Singularity Institute

Research Grants

Grant: Improving the Uncertain Future web application 

Key Aim: Make the Uncertain Future web application more user-friendly via improved text and tutorials.

 

Vision

1. The Uncertain Future application has informally received positive reviews from a number of academics and has generated interest in AI risk.

2. It is currently hard for users to understand what the application does, as indicated by the comments on Hacker News.  Better documentation and a better landing page could make the application easier for users to understand.

3. The more transparent we can make the applet, the more users will engage with it.  This can help users think through AI risks in a fair, reflective manner, instead of remaining attached to the specific storylines that often dominate conversations about the future.

Step 1: Collect feedback.

Determine what currently confuses users and what concepts, if conveyed, would help users understand the Uncertain Future web application.
Deliverable:  Document summarizing previously gathered user confusions and feedback.
Target date: Three days after start date[1].

Step 2: Plan improvements.   

Choose a set of improvements to the Uncertain Future landing page, FAQ, etc. that make it more engaging and more transparent to the target audience.  This could include video tutorials about how to use the software or about subjective Bayesian estimation.
Deliverable: Internal document detailing improvements to be made. 
Target date: One week after start date.

Step 3: Implement planned improvements.   

Deliverable: Improved Uncertain Future application.
Target date: Five weeks after start date.

Step 4: Publicize. 

Link the pages from the SIAI website and other relevant websites. 
Deliverable:  Publicity and dissemination.
Target date:  Six weeks after start date.

Step 5: Assess. 

Assess benefits, as detailed below; write a short written report on the cost-effectiveness of Uncertain Future work to date.

Deliverable:   A written document containing the actual budget, the impact on users, and updates to SIAI action heuristics.
Target date:  Twelve weeks after start date.  (The lag on this step is to allow user feedback to accumulate.)

[1] The "starting date" is the date (guaranteed to be within six months of the receipt of grant money) when we have skilled people to allocate to the project.  Extra donations increase our base of skilled people and thereby increase the number of projects we can get to; the lagged start date allows us to find new people, bring them here, and train them.

Total budget:  $4,800

Estimated time to completion: 2 person-months.

In total, this makes for a budget of $4,800.  This billing rate reflects what it costs SIAI to meet basic expenses (food, housing, often travel) for full-time researchers, writers, or interns willing to work to reduce existential risk without further pay.  This is not reflective of the money these Visiting Fellows or staff members could earn in the competitive labor market. Think of this as a matched donation. You donate the living expenses (which come to about $2,400 per researcher month); our researchers donate the surplus value of their labor.

How this grant, if funded, can be expected to reduce existential risk: 

Benefit 1: Attention to AI trajectories

A good, easily usable Uncertain Future webapp would, if well-publicized, increase public discussion of AI trajectories and AI safety. (The "AI safety" part comes from UF's recommended reading and links with SIAI.)

Benefit 2: Adding probabilities to AI futurism  

A good, easily usable Uncertain Future webapp would help change the type of conversation people have around AI scenarios. It would shift conversation from specific imagined futures, to the entire range of plausible futures and their likelihoods. Such a shift would help interested parties rationally allocate attention and other resources to the right risks

Benefit 3: Increased contact with professional forecasters and other relevant academics  

The Uncertain Future webapp is a good entry point for discussions with others who are interested in the future and who understand probabilistic reasoning. The alpha version already allowed SIAI research fellow Anna Salamon to give an invited talk at the Santa Fe Institute's Forecasting conference; a better version can allow further opportunities.

How these benefits can be assessed.

Roughly a year after project conclusion, we will write up an evaluation of the results of the Uncertain Future project, including the following:

• Cost: How many person-hours actually went into the project

• Impact on public perception:

  • How many Google search results does the Uncertain Future site have?

  • Is the Uncertain Future site referenced in any written conversations about technology forecasting or AI risk?  If so, what conversations, with what caliber of discourse?  Does it help people consider a range of plausible outcomes, rather than focusing on single scenarios? 

• Increased contact:

  • How many professional forecasters and other academics are we in regular conversation with who started talking to us because of the Uncertain Future webapp?

  • Has anyone gotten involved with AI risk reduction and cited the Uncertain Future site as one of the causes of their interest?