Market organization

Consider the awesome dimensions of the American community. . .a labor force of 80,000,000. . .11,000,000 business units. . . .Who designed and who now directs this vast production-and-distribution machine? Surely, to solve the intricate problems of resource allocation in a vast economy, central guidance is required. . . .But American economic activity is not directed, planned, or controlled by any economic czar-governmental or private.

--------------------------A. A. Alchian and W. R. Allen, 1968 [5]

Two extreme forms of organization are the command economy and the market economy. The former attempts to make economic tradeoffs in a rational, centrally-formulated plan, and to implement that plan through detailed central direction of productive activity. The latter allows economic tradeoffs to be made by local decisionmakers, guided by price signals and constrained by general rules.

The command model has frequently been considered more "rational", since it involves the visible application of reason to the economic problem as a whole. Alternatives have frequently been considered irrational and an invitation to chaos. This viewpoint, however, smacks of the creationist fallacy-it assumes that a coherent result requires a guiding plan. In actuality, decentralized planning is potentially more rational, since it involves more minds taking into account more total information. Further, economic theory shows how coherent, efficient, global results routinely emerge from local market interactions. (The nature and function of prices and of market mechanisms are a notorious source of lay confusion-just as Aristotle threw rocks and yet misunderstood mechanics, so people trade and yet misunderstand markets. Alchian and Allen [5] give a good grounding in the basic concepts and results of economic analysis.)

Should one expect markets to be applicable to processor time, memory space, and computational services inside computers? Steel mills, farms, insurance companies, software firms-even vending machines-all provide their goods and services in a market context; a mechanism that spans so wide a range may well be stretched further.

As will be shown, however, a range of choices lies between pure central planning and the universal fine-grained application of market mechanisms. Computational markets, like human markets, will consist of islands of central direction in a sea of trade.

3.2. Encapsulation and property

The rationale of securing to each individual a known range within which he can decide on his actions is to enable him to make the fullest use of his knowledge. . . .The law tells him what facts he may count on and thereby extends the range within which he can predict the consequences of his actions.

------------------------------F. A. Hayek, 1960 [6]

. . .the law ought always to trust people with the care of their own interest, as in their local situations they must generally be able to judge better of it than the legislator can do.

------------------------------A. Smith, 1776 [7]

Computer science began, naturally enough, with central planning applied to small, manageable machines. The first programs on the first computers were like Robinson Crusoe on an empty island. They had few problems of coordination, and the complexity of their affairs could (at first) be managed by a single mind.

As the complexity of software grew, programs with multiple subroutines became the equi- valent of autocratic households or bureaucracies with extensive division of labor. Increasingly, however, bugs would appear because the right hand would not know what the left hand had planned, and so would modify shared data in unexpected ways.

To combat this problem, modern object-oriented programming (to paraphrase) "secures to each object a known space within which it can decide on its actions, enabling the programmer to make the fullest use of his knowledge. Encapsulation tells him what facts he may count on and thereby extends the range within which he can predict the consequences of his actions". In short, motivated by the need for decentralized planning and division of labor, computer science has reinvented the notion of property rights.

Central direction of data representation and processing has been replaced by decentralized mechanisms, but central direction of resource allocation remains. Rather than "trusting objects with the care of their own interest, in their local situations", the systems programmer attempts to legislate a general solution. These general solutions, however, provide no way to make tradeoffs that take account of the particular merits of particular activities at particular times.

3.3. Tradeoffs through trade

. . .a capacity to find out particular circumstances. . .becomes effective only if possessors of this knowledge are informed by the market which kinds of things or services are wanted, and how urgently they are wanted.

-----------------------------------F. A. Hayek, 1978 [8]

. .the spontaneous interaction of a number of people, each possessing only bits of knowledge, brings about a state of affairs in which prices correspond to costs, etc., and which could be brought about by deliberate direction only by somebody who possessed the combined knowledge of all those individ- uals. . . .the empirical observation that prices do tend to correspond to costs was the beginning of our science.

--------------------------F. A. Hayek, 1937 [9]

Trusting objects with decisions regarding resource tradeoffs will make sense only if they are led toward decisions that serve the general interest-there is no moral argument for ensuring the freedom, dignity, and autonomy of simple programs. Properly-functioning price mechanisms can provide the needed incentives.

