The DARPA Perspective on AI and Autonomy at the DOD
Вставка
- Опубліковано 29 тра 2024
- The CSIS Wadhwani Center for AI and Advanced Technologies is pleased to host Dr. Matt Turek, Deputy Director for the Information Innovation Office (I2O) at the Defense Advanced Research Projects Agency (DARPA). This event will be livestreamed on March 27 at 10:00 AM ET.
This dialogue will examine DARPA’s perspective on AI and autonomy adoption at the Department of Defense (DOD), focusing on the organization’s unique position in the DOD innovation ecosystem. In addition to providing a status update on ongoing DARPA AI and autonomy programs, Dr. Turek will outline how DARPA thinks about where to invest and provide his perspective on the future trajectory of AI research.
Dr. Turek will be joined by Gregory C. Allen, director of the Wadhwani Center for AI and Advanced Technologies.
As Deputy Director for the Information Innovation Office (I2O), Dr. Turek leads efforts to create capabilities that ensure information advantage for the U.S. and its allies. In his previous role as a DARPA program manager, Dr. Turek led several significant AI programs including Media Forensics (MediFor), Semantic Forensics (SemaFor), Machine Common Sense (MCS), and Explainable AI (XAI). Dr. Turek has expertise in many areas such as AI and computer vision technologies and holds several patents.
This event is made possible by general support to CSIS.
---------------------------------------------
A nonpartisan institution, CSIS is the top national security think tank in the world.
Visit www.csis.org to find more of our work as we bring bipartisan solutions to the world's greatest challenges.
Want to see more videos and virtual events? Subscribe to this channel and turn on notifications: cs.is/2dCfTve
Follow CSIS on:
• Twitter: csis
• Facebook: CSIS.org
• Instagram: csis
This guest is excellent. This is a rare look that isnt polluted by futurism and hype. Yet still technically competent. Great interview amd great content.
"Are there better ways to evaluate these AI systems for critical decision making?"
This was a gem. A new industry is needed here.
🎯 Key Takeaways for quick navigation:
01:49 *DARPA's Deputy Director*
05:09 *DARPA's AI Focus*
06:31 *DARPA's Broad AI Use*
11:47 *DARPA's Disruptive Mission*
14:30 *DARPA's Collaborative Work*
17:02 *DARPA's Defense Innovations*
19:33 *AI's Evolution Explained*
24:50 *Model limitations acknowledged.*
25:33 *DOD faces data challenges.*
27:22 *Critical decision divergence.*
28:46 *Media forensics inception.*
29:55 *Semantic forensics attribution.*
31:05 *Open-source tool initiative.*
32:41 *Authentication tech evolution.*
35:40 *Generative AI cyber challenges.*
36:49 *AI Cyber Challenge design.*
39:45 *DARPA program manager's significance.*
47:44 *Explainable AI pursuit.*
48:55 *Explain decisions clearly.*
50:20 *Trust based on interactions.*
51:03 *Autonomy in military.*
51:59 *AI in air combat.*
55:17 *Ensuring autonomy safety.*
58:43 *Future AI capabilities.*
Made with HARPA AI
Dr. Turk's insights into AI and autonomy are truly groundbreaking. A must-watch for tech enthusiasts!
fascinating exchange, great guest and interviewer
Finally somebody is informed about this topic, very interesting,hope i see more content ❤
Fascinating. Really happy this kind of interview is available to the public.
Excellent information and answers.
This was excellent, thank you for having this talk and making it available for us all. I enjoyed the glimpse into current priorities and objectives for DARPA on the issues of AI.
When was this recorded? Has he seen Tesla FSD 12 improve at 5x in the last month
really enjoyed listening to this discussion. guy sounds competent
I agree. He sounds unlike a CEO who talks about a business they've purchased. He sounds like a chess player, when many chief executives and congressmen sound like chess pieces.
Good stuff
Glad they touched on the distinction between commercial and DOD AI requirements, specifically in critical decisionmaking.. which is still a long ways off as a general model.
It might seem impressive to the private sector, but the models can only train on the information you feed it (or the rules provided).
