Proposal for an Electronic Journal on Artificial Intelligence Steven Minton Sterling Software AI Research Branch NASA Ames Research Center, Mail Stop 269-2 Moffett Field, CA 94035-1000 The development of personal computing has the potential to radically alter the way that scientific in- formation is disseminated. We propose to establish a new journal on Artificial Intelligence that will be distributed electronically, as well as being published in hardcopy. Electronic publication of scientific articles has many advantages over traditional publication. We are planning to distribute the electronic version of the journal for free, to allow authors to publish code/data along accompanying their articles, to greatly reduce the standard time to publication, and to make use of a variety of online capabilities, including online text searching and an online newsgroup for discussing articles. 1 Scope The journal will cover entire field of AI, rather than addressing a particular subfield. The journal will aim at publishing only articles of the highest quality. All members of the editorial board should be established members of the AI community who have been recognized for their research contributions. In addition to distributing the journal electronically, bound hardcopy volumes will be published on a annual or semi-annual basis. 2 Aims There are four primary goals that we hope to achieve by establishing this journal: -- Free (or very low cost) access to scientific articles for individuals. It appears that much of cost of producing traditional journals is due to the publishing medium and the way that publishers have traditionally operated. If this journal can be distributed to individuals for free, or at very low cost, and if it is successful, then it will have a significant impact on the field. It is still questionable whether it can be distributed for free on an ongoing basis, but this is our current goal. The free distribution of scientific articles would have obvious benefits both to the authors and the academic community. For example, are planning to eliminate the standard requirement that the publisher's permission be obtained prior to copying an article. -- Availability of programs/data accompanying articles. We intend to ``publish'' code and/or data along as appendices to articles, such that empirical experiments described in the article can be reproduced. This will not be a requirement for publication; it will merely enhance current practices for making code/data available. If authors do not wish to make use of this facility, or their system is large or inappropriate for us to publish, we will encourage them to include instructions as to how readers can obtain their code. -- Reduced time to publication. We hope to reduce the entire reviewing/publication time for the electronic version to as little as 3 months (in the best case, and with the authors help, of course). This is the standard practice in other fields, such as biology. Reducing the time to publication cannot be achieved simply by publishing articles electronically. It is also necessary to change the standard practices for soliciting reviewers. As an editor of the Machine Learning journal, my experience has led me to believe we can organize the reviewing and publication process in such a way that the 3-month turnaround time is possible (in those cases where the author is capable of carrying out his revisions in a timely manner). There are a number of key provisions for doing this. The review process will be managed by a group of ``action editors'' each of whom will be responsible for 5-10 papers a year. Editorial board members will serve for a fixed period of time (e.g., two years). Prior to joining the board, invitees will be informed of the expected reviewing load (e.g., 3-4 papers per year), and must agree to return their reviews within three weeks. Papers and reviews will be sent and returned electronically. However, before being emailed a paper, board members will be asked whether they have the time to review the paper, and if they do not, they will not be sent the paper. In general, papers will be reviewed by two members of the editorial board and a third reviewer who is not on the board, such as a recent Ph.D. Special attention will be made in selecting board members (as well as outside reviewers) so that we avoid reviewers who are chronically late or over-committed. -- Exploring the future of electronic publication. Electronic publication opens up some interesting new possibilities. We are specifically interested in exploring the following: 1. Capabilities for online text searching. 2. Allow authors to add online appendices updating their articles and pointing towards new work. 3. An online, moderated bulletin board where members of the community can comment on and discuss articles that are published. Authors would be especially welcome to join in the discussion. 