Singularity Institute

Steps to be taken in order to realize this vision.

Step 1.  Set up.

Set up a webpage on the singinst.org domain advertising the grants; the page should also include background reading and open questions (so as to open up readers' thinking about possible papers they might write) and should contain a list of relevant conferences and conference submission deadlines.

Target date: have a website up by December 12

Step 2. Publicize.

Publicize the grants on the SIAI blog and allied blogs, mailing lists, scholarship and Call-For-Papers aggregation sites, and selected university grant/scholarship compendia. 

Target date: complete the blogs, mailing lists, aggregation sites, and initial compendia by December 26.  Continue publicizing thereafter if needed and as venues become available.

Step 3. Award the grants.

Review the first few grant applications that we get.  Fund good applications; refuse bad applications; and, for intermediate applications, dialog with the submitter to help the submitter figure out how to write a high-quality, high-impact conference paper on their chosen topic.  Seek and incorporate feedback from both successful and unsuccessful applicants.

Give preference to conference presentations that are more likely to reduce risks, but also to conference presentations that are less expensive (e.g., because the conference is in the applicant's home country).

Given the costs of conference fees and flights, the money should fund about two conference grants, with possibly some left-over money for a third partial grant.

Target date: due to the time scales on which academics need to decide whether to attend a given conference, we will send an initial reply to each recipient (a "yes", "no", or "not as it stands, but here are some changes to consider") within two weeks of the receipt of each application.  We'll continue reviewing applications until all the money has been awarded.

Step 4.  Assess the pilot grants program.

Record the impact of grants given in terms of quality of papers generated, the extent to which the papers are useful for improving our or the academic world's understanding of AI risks, and contacts made between the applicant and other AI researchers. This impact analysis should include any continued contact between us and the successful applicant after the conference. 

Target date: a preliminary report will be written a month after the final grant is awarded, but no later than three months after the first grant is awarded.  This report shall be updated slightly a year later, to include any delayed impacts of the conference (e.g., any follow-up papers written by an applicant).

Step 5.  Extend the program.

Publicize the impacts of this program on the SIAI blog or website.  If the pilot conference grants program has positive impacts, continue the conference grants program on a larger scale, either with money from the SIAI general fund or with money obtained by other donors who may be moved by the returns from the pilot program.

Target date: this publicization, and an initial attempt to seek funding in the event that the program has sufficient positive impact to be worth the money, shall be made at the same time as step 4.

Budget:  $2.5k, including:

$500 of staff time, to:

• Publicize the grants via SIAI and allied blogs, mailing lists, scholarship and Call-For-Papers aggregation sites, and selected university grant/scholarship compendia  

• Read grant applications, and either accept applicants, or reject applicants, or give applicants specific suggestions for improving their proposed paper.

$2000 to cover the costs of conference attendance (including conference fees, flights, accommodations, and meals) for applicants wishing to present papers bearing usefully on long-term AI risks, and unable to obtain funding from their universities. 

How this project, if funded, will reduce existential risk:

Benefit 1:  Increase the number of conference talks that bear usefully on AI risks.  By increasing the number of such talks: (a) probabilistically generate useful new ideas and perspectives on AI risks; (b) increase the academic credibility of AI risks, and of research on risks.

Benefit 2:  Indirectly increase the odds of journal articles that near usefully on AI risks (since many who submit a conference paper will later submit a journal paper on the same content).

Benefit 3:  Probabilistically increase the number of researchers who see this type of research as a part of their research interests.  (Sometimes people do something for the first time because there is an external occasion, but then continue in the same vein on their own.  Students, in particular, may find conference mini-grants to be a useful spur that causes them to draft an abstract.)

Benefit 4:  Test the efficacy of conference mini-grants; if the mini-grants have useful effects, we will have evidence of this usefulness which can be used in soliciting funding from other SIAI donors to continue the mini-grants program.

How benefits will be assessed:

An assessment of the grants' effects shall be written up afterward, discussing whether similar grants would be a good use of future funds.

    The assessment shall include:

• The actual costs in SIAI staff hours of running the program, and the extent to which the program managed to stick to the promised timeline;
• The number, positions, and affiliations of applicants;

• The talk titles, slides, abstracts, and/or papers of successful applicants, and the conferences at which successful applicants presented;

• Informal assessment of the value of the grant-backed papers by SIAI staff and at least two external academics, e.g. Future of Humanity Institute faculty;

• Informal assessment of the impact of writing and presenting the talk on the presenter and other conference attendees, based on interviewing the presenter after the conference: how many people attended the talk, what feedback did attendees provide, what follow-up contacts were made between the presenter and relevant attendees?  Also, how likely is the presenter to engage in further academic work around AI risks?  Did the conference grants make a difference?