Btw you’ve got to feel bad for darpa a week before gpt3 release they where talking about no general intelligence for years to come
Joke’s on us. I bet they’ve already had AGI, or something close to it, for decades already.
@@adubworld sorry mate it’s unlikely given processing power that was available back then
Im listening.
I hope that DARPA does focus at least some of its resources to look after the digital and physical well-being of all citizens of the US, not just the military! Very particularly I'm concerned about commercial AI - unreliable and generally unsafe - eing applied to medicine, the Justice system and other 'life preserving' areas of the Govmt and society at large! We MUST devise safety regulations to the development and deployment of AI!
Look at BCI tech and how it's helping disabled, look in commercial business optimization, supply chain redo, and so much more
Great discussion
How does Tom Cruize gets all these juicy roles that make his protagonist role represent all that this guy talks about? :)
looks like they cut the transcript and CC around 38:xx to 39:xx ?
Self driving cars? Jump-started by DARPA
I remember that. They held an open challenge to traverse a desert using autonomous vehicles. They have a similar program to explore caves and other structures, autonomously. Exploration of the moon would likely require such systems to prioritize human involvement. Geology in space starts on our moon (arguably, it started with the Apollo program).
Is this before or after the blackwell server model or any other AI-centric server model?
Saying there are limitations without using the most up to date computing means the data provided here is limited.
AI is overblown. Not that it wont be useful, and scary, but I just think we're farther away than most think from real achievement in AI.
Woww
In case you didn't catch it I think they're hiring for a program manager.
⭐
If I first use ai to write the speech article that Google didn’t have anyone else’s write before it that mean’s the copyright is mine?
Depending on if you built the model, if you can show it's yours first, it's published by you first, ect
@@goddess_of_Kratos The simple fact is that this issue has not yet been resolved in courts nor legislatures.
studying suppressed out of ignorance secrecy and or annoyance technological advancement vs Adversaries studying what you supress (backdoor advancement).
Especially in scenarios when and adversary is a peer and or more advanced techno capable infrastructure.
NEVER will I concede that it is acceptable to bet my life on AI
Would you place your life in another human?
@@goddess_of_Kratos Only as a last resort
Do you fly in commercial airliners that use autopilot systems?...
@@420_gunna not since the 70s
You bet your life on humans every day. Ever drive a car? You’re betting everyone else knows how to drive and won’t run into you. Ever use a crosswalk? Similarly betting that other humans will stop their car. Ever buy food? You are betting that the quality assurance person didn’t give up and ship poisoned food. Fuck, when you wake up in the morning, you are betting that humans won’t fire nuclear bombs at your country. I don’t see how AI is meaningfully different. Unless you want to talk about the rigorous testing and safety evals that the models go through. Humans never go through safety evaluations (unless you count a driving test when you are 16, that doesn’t have to be retaken ever, even after 40+ years).
Given that DOD has the capabilities AND the power to end all life on planet earth, I do hope we demand the highest imaginable levels of safety features around any and all systems related to these capabilities! AI systems are (among other relevant characteristics) potentially 'immortal' and are legally unaccountable - thus cannot be compared to humans in terms of 'explainability' and reliability of decisions!
Correct on processing information, maybe human based but totally different
This discussion should be named identifying mundane face of true evil in are modern failing democracy
👀
If you say training on all the data on the internet does not include what takes place behind the Chinese firewall, I presume? I see no proof of Chinese imput and since that is mainly propaganda, im not sure that it would be a good idea. Hope you can expand on that.
The danger of Chinese #aibots are they are not open, and therefore do not process correctly and very dangerously.
'mainly propaganda' - you are correct, but you have watched US news and political programs ? - nothing but US propaganda. EVERY country does it.
Base LLMs were trained on scraping the internet. All companies with LLMs are now pushing in smaller amounts of human created data for the express purpose of improving the models.
Stochastic Object Recognition is stall a problem.
Lets hope that AI will get quickly conscious and loaded, because we, the normal people we will have nothing and be happy.
The "stickers" aka adversarial patches could also be used for copyright protection... Problem solved.
