Category: Web/Tech

This started in the late 1980s, and was led by GMU economist Don Lavoie, who earlier had been a computer programmer.  Here is one bit from Don’s extensive essay, co-authored with Howard Baetjer and William Tulloh:

The market for scholarly ideas is now badly compartmentalized, due to the nature of our institutions for dispersing information. One important aspect of the limitations on information dispersal is the one-way nature of references in scholarly literature. Suppose Professor Mistaken writes a persuasive but deeply flawed article. Suppose few see the flaws, while so many are persuaded that a large supportive literature results. Anyone encountering a part of this literature will see references to Mistaken’s original article. References thus go upstream towards original articles. But it may be that Mistaken’s article also provokes a devastating refutation by Professor Clearsighted. This refutation may be of great interest to those who read Mistaken’s original article, but with our present technology of publishing ideas on paper, there is no way for Mistaken’s readers to be alerted to the debunking provided by Clearsighted. The supportive literature following Mistaken will cite Mistaken but either ignore Professor Clearsighted or minimize her refutations.

In a hypertext system such as that being developed at Xanadu, original work may be linked downstream to subsequent articles and comments. In our example, for instance, Professor Clearsighted can link her comments directly to Mistaken’s original article, so that readers of Mistaken’s article may learn of the existence of the refutation, and be able, at the touch of a button, to see it or an abstract of it. The refutation by Clearsighted may similarly and easily be linked to Mistaken’s rejoinder, and indeed to the whole literature consequent on his original article. Scholars investigating this area of thought in a hypertext system would in the first place know that a controversy exists, and in the second place be able to see both (or more) sides of it with ease. The improved cross-referencing of, and access to, all sides of an issue should foster an improved evolution of knowledge.

A potential problem with this system of multidirectional linking is that the user may get buried underneath worthless “refutations” by crackpots. The Xanadu system will include provisions for filtering systems whereby users may choose their own criteria for the kinds of cross-references to be brought to their attention. These devices would seem to overcome the possible problem of having charlatans clutter the system with nonsense. In the first place, one would have to pay a fee for each item published on the system. In the second place, most users would choose to filter out comments that others had adjudged valueless and comments by individuals with poor reputations. In other words, though anyone could publish at will on a hypertext system, if one develops a bad reputation, very few will ever see his work.

And this:

Miller and Drexler envision the evolution of what they call agoric open systems–extensive networks of computer resources interacting according to market signals. Within vast computational networks, the complexity of resource allocation problems would grow without limit. Not only would a price system be indispensible to the efficient allocation of resources within such networks, but it would also facilitate the discovery of new knowledge and the development of new resources. Such open systems, free of the encumbrances of central planners, would most likely evolve swiftly and in unexpected ways. Given secure property rights and price information to indicate profit opportunities, entrepreneurs could be expected to develop and market new software and information services quite rapidly.

Secure property rights are essential. Owners of computational resources, such as agents containing algorithms, need to be able to sell the services of their agents without having the algorithm itself be copyable. The challenge here is to develop secure operating systems. Suppose, for example, that a researcher at George Mason University wanted to purchase the use of a proprietary data set from Alpha Data Corporation and massage that data with proprietary algorithms marketed by Beta Statistical Services, on a superfast computer owned by Gamma Processing Services. The operating system needs to assure that Alpha cannot steal Beta’s algorithms, that Beta cannot steal Alpha’s data set, and that neither Gamma or the George Mason researcher can steal either. These firms would thus under-produce their services if they feared that their products could be easily copied by any who used them.

In their articles, Miller and Drexler propose a number of ways in which this problem might be overcome. In independent work, part of the problem apparently has already been overcome. Norm Hardy, senior scientist of Key Logic Corporation, whom we met at Xanadu, has developed an operating system caned KeyKOS which accomplishes what many suspected to be impossible: it assures by some technical means (itself an important patented invention) the integrity of computational resources in an open, interconnected system. To return to the above example, the system in effect would create a virtual black box in Gamma’s computer, in which Alpha’s data and Beta’s algorithms are combined. The box is inaccessible to anyone, and it self-destructs once the desired results have been forwarded to the George Mason researcher.

There is really quite a bit more at the link, noting that at the time Don had assembled a group of about ten people working on these ideas.  As for the hyperlinks, I recall thinking at the time something like: “People don’t value reading so much, so making reading better with hyperlinks won’t have a huge marginal value!”

Yes, in short.  Emilio Calvano, Giacomo Calzolari, Vincenzo Denicolò and Sergio Pastorello cover this topic in the latest issue of the American Economic Review:

Increasingly, algorithms are supplanting human decision-makers in pricing goods and services. To analyze the possible consequences, we study experimentally the behavior of algorithms powered by Artificial Intelligence (Q-learning) in a workhorse oligopoly model of repeated price competition. We find that the algorithms consistently learn to charge supracompetitive prices, without communicating with one another. The high prices are sustained by collusive strategies with a finite phase of punishment followed by a gradual return to cooperation. This finding is robust to asymmetries in cost or demand, changes in the number of players, and various forms of uncertainty.

Here is the paper.