#PCAST has released its report containing recommendations on Patient Safety. A surprisingly large fraction of hospitalized patients in the US experience adverse events from misdiagnosis, miscommunication, and other preventable medical errors; despite widespread awareness of this issue, error rates have not improved significantly in recent decades. We propose the formation of a national patient safety effort involving such initiatives as improved data collection and monitoring, sharing of best practices, partnership with broad communities of stakeholders, and research into emerging technologies (including AI assistance) to systematically reduce or mitigate errors. https://www.whitehouse.gov/pcast/briefing-room/2023/09/07/pcast-releases-report-on-transforming-patient-safety/

#PCAST has released its (congressionally mandated) review of the National Nanotechnology Initiative, which was formed in 2003 to develop the nanotechnologies now used in applications such as microelectronics and mRNA vaccines. Our main recommendation is that, given that the technology is now mature and can now be primarily developed by the private sector, that the Act that authorized the initiative be revised or sunsetted to remove most of the extensive oversight and other federal infrastructure for the intiative, leaving behind a smaller subcommittee to coordinate strategic planning and research and to support some modest training programs. https://www.whitehouse.gov/pcast/briefing-room/2023/08/18/pcast-releases-report-outlining-seventh-assessment-of-the-national-nanotechnology-initiative/

#PCAST is seeking public input on nutrition research, as part of a new working group on using nutrition science to combat diet-related diseases. The key questions being focused on are

1. How can the United States obtain the greatest return from federal investment in nutrition research?

2. How could/should research-based interventions for primary and secondary prevention of diet-related chronic diseases be introduced into federal programs?

3. What can be done to assure equitable access to the benefits of the federal nutrition research investment?

Further information on the public call can be found at

https://www.whitehouse.gov/pcast/briefing-room/2023/07/18/pcast-welcomes-public-input-on-nutrition-research/

Inquiry Into Inquiry • On Initiative 4
• https://inquiryintoinquiry.com/2023/05/15/inquiry-into-inquiry-on-initiative-4/

Re: Terry Tao • PCAST Working Group on Generative AI Invites Public Input
• https://terrytao.wordpress.com/2023/05/13/pcast-working-group-on-generative-ai-invites-public-input/

My Comment —
• https://terrytao.wordpress.com/2023/05/13/pcast-working-group-on-generative-ai-invites-public-input/comment-page-1/#comment-678057

I think a lot of people who've been working all along on AI, intelligent systems, and computational extensions of human capacities in general are a little distressed to see the field cornered and re‑branded in the short‑sighted, market‑driven way we currently see.

The more fundamental problem I see here is the failure to grasp the nature of the task at hand, and this I attribute not to a program but to its developers.

Journalism, Research, and Scholarship are not matters of generating probable responses to prompts or other stimuli. What matters is producing evidentiary and logical supports for statements. That is the task requirement the developers of recent LLM‑Bots are failing to grasp.

There is nothing new about that failure. There is a long history of attempts to account for intelligence and indeed the workings of scientific inquiry based on the principles of associationism, behaviorism, connectionism, and theories of that order. But the relationship of empirical evidence, logical inference, and scientific information is more complex and intricate than is dreamt of in those reductive philosophies.

#Peirce #Logic #Inquiry #InquiryDrivenSystems #InquiryIntoInquiry
#TerryTao #PCAST #AI #IntelligentSystems #LLM #LargeLanguageModels

Inquiry Into Inquiry • On Initiative 4
• https://inquiryintoinquiry.com/2023/05/15/inquiry-into-inquiry-on-initiative-4/

Re: Terry Tao • PCAST Working Group on Generative AI Invites Public Input
• https://terrytao.wordpress.com/2023/05/13/pcast-working-group-on-generative-ai-invites-public-input/

My Comment —
• https://terrytao.wordpress.com/2023/05/13/pcast-working-group-on-generative-ai-invites-public-input/comment-page-1/#comment-678057

I think a lot of people who've been working all along on AI, intelligent systems, and computational extensions of human capacities in general are a little distressed to see the field cornered and re‑branded in the short‑sighted, market‑driven way we currently see.