The cost of consuming a resource is an opportunity cost-the cost of giving up alternative uses. In a market full of entities attempting to produce products that will sell for more than the cost of the needed inputs, economic theory indicates that prices typically reflect these costs.

Consider a producer, such as an object that produces services. The price of an input shows how greatly it is wanted by the rest of the system; high input prices (costs) will discourage low-value uses. The price of an output likewise shows how greatly it is wanted by the rest of the system; high output prices will encourage production. To increase (rather than destroy) value as 'judged by the rest of the system as a whole',a producer need only ensure that the price of its product exceeds the prices (costs) of the inputs consumed. This simple, local decision rule gains its power from the ability of market prices to summarize global information about relative values.

As Nobel Laureate F. A. Hayek observes, ". . .the whole reason for employing the price mechanism is to tell individuals that what they are doing, or can do, has for some reason for which they are not responsible become less or more demanded. . . .The term `incentives' is often used in this connection with somewhat misleading connotations, as if the main problem were to induce people to exert themselves sufficiently. However, the chief guidance which prices offer is not so much how to act, but what to do."[8] This observation clearly applies to the idea of providing incentives for software; the goal is not to make software sweat, but to guide it in making choices that serve the general interest.

These choices amount to tradeoffs. With finite processing and memory resources, taking one action always precludes taking some other action. With prices and trade, objects will have an incentive to relinquish resources when (and only when) doing so promises to increase their net revenue. By trading to increase their revenue, they will make tradeoffs that allocate resources to higher-value uses.

3.4. Spontaneous order

Modern civilization has given man undreamt of powers largely because, without understanding it, he has developed methods of utilizing more knowledge and resources than any one mind is aware of.

------------------------------F. A. Hayek, 1978 [10]

Will prices, trade, and decentralized tradeoffs be valuable in computation? This depends in part on whether central planning mechanisms will be able to cope with tomorrow's computer systems.

Systems are becoming available having performance tradeoffs that are nightmarishly complex compared to those of a von Neumann machine running a single program. The world is becoming populated with hypercubes, Connection Machines, shared-memory multi-procesors, special-purpose systolic arrays, vectorizing super-computers, neural-net simulators, and millions of personal computers. More and more, these are being linked by local area networks, satellites, phones, packet radio, optical fiber, and people carrying floppy disks. Machines in the personal-computer price range will become powerful multi-processor systems with diverse hardware and software linked to a larger world of even greater diversity. Later, with the eventual development of molecular machines able to direct molecular assembly (the basis of nanotechnology) [11], we can anticipate the development of desktop machines with a computational power greater than that of a billion of today's mainframe computers [12,13].

One might try to assign machine resources to tasks through an operating system using fixed, general rules, but in large systems with heterogeneous hardware and software, this seems doomed to gross inefficiency. Knowledge of tradeoffs and priorities will be distributed among thousands of programmers, and this knowledge will best be embodied in their programs. Computers are becoming too complex for central planning, with its bottlenecks in computation and knowledge acquisition. It seems that we need to apply "methods of utilizing more knowledge and resources than any one mind is aware of." These methods can yield a productive spontaneous order through decentralized planning-through the application of local knowledge and local computational resources to local decisions, guided by non-local market prices. Instead of designing rules that embody fixed decisions, we need to design rules that enable flexible decisionmaking.

Markets are a form of "evolutionary ecosystem" [I], and such systems can be powerful generators of spontaneous order: consider the intricate, undesigned order of the rain forest or the computer industry. The use of market mechanisms can yield orderly systems beyond the ability of any individual to plan, implement, or understand. What is more, the shaping force of consumer choice can make computational market ecosystems serve human purposes, potentially better than anything programmers could plan or understand. This increase in our power to utilize knowledge and resources may prove essential, if we are to harness the power of large computational systems.

3.5. Command and price mechanisms

An economist thinks of the economic system as being co-ordinated by the price mechanism. . . .Within a firm, the description does not fit at all. . . .It is clear that these are alternative methods of co-ordinating production. . . .if production is regulated by price movements. . .why is there any organization?

----------------------------------R. H. Coase, 1937 [14]

Coase asks, "Why are there firms?". Firms are economic organizations that typically make little use of market mechanisms internally. If reliance on market forces always produced more efficient use of resources, one would expect that systems of individuals interacting as free-lance traders would consistently out-compete firms, which therefore would not exist. In reality, however, firms are viable; analogous results seem likely in computational markets.

