A bipartisan working group led by Charles Schumer has introduced what this article calls a “long-awaited AI roadmap.” The document calls for at least $32 billion to be allocated for nondefense AI innovation.
Bill Gurley, a venture capitalist of long standing, says: In the entire history of the VC industry has there ever been a category LESS in need of incremental $$$$$.
Indeed. Corporations and individuals with money to invest are falling all over themselves to invest in things AI-related. Meanwhile, there are all kinds of serious issues–the hardening of the electrical grid against both enemy-caused EMP and natural magnetic storms, for example–that are not being adequately funded by the private sector and could benefit from some of that $32 billion. But they’re not as trendy at the moment.
Today’s WSJ includes an op-ed by Martin Casado and Katherine Boyle, both of Andreessen-Horowitz. They write about the Department of Homeland Security’s formation of an AI Safety and Security Board, whose purpose is to advise the department, the private sector and the public on “safe and secure development and deployment of AI in our nation’s critical infrastructure,” and they note that:
Of the 22 members on the board, none represent startups, or what we call “little tech.” Only two are private companies, and the smallest organization on the board hovers around $1 billion in value. The AI companies selected for the board either are among the world’s largest companies or have received significant funding from those companies, and all are public advocates for stronger regulations on AI models.
Much of the discussion of AI risks reminds me of the parable of Baptists and Bootleggers. And when regulation becomes a dominant competitive factor in an industry, it becomes very difficult for new players to survive and thrive unless they are exceptionally well politically-connected.
Your thoughts?
Looking back at the last half century, two trends are indisputable in the US.
(1) There has been an explosion of regulation across the board.
(2) The US has lost its former world-leading role in education, industry, health care and more.
Correlation or Causation? Any objective evaluation would point to excessive regulation as a leading contributing cause of the sorry decline of the US. So let’s have more regulation!
Of course, it does not matter what Our Betters do. What matters is what the Chinese, Indians, Russians. Brazilians choose to do — and we have little influence on that.
So, the naturally stupid want to pump $32 billion of taxpayer money into artificial intelligence and regulate it. How much would you be willing to bet that Schumer could give a reasonable explanation of how a toilet flushes? Now he’s going to regulate large language models.
So far, the only indisputable fact about this version of “artificial intelligence” is that it is fairly easy for whoever deploys it to manipulate or distort the “intelligence” to give whatever answers they want. Let them prove it’s worth something by finding paying customers.
Has one of these government initiatives/partnerships ever been anything but a total $h#t show, with the money siphoned off by various grifters.
I have a different take on the article. I don’t doubt that many of the things in the law are true and I also have concerns about the way government warps the private sector every time it tries to “help.” However my concern has to deal with what isn’t being said.
Much of the business model for social media companies is indistinguishable from intelligence agencies in that they are built around the accumulation of data and its translation into useful information. That combined with the ongoing digitization of our interaction with the environment becomes very worrisome, in fact it creates a situation that Erich Mielke of Stasi fame could only dream of.
A while back somebody pointed out to me that if you could put on special goggles, but instead of IR it saw the world in data; you would see the tremendous amount of data being thrown off by objects in the environment. It is the dream of social media companies to gather that data, organize it and turn it into useful information for monetary purposes. The first step of this, “information capture”, is facilitated not only by your cell phone but by Tim Berners-Lee’s “Internet of Things” which connects everything from your car to your door bell into a network so thatdata can be captured and analyzed.
We are rightly concerned that not only social media companies are censoring information at the behest of government, but the less remarked aspect of those stories is the mere presence of “former” US intelligence officers from DHS and FBI in their corporate structures. I have noticed few people have taken to the logical conclusion the recent controversies regarding TikTok scraping data and sending it China. Once again social media companies and intelligence agencies are very much alike
One of the most obvious uses for AI is information analysis, plug in the right data sets to train the system and allow it see patterns in near real-time that would take a building of analysts a week to figure out. I know the spending bill mentions “nondefense” but much of what is here is dual-use as the most probable defense use is not in operating weapons but in ISR. A good example would be the use of AI to analyze the data fed by Gorgon Stare systems which provide persistent aerial surveillance over 100 sq. km area. Train AI on enough data sets and it would be to recognize in near real-time much of what Gorgon Stare is gathering.
All of that legislation and spending provides a lot of cover for Deep State collaboration; most of what’s in the bill is probably just genuine pork and graft, but the concept of “10% for the big guy” works also to cover up black programs. There are already existing models of secretive collaboration between Silicon Valley and government, mutual interest in developing dual-use technology, and a past history of government funded black programs.
Paranoid? Well there’s always judging by capability and not intent and also all that recent history. There is also strong pressure to further digitize other aspects of society such as digital currencies and self-driving cars; all that’s missing is a way to store and analyze the data generated. That’s where AI comes in. The Stasi was the most efficient secret police force on the planet and was said that 1 in 8 East Germans worked for it. With AI that would make Stasi look prehistoric and allow the creation of a social credit system beyond the dreams of avarice.
Edward Snowden warned us of this. More than once. But what have the Rs done to defang this IC Monster? Nothing except keep it going when sensible and fear respecting people would’ve killed it.
