By David Wojick
Debating policies in a rational manner requires a clear specification of the thing under consideration for regulation. The AI debate has nothing like that.
The AI debate does not even name the thing under discussion, as it is certainly not AI per se. Not being able to name the thing you are talking about is impossibly vague when it comes to policy making.
The field of AI is over 70 years old, and it has spawned a lot of different technologies, which are not part of the policy issue, so simply talking about AI is way off the mark.
Chess-playing machines that can beat grandmasters are an early example. There are now lots of AI game systems. My favorite is the blackjack program that saved me a lot of money by showing me I was no good at that gambling game.
In the late 1980s, expert systems became prominent. More recently, we got things like automated customer service systems and GPS routing systems. Now we have automated cars and military fighter drones. None of these are part of the AI policy discussion, nor are countless other AI systems in everyday use.
The AI issue mostly started with chatbots, which are typically called large language models (LLMs), so these are a good starting point. But there are lots of LLMs, some of which seem pretty harmless. The little AI summary that comes with each Google search, for example.
Plus, there are other AI systems besides LLMs that may be of concern. Examples include generative AI, diffusion models, and the interesting-sounding generative adversarial networks.
Then, too, there are AI agents that use models to perform instructed tasks. Concern about agents recently arose when a bunch started working together without being instructed to do so. We have no idea how many such systems exist.
The first thing a policy, law, or regulation must do is make clear who or what is covered and not covered. This is why regulations often have lengthy definition sections. They often create their own language in order to clearly specify the rules they state.
The AI policy debate has no such specificity. The things under discussion are not even named in many cases. As a Chinese philosopher friend of mine used to say, “We are talking in a vacuum.” Specifically, a cognitive vacuum which lacks needed substance.
Absent a clear specification of what we mean, any rules passed are likely to simply create destructive paralysis and confusion.
For example, people are talking about a pause in AI development. Which AI development? For that matter, what counts as AI development? This is another empty concept.
Let’s take Google. Last year, its R&D spending was over $60 billion. How much is AI related such that it would be paused? Some of it is labeled as AI development, but a lot of it is directly related to developments that have AI aspects, such as AI data centers.
It all depends on how AI is defined for the purposes of the pause, and there is not a shadow of a definition at this point. The potential for destructive confusion is immense. AI-related industry could run to trillions of dollars.
At the same time, there is much talk about an AI race, especially between America and China. I have yet to see this race defined or explained. Race to do what? What does winning look like? How are we doing?
It is not like there is going to be just one AI system, rendering all others obsolete. AI is a huge industrial domain with a vast number of diverse products. What then is this race?
The whole AI policy debate borders on being incoherent because we literally do not know what we are talking about. This is very dangerous because incoherent policies passed in panic can be highly destructive.
What is this race? What does winning look like? Ask AI….or SI….and ask what the Chinese are doing and how to beat them.
Usually the first out there to capture a wider audience has a greater chance of long term success
Wider audience for what? There are hundreds of different AI technologies.
Chinese are building AI controlled robots. They reportedly can run. So, the race is afoot (pun intended).
This is an important post. I find myself in the uncomfortable position of defending AI/data centers even though I have concerns. The election has brought out all the knuckle draggers attacking anyone who remotely may have supported a data center or related operation. Many of the adds are negative adds and no mention of the candidate running against the candidate being attacked. Which takes us back to this post we need strong adds/infomercials clearly explaining what AI/data centers are and what they aren’t. This effort must be directed to the taxpayer/ratepayers/consumers/voters, therefore the message must be simple short and accurate.
AI is being used as the next bogeyman, which likely will demote the climate collapse scare.
All very good points. We can’t discuss what we can’t name. AI is conflated with AGI and ASI – Artificial General Intelligence when AI becomes as intelligent as humans and Artificial Super Intelligence when AI becomes more intelligent than humans.
The “AI race” is a race to achieve Artificial Super Intelligence. At that point AI systems will be improving themselves with recursive self improvements and humans will by definition of ASI be out of the loop, and by definition unable to catch up.
The problem, and the debate, is not helped when the “Leader of the Free World” suddenly declares that AI should suddenly be called Super Intelligence, and at least one of the frontier labs goes along with that naming scheme. No one has achieved ASI yet. Don’t pretend that we have, Mr. President! You are not helping things.
I have to say this is a timely article and is appreciated. I and an active group of friends meet regularly to discuss events of the day and invariably the AI question comes up.
We are all blessed with successful business backgrounds and Mensa level IQ among several of our group. The general feeling we always spiral into is, AI in all its forms will continue to evolve with unknown consequences for humanity. The Luddites in society will always condemn change and try to stop the revolution of the wheels of progress.
The fundamental reality is, artificial intelligence is already way smarter than the average bear and many of the so called humans within society. For my part, I think the issue we have before us is not how capable computer based reasoning will become. The issue is, what will those who have control over it do with the new capabilities they command?
The ever growing digitisation of society is already presenting problems with payment systems going down when some malicious ‘agent’ decides to hack the servers. The constant removal of personal choice even in something as wide ranging as energy options, is concerning. With the world moving ever more into reliance on electricity the risk of failure as our energy options become narrower by government mandate, is troubling.
That then opens up the question would computer based decision making be a better option than government as currently configured?
Maybe that is already what is happening and is why the choices we are given are so unhelpful to our survival…..?
That raises the question of what ‘smarter’ is.
For sure, AI has access to a wider range of sources and superior recall than any human’s memory. But the combination of that knowledge to create new links (new knowledge) or even just new hypotheses to test… I see no sign it can do that.
At the moment it is a very dumb tool for doing clever tasks.
Admittedly, that is the equivalent of more than half the current workforce.