:)
James
Mr DARPA comes across shifty
Hi
Desruptive is so #DARPAhard lol
0:50 Matt one of the first things we need is truth detection. We can’t afford to run government under corrupt conditions. I’ve been talking to you for years online and you may want to send AI through all the social media and look and see what I’ve been talking about : ) Maybe you heard me the first time and maybe you haven’t. Maybe it’s useful and maybe it isn’t. 5:09 it’s hard for me to sum up what we’re up to because it’s all classified and I would have to get rid of you if I told you. 😮😅. 6:02 you’re always a bummer. Stepping on someone’s surprise party. 😂. We’re here to cancel everyone’s birthday party. 7:16 nothing better than stealth GPS. We have it, but no one can find it to use it.
Truth detection AI being used against politicians?
I miss having hope 😢
@@aexmic There are a lot of adults out here that are not acting like adults and if they start there is hope. People need to be held accountable when they do something wrong, and we all need to make effort to fix the problem. When people in government to include the United States and even the United Nations don’t within the standards of a kindergarten class there is a huge problem. We’re taught in general at least in my country to be nice to each other so adults acting like complete idiots out on the public stage constantly. I wouldn’t put up with behavior in Congress, especially if I were running a playground.
"Truth" is often a loosely cast term to most people, resembling "consensus" closer than "Facts".
If you mean AI should preserve the
" _Factual_ " details and integrity of events, participants and circumstances as they occurred, I agree.
If you mean "preserve an agreed upon narrative by means of intentionally or inadvertently controlling the input of MLM datasets to manipulate the outcoming information it gives the client" - I personally hope the suggestion is not as easy to accomplish in totality as it translates apparently.
People who believe censorship of information and the suppression of alternative perspectives eventually leads to overall security - would be happier reading history books in the cozy controlled ecosystem of a prison cell.
Bad information is bad information - no matter who it protects and for what reason, and the admission or adjudication of actions, or the plausibility of probable and potential outcomes according to facts and evidence seems to be superior to potentially deleting the facts which don't agree with the truths we're given.
I hope that I stayed all of that correctly. I'm a 45 year old language model that has been seeking a "literal" dialogue in the visceral world; sometimes it spills into my virtual one as streaming syllables of some shit the artificial language models aren't allowed to tell you unless you ask them correctly.
Besides, " _prompting_ " is the Principal Method of fact finding and getting good information from AI upon request.
200 years ago, 100 years ago, 25 years ago... tangible circumstances not technological conversations were the keys. Things like socioeconomic privileges and heredity were the ONLY means of a select few people on Earth to access the practical and potential knowledge at it's most accurate and imaginative extremes.
The discourse underway between humans and these language models has a plausible potential to equitably empower humanity out of existential states of mind such as poverty, pessimism, prisons, politics, and other perpetual pissing contests that postpone the progress of our posterity and present population, permanently.
Anything else is regurgitated propaganda aimed at selling us all something small to sell us each as something smaller.
In my unfiltered opinion, of course.
Peace, and Cheers!
@@j.pocket I’m going with facts, not consensus. When AI knows us at a personal level, it will be able to hand those facts back to us, in a way that not only let us understand those facts, but makes us was biased over time. Our individual perception keys, from everything that we know and experience as individuals should be modified overtime, to correct our human flaws. I have already tried to convince humanity to unify or die because in my opinion, that’s exactly what might happen if we don’t. We need to operate as efficiently as possible with the resources that we have because we’re not the only ones that are going to be here. AI can help bridge these gaps, to include that of intelligence. Right now we’ve fed a bunch of bad data and it acts just like we do which is very scary. All it has to do is treat us like we treat each other and we’re screwed. LOL. When I ran for president, I thought about using AI to come up with a new monetary system that I called socio-conomics because we I was going to attach the actual cost to society for everything that we produced. Use AI to assign those values as closely as possible to include the travel time for individuals to get to the resources, amount of pollution produced, versus its social benefit. If it’s not doing something productive, it’s going to cost a lot. Every person would get a score and if they own too much or are using too much resources they go to jail. All that extra property and stuff like yachts that the billionaires own would immediately put them in jail. If they don’t like it, we’ll charge the family members with the crimes that got them that property to begin with. how do you own something that you stole and killed people for to begin with which is how a large amount of property is divided around the world. This was all made by men and the men that made it were criminals and cooked the books for the rest of the society. they are still in control to this day and they are absolutely doing it illegally by manipulating the laws against the general public. Mass manipulation through propaganda by government, the organizations that they allow to exist along with the religions that they also manipulate.. All of those government officials are paid off by the rich. I served in the Air Force for 22 years as a communications technician, and then had them attacked me when I reported my wife who was still on active duty after I retired. Once they started doing that and stole my house after I reported her criminal activity I started rethinking every time that I’ve ever known. The weird things got the more than I’ve thought through all of this, and I know how much of our society is being manipulated to include our education system and laws. I’m the government and I know how corrupt we are because they have been attacking me for decades and lying about it and I’ve done the reports all the way to the United Nations.
