Senator Chuck Schumer and a bipartisan group of senators unveiled a plan to ensure US security in Artificial Intelligence. NPR tells us:
The working group's road map includes the following proposals:
— Increasing funding for AI innovation to "maintain global competitiveness"
— Ensuring enforcement of existing AI laws and to address any unintended bias
— Considering the impact AI will have on the U.S. workforce, including potential job displacement and demands to train workers
— Addressing issues related to deepfakes, particularly with regard to election content and "nonconsensual intimate images"
— Mitigating threats of "potential long-term risk scenarios"
“Mitigating threats of ‘potential long-term risk scenarios’ “ - that’s my favorite. What exactly does that mean? How would you know if you’ve mitigated a threat of a risk scenario? “Word salad” is my summary of that phrase.
The report's goal of $32 billion by 2026 “aligns” with a goal laid down by some august commission: I put the number in bold because that’s really all the senators care about: “how much money for my state?” or maybe “how many PAC’s will contribute based on that number?”
If you follow that link up, you come to a 786-page report. Now, honestly, does anyone really read this leaden bureaucratic sludge? Page 164-65, to take a random example:
The government will have to orchestrate policies to promote innovation; protect industries and sectors critical to national security; recruit and train talent; incentivize domestic research, development, and production across a range of technologies deemed essential for national security and economic prosperity; and marshal coalitions of allies and partners to support democratic norms. Some elements of a national strategy will need to be coordinated and replicated at the state level, through state-specific strategies to support Al research, commerce, and education. This will require a complex sequencing of promotion and protection actions to minimize costs and risks of punitive actions; ensure basic and applied research agendas are mutually reinforcing; coordinate approaches with international partners; and align executive priorities with legislative powers. It will require identifying technology trends and assessing the relative strengths of the United States and its competitors. It will require, above all, strong and consistent White House leadership.
I’m sure there are some people who eat up that sort of jargon (complex sequencing of promotion and protection actions! marshal coalitions of allies and partners to support democratic norms!). They’re probably the “hieratic persons (from Harvard or some such place)” [hieratic. I think “highly stylized or formal” is the definition Taleb means]
Nessim Nicholas Taleb, in Antifragile, tells us what’s going on. It’s Ivy League-educated people believe nothing worthwhile can ever happen unless they lead it:
Let us return to the metaphor of the birds. Think of the following event: A collection of hieratic persons (from Harvard or some such place) lecture birds on how to fly. Imagine bald males in their sixties, dressed in black robes, officiating in a form of English that is full of jargon, with equations here and there for good measure. The bird flies. Wonderful confirmation! They rush to the department of ornithology to write books, articles, and reports stating that the bird has obeyed them, an impeccable causal inference. The Harvard Department of Ornithology is now indispensable for bird flying. It will get government research funds for its contribution.
Mathematics → Ornithological navigation and wing-flapping technologies → ungrateful) birds fly
….
The problem is that what I wrote above looks ridiculous, but a change of domain makes it look reasonable. Clearly, we never think that it is thanks to ornithologists that birds learn to fly and if some people do hold such a belief, it would be hard for them to convince the birds. But why is it that when we anthropomorphize and replace “birds” with “men,” the idea that people learn to do things thanks to lectures becomes plausible? When it comes to human agency, matters suddenly become confusing to us.
So the illusion grows and grows, with government funding, tax dollars, swelling (and self-feeding) bureaucracies in Washington all devoted to helping birds fly better. Problems occur when people start cutting such funding—with a spate of accusations of killing birds by not helping them fly.
Clearly, without that $32 billion, AI will fall hopelessly behind; or to keep the metaphor going, those birds would fall out of the sky.
Who’s Writing This Stuff?
This “word salad” I called out is something that appeals to the Ivy League-educated classes, while actual engineer talk turns them off.
Taleb tells us the flaw, as usual:
Clearly, it is unrigorous to equate skills at doing with skills at talking. My experience of good practitioners is that they can be totally incomprehensible—they do not have to put much energy into turning their insights and internal coherence into elegant style and narratives. Entrepreneurs are selected to be just doers, not thinkers, and doers do, they don't talk, and it would be unfair, wrong, and downright insulting to measure them in the talk department. The same with artisans: the quality lies in their product, not their conversation, in fact they can easily have false beliefs that, as a side effect (inverse iatrogenics), lead them to make better products, so what? Bureaucrats, on the other hand, because of the lack of an objective metric of success and the absence of market forces [emphasis added], are selected on the "halo effects" of shallow looks and elegance. The side effect is to make them better at conversation. I am quite certain a dinner with a United Nations employee would cover more interesting subjects than one with some of Fat Tony's cousins or a computer entrepreneur obsessed with circuits.
What’s the Goal?
The beginning of wisdom is to look at the goals of a profession. How do they react to a societal problem, e.g. the supposed failure of the US to keep up in AI?
An engineer or businessman wants to solve the problem. That’s our nature.
A lawyer wants to make it one of his or her firm’s practice areas.
A politician wants to fundraise off of it, so as to keep himself or his party in power.
A civil servant wants to set up a permanent department with its own line item in the Federal budget.
In the second, third, and fourth cases, solving the problem would defeat the purpose. They want to colonize the problem.
A Senator like Charles Schumer considers his job done when he gives some speeches about the problem and gets on television. Passing legislation is just a bonus. What that legislation actually does in the real world is of no importance, except insofar as it might indicate a need for even more spending.