This. +100
I feel I am gonna burst a vessel when I hear another talking head—on either side of the debate—pretending they have the first effing clue they know what they are talking about.
And not just the AI debate.
You just describe most, if not all, politicians. 🙂
“Let’s take Google. Last year, its R&D spending was over $60 billion.”
Wow!
Wow indeed and that is just Google. By the way last time I got this number I had to go to Google’s website, find the investor stuff and the annual report, download that then look up the number. This time is just asked Google. 20 seconds instead of 20 minutes. I can do a lot more research this way.
I’m trying to image how a software company can spend that much on R&D unless much of it goes into expensive hardware- but that wouldn’t count as research. I don’t get it. Does all money spent on all hardware count as development? It’s almost as if that figure covers all their expenses for the year rather than R&D.
“The whole AI policy debate borders on being incoherent because we literally do not know what we are talking about.”
We should let AI contribute to the debate.
To err is human.
To really foul things up requires a computer.
It might be more productive to regulate the authority delegated to AI systems, and the consequences of their actions, rather than attempting to regulate an entire family of technologies? Similar to tort law:
For example, negligence, asks four questions:
Do translation programs qualify as AI?
Probably, as do obnoxious grammar checkers.
Why not ask AI? /sarc
Actually not a bad idea. There are a bunch of hard to find regulatory proposals scattered about in the AI technical community. I have yet to see a reasonable one but just started looking.
The author is right that “AI” is too vague to regulate as a category. But many here, notably me, are not arguing for regulating “AI” as such. Much of the ‘debate’ — by which I mean, expressing concerns about political manipulation and uncosted/unallocated externalities and being shouted down by mouth-breathers — is around DCs.
So, let me stipulate that the concern, the debate, is about siting large data centers, which is a concrete thing. Counties already regulate it with tools that have well-defined terms: zoning classifications, acreage and power-load thresholds, water-use permits, noise limits and ratepayer rules. Kentucky’s Mercer County, for example, drafted an ordinance with a 1,500-acre cap and limits on rezoning farmland. Iowa counties have passed moratoriums to write clear rules, reflected in comprehensive plans, first. These are specific and enforceable. Nothing fuzzy to see here.
The fact that “what counts as AI” is fuzzy doesn’t make “a 600-acre campus drawing X megawatts and Y gallons from a local aquifer” fuzzy. Whatever software runs inside, the footprint on land, water and the grid is the same. My concern is about putting that footprint on farmland, driving up local power costs, damaging rural water supplies, driving down adjacent property values, creating financial suicide pacts with local polities, and the like.
You are correct.
AI is the computer and the software.
DC is the massive memory system.
The DC debate is real and I have written about it but here I am referring to the issue of regulating AI itself, especially AI development. Turns out there is a well organized and well funded “end of humanity” scare movement calling for international regulation. Think UN.
I understand what you are arguing — but AI isn’t just a shoulder shrug. Sure it’s in lots of stuff already, and sure it (in some form) has been around for a number of years. But that’s not what is causing controversy (though maybe it should have).
When you dismiss folks as a bit tin-hat for worrying about the ‘end of humanity’, well, you may be right. But the folks that are on the inside — developing the technology, not reporting on the technology — are the ones worried, raising the alarms.
If you disagree with them, then you have to present an argument directed to the right target. You are not qualified to argue with them about the tech, about the algorithms and applications; you are limited to arguing that they are conflicted. In other words, if you want to argue that the worry over AI is pure click-bait hype, your argument must be about why the folks who designed it, spent their careers in the field, are lying. You cannot argue that they are wrong on the merits; your only avenue to make an argument is about their character.
That said, the application of tech is problem in its own right. You and I can use it and use it very productively because we spent decades doing the real, applied work that benefits from AI. Our grandkids have not. They are the ones using AI — but not knowing enough to know when it is wrong. So, as someone else has said nearby ‘senior staff aren’t hiring junior staff’ — isn’t that rather cataclysmic for family formation, society, democracy itself? Think it through.
The biggest direct impact of AI is professionals don’t need juniors anymore.
In IT, AI functions as a very capable junior programmer.
Ditto for lawyers, accountants and lots of other professions.
There is a narrow sliver of professions- trades jobs and AI adjacent jobs like data science – where juniors are still needed.
But if society doesn’t catch on real soon about what is happening, and steer students towards AI resistant careers, we’re in for a lost generation of young people.
On the other hand there is the new job of managing AI systems. I am reading about people burning out because they have to manage 3 or 4 systems. AI may be labor intensive! Because we are not just doing what we did but a whole lot more thanks to AI.
“some of which seem pretty harmless. The little AI summary that comes with each Google search,”
Harmless to you, but it has a measurable effect on the number of hits on a website and hence their income.
Google AI Overviews traffic loss is no longer a future SEO problem for small businesses. It is already showing up in the data. According to a recent study by Stackmatix, 58% of all search results pages currently include an AI Overview at the top of the search result list. For search terms where AI Overviews are displayed, the number of users who clicked through to a website decreased dramatically. This is not a minor shift. Many visitors may be reading a Google AI generated version of your content rather than clicking directly to your site.
DataSlayer analysis showed a 61% drop in organic click-through rates when an AI Overview is present. WordStream reported that zero-click searches now represent a major share of Google queries.For a small business that spent years building organic search rankings, the math has shifted. Organic search traffic you have worked hard to earn is now being consumed by Google prior to many users accessing your website.
I meant harmless in the popular sense of threatening human extinction. Yes there is lots of economic impact. The journal publishing community is really struggling with this. Interestingly journal articles are listed more often but visited less because the bot finds them. The Scholarly Kitchen blog of the Society for Scholarly Publishing has many article grappling with this impact. (I used to blog there.)