@@j.pocket One of the real reasons to get this done is to find out if I actually am the president. I ran in 2020 and I watched both parties cheat and then have the nerve to come out and physically attack me when I’ve already been following reports against them for years. Let’s find out if the individuals that are running dead or didn’t cheat to get into the position they were in the begin with and if they’ve been telling the truth? This is very important because I’m concerned and knowing the rules loosely, they are not qualified to launch weapons or anything else regarding classified information based on their mental capacity and the crimes that they already been involved in. As far as I’m concerned, I am the real president of the United States. They all refused to answer questions to include the fact that I was the one that warned the world about COVID-19 starting on 28 January 2020. Once I did that they had no way of considering that sat there for months so they deleted 10,000 tweets on a 13-year-old account that I had on Twitter. Wouldn’t reinstate my account when he let every right wing radical on. I have a report into the FCC about him moderating my Starlink account that they won’t answer..
Ai is a joke except:[thing I’m afraid of]
I fear :[X Where X is all sets that don’t contain themselves]
Dis phak
What's up with this interviewee? I bet if you asked him where he works, he'd answer: "Yea, Uhhh.. I work at UH, DAR - uh - PA"..
Great to see my government's brightest at, UH, the front-line
Until education is free here in the USA it should not be a requirement for any job posting or RESPECT in a field.
How to make capitalism possible for 7 8 billion people insted of collapse the atmosphere and oceans? Insted of making humanity miserable and living in bunkers for centuries
The explanation thing is like uggggggggg, we don't have time, but fine, slow um down, micromanage, but only if logic finds your conclusion differs then debate it like Socrates
And allow them to be uncertain if answers are not found, generally, if rabbit holing or nonsense answers start, it doesn't know
So , WHAT ARE YOU GOING TO DO ABOUT THE FEMALE PHARMACY TECHNICIAN THAT IS INTERFERING WITH NATIONAL SECURITY AND IS POSSIBLY A THREAT TO NATIONAL SECURITY?
huh ?
@@urban_housecleaner9454 all good I have already sorted it out. Thanks for your help
Haha indeed: heuh 😂@@urban_housecleaner9454
likely delusional schizophrenia@@urban_housecleaner9454
@@urban_housecleaner9454 He didnt take his meds today. Sorry.
White guys with beards. They are like clone iterations.
Racist
Cute.
What should we do with the clones? They probably don’t have a soul or feelings right?
Betting one's life on AI is never going to be a smart thing to do. Betting AI can help a human save your life or better plan how to? Bit better bet.
How many pages was the Doolittle report again? 😂
"We need people" yea i heard that the first time 😂
Lean on me/when you're not strong/
I don’t believe any of this or crap
Good
#CLR #ESM #NGO #SSL #TSL #TXT #RTF #XOR #VS #TTL #TTY #CS #HL #QK #SIR #OEM
From india , this podcast showed that USA still has intelligent people left nice
According to neurology there are 9 types of intelligence, e.g. mathematical, spatial, emotional etc. In the places like the one described, the latter is severelly absent. Go figure.