In a different era, Margaret Thatcher said,
When you want something said, get a man. When you want something done, get a woman.
Well, I think there are plenty of women in politics nowadays who consider “saying something” a completion of the job. Maybe we should change Mrs. Thatcher’s quote to
When you want something said, get a politician. When you want something done, get an engineer.
Engineers got AI to the amazing state it’s at now. Lawyers and politicians want to teach the engineers how to write code (or teach the birds to fly). It’s as simple as that.
Other Government Crash Programs
“Bipartisan group of Senators announce $32 billion program”: do you wonder if you’ve seen this movie before? You have.
Counter-intuitively, my review here doesn’t show that it never accomplishes anything.
Sematech
In the 80’s, Japan was the big threat. The Japanese were going to steal all our technology and destroy the industry that the US built. Wikipedia gives the story. Here’s another analysis of how it went, and whether it’s relevant to the more-recent CHIPS Act.
The latter analysis quotes only the best bits from the National Defense Magazine’s analysis (I think in mainstream “journalism” they call that “cherry-picking”):
"In 1987, the United States created SEMATECH, . . . a public-private partnership that was designed to direct research on semiconductor manufacturing between major industry players.
Today’s situation is reminiscent of the 1980s, when U.S.-based companies were losing ground to global competition due to major investments and subsidies by the Japanese government."
If we look at that actual report, it’s much more nuanced:
The importance of strong leadership from day one is the first lesson to learn from SEMATECH. In its early years, differences in corporate culture, business rivalries and secrecy over proprietary technology led to friction between employees assigned to the partnership. Additionally, member organizations were reluctant to lose top talent for the two years that was required when assigned there.
These challenges were not overcome until 1990, when Xerox CEO William Spencer took charge. Spencer identified specific people in industry who fit his vision and used his connections to convince CEOs of the benefits of participating in the organization.
A major benefit that emerged was the ability to create a place where researchers, manufacturers and end users all worked together. Any new efforts will face similar challenges that stem from company cultural and competitive differences.
A key weakness of SEMATECH was its annual fee of $1 million, which placed a high barrier to entry on all but the largest manufacturers. Small and medium-sized firms — an important part of the semiconductor supply chain and sources of innovation — were effectively excluded.
Unusually, the Federal government did not attempt to run the whole show. The investment by DARPA was only $500 million, and there were hefty dues for the member companies, who sent their own people to do the work.
Conclusion
I’m not as conversant with semiconductors as I am with software. I tried asking the one person I know who works at a major chip firm, and he just said that companies aren’t eager to share proprietary knowledge with their competitors, as that National Defense Magazine analysis said. Also that Sematech’s work does not seem to have affected what companies were actually doing. It did generate an 800+ page Ph.D. dissertation analyzing what a great “roadmap” Sematech produced, without any proof that anyone actually followed the roadmap.
It appears that Sematech accomplished some fairly modest goals, but “defeating the Japanese” wasn’t one of them. Japan did that to itself.
The Fifth Generation
Much close to Sen. Schumer’s AI effort is something that everyone involved would rather forget: the Japanese “Fifth Generation” project. This was Artificial Intelligence, 40 years ago.
“The Fifth Generation” is the title of a book from 1983 that I actually owned once, until I got rid of it:
The Same Movie, Once Before
A contemporaneous review of this book says:
Throughout all this, however, emerges a clear line of argument that leads steadily to the urgent message that the authors present as a tangible response to the Fifth Generation stimulus,. The message is that it is essential for America’s national interest that a large-scale concentrated project be mounted, similar to the Fifth Generation Project. I see the book as an unambiguous plea for rational planning and serious preparation for the “real computer revolution” which will come with the mass production of machine intelligence.
One review says on Amazon:
Even when Prof. Feigenbaum was awarded The Turing Award there had not been any mention of The Fifth Generation Project outlined in this book.
It is interesting to note, that Feigenbaum work on expert systems had quite a few local amazing successes. They were local. because they could be generalized or reproduced. His "Knowledge Engineering" methodology was faulty.
In 2017, Erik [sic] Schmidt pronounced that China began its AI National Project and phrased a wake-up call for the U.S. almost as a copy paste of Feigenbaum's call. Interesting indeed.
Software as Opposed to Hardware
Software is soft. That’s why they call it “software.”
What that means is that any college student in Romania or Kazakhstan or Sri Lanka can get a cheap computer, download some free software, and make a contribution, which might even turn out to be important.
A government program might have some success with hardware, because those students can’t as easily hack a semiconductor fab. But software moves much too fast for anything that Senator Schumer and his bipartisan coalition can come up with. They’re thinking, “Oh, great, AI takes billions of dollars worth of investment! Perfect for us!” But the ingenuity of those millions of programmers will render obsolete anything they can come up with.
Albert- Thanks so much for sharing this, particularly the reflection on AI's impact and the references on "birds" and "men." Definitely a piece that has brought more clarity to a new area (and era) of humanity. Hope you're well this week, Albert-
I read that AI of the 1960's was not envisioned in the same way of the 2020s. While there were AI researchers back then, I feel like several newcomers jumped ahead in line for research grants or investments simply by being in the right place at the right time. Or more effort was made at selling the idea than the actual product. If superconductors and cold-fusion were ubiquitous, perhaps this would be the best decade to develop AI. But because some regions still rely on certain types of power plants, there is not a harmonious enthusiasm for its utilization towards AI. I remain open minded though. 😉