Artificial Intelligence gives me the Heebie Jeebies, with it’s potental for misuse.
well the apparats, much like their analog in Russia, China, et al, went private, there was a mass exodus into the tech companies, the word is sysgy, which is a merger of two disparate things, in this case public and private power,
in the 90s in college around the time of the NAFTA debate, I was firmly on the side of free trade, taking the argument against Perot I didn’t consider why it took hundreds of pages to create a free trade compact,
There’s another dimension (besides ISR analysis) to this proposed $32 billion slush fund of a public-private partnership
AI not only needs a lot of energy for its training and inference tasks, but more importantly it needs reliable energy. That is of course the very energy that will become rarer as government green energy mandates further destabilize the grid. So we will have a one business client of government (AI) in conflict with both another commercial (green energy firms) client and ideological ally (environmentalists)
Some people like Richard Fernandez think that conflict will lead to one side or the other getting the chop, betrayed like that meme where the guy’s attention is drawn from his current gf to the hot new thing walking down the street. In a pure ideological sense perhaps, but only if you view things as a Trotskyite would.
In reality partnerships like this and the various green energy legislation create conflicts which aren’t bugs but rather features of the system because they create opportunities for government and various swamp creatures to broker compromises among the various competing factions, in this case AI/Silicon Valley vs. Green Energy/environmentalists.
I can easily imagine within a year or two of this roadmap going into effect, various parts of the AI industry joining not so much for the cash but because they need access to reliable energy and this “partnership” will be the easiest way to get it. Much like the massive data farms of northern Virginia got their start because that’s where the fiber networks were, we’ll see AI facilities cluster where the feds will ensure reliable energy will be made available.
For the various companies in AI the message will be clear. Join up and be a chum… things will go okay for you. Be ideologically pure, keep government at an arm’s length and see how you like running your data infrastructure on wind and solar.
Mike…It’s happening. Amazon Web Services is building out a data center adjacent to a nuclear plant in Pennsylvania. Equinix is doing a power-supply deal with the nuclear startup backed by OpenAI founder Sam Altman. Microsoft has committed to buy 10.5 GW of ‘renewable’ capacity to be developed by Brookfield…given that Brookfield is one of the owners of the Westinghouse nuclear business, I’m pretty sure that much of this 10.5 GW will be nuclear.
https://electrek.co/2024/03/05/amazon-just-bought-a-100-nuclear-powered-data-center/‘
https://www.bisnow.com/national/news/data-center/data-center-reit-equinix-inks-power-deal-with-sam-altman-backed-nuclear-startup-oklo-123688
https://www.theverge.com/2024/5/2/24147153/microsoft-ai-data-center-record-renewable-energy-purchase
David, I heard about the Equinix/Oklo deal in the datacenter press. It’s good to hear that people like Altman and at Amazon are thinking ahead on these issues.
I guess I’ll believe it when I see it… meaning a fully operational, fueled, and permitted reactor supplying juice. The best guess on a planning-to-operational cycle is upwards of 10 years, which is going to be too long. Then of course there are the flying monkey NGO squads fielded by environmentalists, red tape, and the opposition of the green energy firms. I’m sure Chuck Schumer and his slush funded connections might be able to help for a small consideration
I hope Altman and the rest can speed this thing up but when I’m asked to bet between heroics and the red tape/flying monkey side, bitter experience has led me to pick the latter.
Good luck getting the first spade of dirt turned on a conventional plant in five years. I don’t think anyone’s tried on a nuclear plant in 30-40 years. I’d be very surprised if there is 10+ GW wandering around somewhere, not already spoken for. It’s nice to imagine returning to a mind set like the 50’s-60’s where people understood that having electricity meant power plants and power lines but I don’t see that happening soon.
When you need the output of a couple power plants to mimic (badly) a human intelligence that will run for hours on a burrito and a milkshake, you’re probably barking up the wrong tree.
Here’s a take on the inscrutability of AI, or more properly, LLM’s.
https://www.nature.com/articles/d41586-024-01314-y
Related:
https://x.com/josephflaherty/status/1793262203421811054
If there is one positive aspect of the California AI Accountability Act that Flaherty links to is that penalties for noncompliance are out in the open.
An underappreciated aspect of the soft censorship regime surrounding COVID and the 2020 Election was the background work performed in the years prior. During various Congressional hearings, politicians threatened the social media companies with legislation if they didn’t self-regulate. It didn’t take a genius to predict what happened next; the various companies realized their political vulnerability and began to both preemptively censor and be receptive to messaging from DC. That’s what “smart” companies who are vulnerable to regulation do; your government relations shop isn’t just about lobbying government, it’s a two-way street that transmits “messaging” from government back to you
One problem in Schumer’s roadmap is stated at the end of David’s post, that if you want to survive in the industry you have to join what amounts to cartel. The second problem is that regulation becomes part of a soft corrective process where issues are headed off within the cartel and before they make headlines and we in the public become aware of what’s going on. I can imagine a year or two from now this back channel communication being used to alter data sets and algorithms that our forward thinking elites find problematic. I doubt there will be an AI version of the Twitter files so we deplorables will never find out.
When politicians like Elizabeth Warren go off regarding social media platforms or Big Snack, this isn’t (just) about the ravings of some crazy loon. It’s also a coded message to comply that becomes part of the targeted companies decision-matrix which then becomes open to less overt influencing operations. Tyranny is more than just the mailed fist, it;’s about creating uncertainty so that people and companies self-regulate
Richard Hanania has a good description of this regarding civil rights policy in his book “Origins of Woke” but the model can be applied elsewhere