The next existential catastrophe is likelier than not to wipe off all animal sentience from the planet

Not even the great dying got everything, so of the known natural (defined as coming from nature not technology) risks such as asteroid impact, climate change, etc. I don't give a better than 50% weight to them "wiping off all animal sentience". Nuclear weapons... we don't have enough to saturate the planet to the needed level; so they are also below 50%. That only leaves AI, and while I have a higher than 50% chance AI takes out all humanity, I suspect rather a lot of intelligent animals will get through it. At the risk of being the chimp that thinks it is safe from humans up in the tree because it's not smart enough to understand guns and helicopters, it seems that even for a paperclip optimizer, at nearly every point in time, it will be more efficient/optimal to go out into space and get resources to make more paperclips than to dive into more remote parts of the ocean or expand into more remote environments on land. There are many remote islands and tribes and animals that are in places where it just seems impractical to be looking for resources compared to the moon, asteroids, the orbital clutter, etc. At what point is the extremely small amount of resources on pitcarin island, for example, worth harvesting vs the cost? Return on Investment seems like something an optimizer would care about, and I think that would get the most remote locations significant time, especially as the optimizer's capabilities scaled up more and more making much larger resource deposits accessible. Eventually, maybe, it will get around to hunting down every last atom, but is that still the same event or another one? I am not sure. This makes me move my estimate below 50% for the extreme claim "all animal sentience". "most" or "over 98% of animal sentience" I would definitely be above 50% likely. "all" is a very extreme qualifier.

IIRC you see similar phenomena (although I can't recall any examples off hand) where some government-mandated software has massive security flaws but nobody does anything about it because the software is too entrenched.

Tyler Technologies.

But this is local government not federal.
 

I assume there are still gaps in the marketplace? Do you have a sense for what these gaps are? is there some large segment of the government which would use generative AI if only it was compliant with standard X?

Government Procurement itself. 

The process is incredibly burdensome and time consuming for the procurement officer. Templates exist but if you have never seen one before you would swear it was designed explicitly to prevent the successful procurement of anything. In truth, they are drowned in legalease in order to help them weather protests from unsuccessful bidders. But this makes is so that sometimes it takes 10 to 15 minutes of reading to even figure out what industry a bid opportunity is in and what the agency wants to buy. Then you have to go find out what the minimum qualifications are so you know if your company is even allowed to bid. 

Some poor procurement officer had to add all that gobbledygook to the template, at least one person has to approve it, and then it has to be added to at least one online procurement system. The Feds have many (though GSA is the largest grants.gov is another and several agencies have their own stand-alone procurement sites like Dept of Education and Health and Human Services). States have a ridiculous amount of them because the state level will have one or two, counties might use the state procurement system or have their own, and cities virtually always use aggregator sites rather than having their own. And that's before you get into the cooperative bidding tools/communities like NASPO, Managed Service Providers (MSPs), etc. which support groups of agencies or groups of states. Also, there are some 20-ish different types of bid requests that all have different rules and forms, and not every region uses the acronyms the same way.

And the biggest pain in the *** "statute X" is 2 CFR 200 period of performance. This is the infamous "use it or loose it" clause in federal procedures. The very definition of this statute in the regulations effectively says "even if we say you can use these funds over the next three years, we might change our mind and take anything that you have left back on September 30 of any given year.":

Period of performance means the time interval between the start and end date of a Federal award, which may include one or more budget periods. Identification of the period of performance in the Federal award consistent with § 200.211(b)(5) does not commit the Federal agency to fund the award beyond the currently approved budget period.

Source: eCFR :: 2 CFR Part 200 -- Uniform Administrative Requirements, Cost Principles, and Audit Requirements for Federal Awards

Therefore, a bit of extremely low hanging fruit for someone with the skills:

• An AI agent that can turn a bullet list of specs into a compliant procurement document aka RFP (Request for Proposals) or RFQ (Request for Quotes) or ITB (Invitation to Bid) or Tender or etc.
• Generate an appropriate list of potential evaluators from the agency address book such that they have: time on calendar to invest in the review, appropriate and active procurement authority, relevant background to understand the procurement
• Draft and route request for participation in the committee to potential reviewers, including (when necessary) permission from their supervisor if an evaluator is in a different silo'd team from the procuring team
• Repeat as needed until a full team is assembled
• Create a scoring rubric for the evaluation committee to use
• Review the prior bids for "same, kind, or like" procurements and generate an estimate of cost, as well as allowability and allocability of fund sources (not all government funds can be used for all activities, an activity may be allowable for an agency to perform but they may not have any fund sources to which the cost can be allocated because it is not allowable for those fund sources to pay for)
• Route the resulting bid request and funding proposal to the appropriate people for approvals via a digital approval process (vs some agencies still getting ink signoffs)
• Post the approved procurement document on the correct procurement board(s)
• Send courtesy notifications to vendors via that same tool that the bid is up
• During the Q&A period, field questions from potential bidders and generate sample answers for approval by the Procurement Officer, who can then just approve or edit before the AI posts the answer to the procurement board(s) as an addendum, either individually or in batches
• Confirm receipt of bids from bidders so they know if they actually used the board correctly and whether any critical document was missing from their bid
• On the closing date, archive all bids both complete and incomplete and notify the procurement officer, and create redacted versions ready to be sent out in response to FOIA requests
• Route complete and timely bids to the evaluation committee with the scoring rubric
• Route completed scores to the procurement officer for final decision
• Draft and Post award notice to procurement board(s)
• As needed draft protest responses for the procurement officer and route them for approval/editing before sending to the bidders
• As needed draft debriefing notes for the procurement officer to review with bidders who want to know why they lost and how they can do better next time
• As needed send redacted responses and draft letters in response to FOIA requests
• Archive complete procurement file in appropriate secure repository, and make FOIA-ready version of the whole thing

This is a reasonably comprehensive-but-generic procurement lifecycle (not counting the contract and implementation portions) that should work for any federal, state, or local entity.

Except for the interfacing with existing software, all of this is doable with ChatGPT prompts currently AFAIK. Though when trying I've had limited success getting it to understand the templates. AFAIK no agency is using AI to help in any step of this process in a meaningful way. I'd love to be proven wrong.

There's many other areas that could also provide huge gains for the government, but this one is so completely within the capacity of current AI that I'm wondering whether the government will start using AI to speed up the task or Microsoft will make it entirely possible within MS Word/Copilot first.