The more fundamental problem I see here is the failure to grasp the nature of the task at hand, and this I attribute not to a program but to its developers.

Journalism, Research, and Scholarship are not matters of generating probable responses to prompts or other stimuli. What matters is producing evidentiary and logical supports for statements. That is the task requirement the developers of recent LLM‑Bots are failing to grasp.

There is nothing new about that failure. There is a long history of attempts to account for intelligence and indeed the workings of scientific inquiry based on the principles of associationism, behaviorism, connectionism, and theories of that order. But the relationship of empirical evidence, logical inference, and scientific information is more complex and intricate than is dreamt of in those reductive philosophies.

#Peirce #Logic #Inquiry #InquiryDrivenSystems #InquiryIntoInquiry
#TerryTao #PCAST #AI #IntelligentSystems #LLM #LargeLanguageModels

#PCAST has released another report on supporting the US public health workforce: https://www.whitehouse.gov/pcast/briefing-room/2023/05/01/pcast-releases-report-on-supporting-americas-public-health-workforce/ (see also the video presentation by the working group cochairs at https://youtu.be/LqpvVWLP29s?t=3566 ). (I was not a member of this working group, but served as a referee for an early version of the report.)

The working group identified structural weaknesses in the US public health system as one of the reasons why US health outcomes such as life expectancy lag behind peer nations despite the highest per capita spending on health care in the world. The report contains recommendations to recruit, train, and retain public health workers and develop career pathways for community health workers.

One of the surprising findings was that there was not even a standardized lexicon for the public health sector, so that basic information such as the number of health care professionals working in this sector were not reliably available. So the first recommendation is in fact to develop such a lexicon in order to accurately count and measure this workforce, and identify critical workforce needs.

1/ The #PCAST report to the #President on #extreme #weather #risk in a changing #climate is fine, but a missed opportunity to move the needle.

Let's start with the recommendations.

#PCAST has released its report to the President on enhancing prediction and protecting communities against extreme weather risk, in an era where historical data on extreme weather events are no longer reliable predictors of future risk. https://www.whitehouse.gov/wp-content/uploads/2023/04/PCAST_Extreme-Weather-Report_April2023.pdf

An executive summary is available at https://www.whitehouse.gov/wp-content/uploads/2023/04/PCAST_Extreme-Weather-Report_April2023_Letter-ExecSumm.pdf

Our main recommendations are: 1. To task the US science agencies (NOAA, NASA, DOE, and NSF) to generate high resolution climate models covering the next few decades, and make them available on a public portal; 2. Make public (after suitable anonymisation) relevant federal data to develop weather-hazard and hazard-loss models, as well as to develop standards to score such models; and 3. To develop a national adaptation plan to mitigate harms from climate risks to vulnerable communities.

I was personally happy to serve on this PCAST working group, which was quite an educational experience for me (in particular, to see how all the theory behind solving fluid equations is actually used in practice!).

Was part of a productive meeting on the risks and benefits of #AI between the US President and #PCAST yesterday. I've recently been tasked to co-chair a PCAST working group on generative AI policy recommendations and hope to have some activities and reports on this to announce in the coming months. We will likely build upon existing efforts in this space, such as the #WhiteHouse blueprint for an AI bill of rights at https://www.whitehouse.gov/ostp/ai-bill-of-rights/ , and the AI Risk Management Framework developed by the National Institute for Standards and Technology #NIST.

#PCAST is initiating a working group on Cyber-Physical resilience and is soliciting thoughts and actionable proposals in topics such as the following:

* Recovery and survivability in the face of attacks and events.
* Approaches to assure continuity of operations in degraded states.
* Mechanisms to measure and assess modularity and limitations of scope or costliness of failures.
* Incentives to balance efficiency which can reduce resilience vs. the investment needed to maintain sufficient resilience.
* Out-of-band or systems-independent means of assuring physical control in the event of digital failures.
* Methodologies and standards to encourage resilient systems design and adoption.