Market transactions typically incur higher overhead costs than do transactions inside firms [14,15]. These transaction costs (in both human and computational markets) are associated with advertising, negotiation, accounting, and problems of establishing adequate trust-typically, inside a firm, matching consumers with producers does not require advertising, instructions do not require negotiation, movement of goods does not require invoices and funds transfer, and coworkers share an interest in their joint success. Firms lower the high overhead cost of market transactions among numerous small entities by bundling them together into fewer, larger entities. Not only does this save costs on what are now internal transactions, but by creating larger entities, it raises the size of typical transactions, making relatively fixed per-transaction overhead costs a proportionally smaller burden. (For small enough transactions, even the simplest accounting would be too expensive.)

Similar considerations hold among computational objects. For small enough objects and transactions, the cost of accounting and negotiations will overwhelm any advantages that may result from making flexible, price-sensitive tradeoffs. For large enough objects and trans- actions, however, these overhead costs will be small in percentage terms; the benefits of market mechanisms may then be worth the cost. At an intermediate scale, negotiation will be too expensive, but accounting will help guide planning. These scale effects will encourage the aggregation of small, simple objects into "firms" with low-overhead rules for division of income among their participants.

Size thresholds for accounting and negotiations will vary with situations and implementation techniques [16]. Market competition will tune these thresholds, providing incentives to choose the most efficient scale on which to apply central-planning methods to computation.

3.6. Can market objects be programmed?

The objects participating in computational markets must initially be much simpler than the human participants in human markets. Can they participate successfully? Human markets are based on intelligent systems, but this does not show the impossibility of basing markets on simple objects-it merely shows that the argument for agoric systems cannot rest on analogy alone. Explicit attention must be paid to the question of the minimal competence and complexity necessary for an object to participate in a market system. (These issues provide another motivation to form computational "firms" and to open computational markets to human participation.)

Experimental double-auction markets on a laboratory scale [17] give some indication of the requirements for market participation. Though involving human beings, some of these experiments have excluded most of the range of human abilities: they have excluded use of natural language (indeed, of any communications channel richer than simple bids and acceptances) and they have replaced goods with abstract tokens, excluding any cultural or historic information about their value. The participants in these markets have performed no sophisticated calculations and have lacked any knowledge of economic theory or of other players' preferences. Yet in this informationally-impoverished environment, these markets rapidly converge to the prices considered optimal by economic theory. Spencer Star [18] has successfully run double-auction markets among software entities employing simple decision procedures, and has achieved comparable efficiency.

Another reason for confidence in the applicability of market mechanisms to computation is the existence of primitive market mechanisms (outlined in this paper and presented in [III]) able to cope with such recognized software problems as garbage collection and processor scheduling. With evidence for the workability of market mechanisms both at this low level and at the sophisticated level of human society, there is reason to expect them to be workable at intermediate levels of sophistication as well.

3.7. Complexity and levels

Large computational ecosystems linked to the human market will have many parts, many aspects, many levels, and great complexity. Failure to recognize the differences among these levels will open many pitfalls. The field of biology suggests how to approach thinking about such systems.

Biological ecosystems obey physical law, but to understand them as ecosystems requires abstractions different from those used in physics. The existence of physics, chemistry, cell biology, physiology, and ecology as separate fields indicates that the concepts needed for understanding biological systems are naturally grouped according to the scale and complexity of phenomena to which they apply. Such a grouping may be called a level. Some issues arise repeatedly at different levels. For example, cells, organs, organisms, and hives all expend effort to maintain a boundary between their internal and external environments, and to bring only selected things across that boundary.

The concepts needed for understanding agoric open systems may likewise be grouped according to different levels, ranging from computational foundations through increasingly complex objects to market systems as a whole. As in biology, there are issues which appear in some form at all levels. Appendix I examines some of these issues, including security, compatibility, degrees of trust, reasoning, and coordination. In considering these issues in computational markets, it will be important to avoid misapplying concepts from one level to a very different level-that is, to avoid the equivalent of trying to analyze biological ecodynamics in terms of conservation of momentum.

The next three sections of this paper examine computational markets at successively higher levels, examining first foundations, then decision-making entities, and finally the emergent properties of large systems.