3 Distribution The journal will be distributed both electronically and via a bound volume (a book). The bound volume will help establish the legitimacy of the journal. We are attempting to enlist an established publisher to distribute the bound volume. (See section on costs). The electronic journal will be organized into volumes and issues, as with a normal journal. As an issue becomes ``available'', it will be advertised online (via one or more bboards and newsgroups). Articles will be made available in postscript and at least one other format. ASCII text files will also be available (to facilitate online text searches). Although there are a number of technical issues that remain to be worked out as to which formats should be supported, we believe that in the next few years, appropriate standards will emerge. We are exploring two avenues for electronic publishing. The first, and simpler of the two, is to make the articles available via anonymous FTP and automated email from some publicly accessible machine. We plan to explore this avenue first. If the costs of operating the journal proves to be too expensive to rely purely on grants, our second option is to use an ``electronic publishing company''. (There is at least one being established on Internet). The company will store the online journal for us and charge an access fee. Unlike the first option, this will mean that ``subscribers'' have to pay a fee. I estimate that we need charge only a $20 subscription fee. (Note that this is an estimate.) Availability will be via FTP and email. Our current plan is to offer the journal for free initially, in order to determine whether this is practical for the long term. If it is not practical, we can switch to a subscription basis. (Readers will be advised of this possibility from the start, so they will not be disenchanted if we eventually need to charge for subscriptions). 4 Costs There are three basic costs: 1. Administrative costs. If we run a ``bare bones" organization, the administrative costs can be kept quite low. For the time being, NASA is willing to pay for part-time administrative help (as well for my time). I would prefer to have more substantial administrative help, of course. 2. Publishing costs. We are planning to ask authors to do all of the formatting, etc. that is required for publication. To facilitate this, we plan on giving authors latex skeletons (as does AAAI). Latex would not be required, of course, provided that a postscript version of the article is produced that satisfies the specified format. Diagrams remain slightly problematic at the present time, but should become less so in the future, as methods for producing diagrams and incorporating them into postscript documents become more standardized. In some cases, additional formatting help and copy-editing may be required for certain articles. If necessary, authors can be asked to pay for the expense of copy-editing. Another cost is getting an account on a machine on the net where the articles can be kept and where they can be made available by ftp. One possibility is that the account will be donated. (Faculty members at several universities, including CMU and the University of Washington have indicated that they would be interested in supporting this effort at their site.) A related issue is the cost of backing up the account. NASA has agreed to donate disk space for this. Another expects is the cost of producing the bound volume. We estimate that an initial 500 copies of a bound volume can be produced for under $5,000. We hope to defray the cost of the initial production run by a grant and to convince an established publisher to distribute and sell the bound volume. (At least one publisher has expressed interest in this). In addition to any advertising done by the publisher for the bound volume, we will advertise it electronically. Potentially, we can earn enough on the bound volume such that it will pay for itself, and thus future volumes can be printed without relying on outside funding. Obviously, if the electronic version of the journal is free, there will be a limited market for hardcopy. However, there should be at least a small market for the next few years, consisting of libraries, individuals for whom net-access is inconvenient and individuals who simply prefer to have a bound, printed volume. Eventually, we also hope to have a CD ROM version of the journal available for libraries. 3. Legal Costs. We plan to establish a non-profit corporation to run the journal, in part to handle questions regarding liability and copyrights. The initial costs to set up the corporation are small (some legal help will be donated). However, we will also require legal advice regarding electronic publishing of software. 5 Funding There are two possible sources of funding, subscriptions and grants. Subscriptions can be implemented by charging a yearly access fee (as mentioned above), but it would be greatly preferable if the journal were free. In this case, one or more grants are essential. We estimate that the minimal scenario described above would require approximately $40,000. (We expect NASA to contribute part of this). As outlined above, the majority of the expenses are startup costs. With luck, and low operating costs, the journal would then be self-sustaining through the sale of the bound volume (and later CD ROMS). Continued reliance on grants would be less preferable. 6 Acknowledgements This proposal has been drafted by Steven Minton with the help of Matt Ginsberg, Dan Weld, Oren Etzioni, Richard Korf, Paul Rosenbloom, and Bart Selman. Jaime Carbonell, Ken Forbus and Pat Langley have also indicated interest in contributing to the project and they helped generate the ideas described here. In conversations with many members of the AI community, this project has been almost universally supported. We are currently in the process of exploring the ideas in this proposal in more detail. We welcome comments and feedback. We thank those members of the AI community who have already contributed their suggestions. 3