Further information about this group and directions on how to submit input can be found at https://www.whitehouse.gov/pcast/briefing-room/2023/03/15/pcast-initiating-working-group-on-cyber-physical-resilience/

I forgot to mention this when it first came out last month, but #PCAST has released its third report, this time on recommendations to modernize the national #wildfire response system, with a combination of administrative and technological proposals. For instance, we recommend using modern AI modeling techniques, combined with historical satellite data gathered from civilian and declassified military sources, to significantly improve the existing capability to predict wildfire propagation, both for real-time firefighting efforts and in planning the deployment of resources and prescribed burns. There is also some promising future technology in autonomous or semi-autonomous firefighting whose research and development we recommend supporting.

The executive summary of the report can be found at https://www.whitehouse.gov/wp-content/uploads/2023/02/PCAST_Wildfires-Report_Feb2023_Letter-ExecSumm.pdf and the full report can be found at https://www.whitehouse.gov/wp-content/uploads/2023/02/PCAST_Wildfires-Report_Feb2023.pdf

@tao (2/2)

Scenario 2: An #AI assisted cyber attack happens, society adapts, but the response is inadequate. For example, perhaps the cyber attack makes powerful AI taboo, but then secret labs try to create powerful AI away from the view of the public. You are then back in a similar situation as if the cyber attack didn't happen, just with a delay and a smaller set of actors.

(A possibility for a response that *is* adequate is the creation of safe AI systems that are designed to defeat the malicious AI systems. Or it might be inadequate if in the process of creating such a system, we run into scenario one first.)

This scenario is rather pessimistic though. Any solution we could propose by definition comes from human society, since we are humans in that society.

---

Both of these scenarios are just speculation though. A cyber attack makes a lot of sense in light of the #PCAST meeting you shared.

@theking I think that the more substantial and nearer-term threat will be from numerous lower grade AI-assisted malware attacks on various components of our modern world, rather than from some theoretical grand attack on our civilization as a whole. (And the experience we will gain in defending against the former will help inoculate us against the latter.)

EDIT: See also this #PCAST meeting on #Cyberresilience from last year: https://youtu.be/QkfXUNwyOxw?t=8666 . (A formal PCAST report on this topic should be finalized before the end of this year.)

Marv Adams (who used to serve on #PCAST with me, before he was called to serve at the National Nuclear Security Administration) explains the recent Q>1 #fusion breakthrough at the National Ignition Facility. (Still many steps needed to get to a viable fusion reactor, but one now has proof of concept that the energy produced can exceed the energy directly introduced.) https://www.youtube.com/watch?v=w-5bNFg50KU

The presidential council of advisors on science and technology (#PCAST) that I serve on has released its second report of this term, on recommendations to support the biomanufacturing industry (see https://www.whitehouse.gov/pcast/briefing-room/2022/12/08/pcast-releases-report-on-strengthening-biomanufacturing-to-advance-the-bioeconomy/ ). [Our first report was on recommendations for the semiconductor industry, released at about the same time as the passage of the CHiPS act.]

Our very own @Keasling_Lab@twitter.com contributed to this #PCAST report on strengthening biomanufacturing! @jbei@twitter.com @LBNLBioSci@twitter.com

RT @francesarnold@twitter.com

Newest #PCAST report on supporting our growing #bioeconomy. #Biomanufacturing is integral to solutions for many of our national and global challenges...
https://www.whitehouse.gov/pcast/briefing-room/2022/12/08/pcast-releases-report-on-strengthening-biomanufacturing-to-advance-the-bioeconomy/

🐦🔗: https://twitter.com/francesarnold/status/1601244625960046593

Dr Eric Horvitz, Chief Science Officer at Microsoft and a member of President Biden’s council of advisors on science and technology, gave the University of Michigan Tanner Lecture on Artifical Intelligence and human values last week. It’s well worth your time if you’re looking for an accessible and broad overview of where things stand with AI, opportunities as well as risks. https://m.youtube.com/watch?v=gK-xNridrws #AI #ArtificialIntelligence #PCAST #Science #TannerLecture #UofMichigan