Saturday, September 28, 2024

Reflecting on AI accountability for misinformation - and solar powering the revolution

The Great Big AI Crisis of 2024 will likely wreak its worst harm via political misinformation and manipulation this year (next month!) But it’s swamping its way into science, as well. 


It is not always so easy to spot the use of AI. But one clue is that ChatGPT tends to favor certain words… such as meticulous, intricate or commendable.” 


Of course all such detection cues are temporary.  Even when we point them out to each other (as I just did), that only helps train the systems to avoid overusing them.


So, is it hopeless to dream of escaping the Age of Tsunamis of Lies? 


None of the palliatives proposed by the geniuses and mavens of AI - ranging from 'moratoriums' to EU style regulations to 'privacy rules' stand any chance of helping much. Only one thing will even possibly work and that is siccing AI programs onto each other, competitively, with incentives for them to tattle on misinformation, as I describe here in Wired: Give Every AI a Soul - or Else.


And more vividly detailed? My keynote at the May 2024 RSA Conference in San Francisco – is now available online.   Anticipation, Resilience and Reliability: Three ways that AI will change us… if we do it right.”  


Let's dive in to how that could work.



== AI, Ai Ai!!! ==


In all of those places, I've pointed out that we already developed one fairly effective method for detecting and deterring liars, foiling harm doers and preventing a return to 6000 years of lobotomizing feudalism. Imperfectly, by far! But light years better than any prior culture.


Was it carefully-deliberated and designed laws? Those can help, but no. Paternalistic protection by the state? Ditto, and dangerous. The method I'm talking about – the one innovation that gave us everything – has been to flatten the playing field and empower a wide enough diversity of citizen-players, so that we can pit elites and potential predators against each other. 


It's called reciprocal accountability.

Lawyer vs. lawyer, for example. And the one area where we’ve made the greatest advances? The most effective reciprocal accountability system of all: science.

Scientists are the most competitive creatures our species ever produced. Young scientists are like top guns roaming Main Street, looking for a paradigm or pompous theory-pusher to topple with the six-gun of evidence. They make their mark by finding at least a small chink in the current standard model to critique. 

(And yes, this is diametrically opposite to the slanderous image of wimpy consensus-hugging eggheads that's pushed by anti-science cult media.)

The results are very often positive sum. Negatives and falsehoods get canceled out by competition, while positives can combine through better-tested models, new alliances, and cooperation. 

In other words, a healthy market.

(Politics is supposed to be like that, by the way. And it was, overall, till a mad cult waged open war against the very concept of negotiation, turning our political institutions zero-sum and now negative sum.)

All of which leads back to my notion about Artificial Intelligence. That we should try emulating, in this new ecosystem, the one and only method that ever reduced misinformation and predation among organic humans. Again, I am calling for reciprocal accountability on cyber beings, applied by cyber beings.


While I may be alone in offering institutional innovations to achieve this, some groups have lately been coming up with practical methods.

To pinpoint when a language model might be confabulating, the new method involves asking a question multiple times to produce several AI-generated answers. Then a second LLM (Large Language Model) groups these answers according to their meaning; for instance, “John drove his car to the store” and “John went to the store in his car” would be clustered together.” Leading to a new metric “semantic entropy.”


“Other anti-hallucination methods have used LLMs to evaluate generated answers, through approaches such as asking a single model to double-check its own work. But the paired system improves on this...” 


So, yeah. The idea is starting to gain (a little) traction. Sic em on each other. But with incentives that reward those who do us all - and Truth - the most good.


When did I speak of a 'new ecosystem'? Here's an earlier posting where I talk about that...

Are we making new kinds of 'ecosystems'? 


... and how we organic humans (orgs) will still control - maybe for another decade or two - the new ecosystem's 'sun.' 


And if we handle this power wisely, it may shine upon the fabled 'soft landing' alongside our new cyber children.


And speaking of the sun...



== The solar revolution is here ==


This article in the Economist discusses how solar power's recent huge boom may be only the beginning. 70 years after first introduced by Bell Labs... and frequently sabotaged by filth merchants:


"Today solar power is long past the toy phase. Panels now occupy an area around half that of Wales, and this year they will provide the world with about 6% of its electricity—which is almost three times as much electrical energy as America consumed back in 1954. Yet this historic growth is only the second-most-remarkable thing about the rise of solar power. The most remarkable is that it is nowhere near over..."

...Much as I portrayed in EARTH in 1990. Though by now the question is why are the lords of carbon still backing the Denialist Movement, when this revolution can't be delayed any longer? And when denying all the heat waves and tumbling glaciers and acidifying oceans and super storms only makes you look like a jibbering loony?


Economies of scale have taken over now and batteries and methane and new nuclear will handle surge capacity, so why do they continue to back a cult bent on wrecking the planet that they must live on, too?

Well, the Russians and Saudis are still utterly fossils (and fossil-dependent) and plausibly they are the ones holding blackmail on almost all high Republicans. And unlike mere corruption (which can sometimes think long term), blackmail is always imminent and near term. It is only about satisfying the blackmailer today.

THAT - fundamentally - is what the Ukraine war is about and it is why the GOP is slavishly devoted to a Kremlin and slightly-relabeled KGB that they once despised. And it is why Ronald Reagan would spit in their eyes



== Future Tech ==


An interesting report - The Battery Mineral Loop - suggests that we may be recycling precious battery elements much more efficiently pretty soon, and that the race to mine them may be a temporary thing, lasting as little as 1.5 decades. Which is still a bridge we need to accomplish with careful management, good politics and lots of science. 


Australians have developed night vision optics so thin they might be barely noticeably different from your normal glasses.


Great news that soon computer chips will be able to store the energy they need for rapid operations right on the chip, itself, through new micro-capacitors. 


Oh, is it possible to find something positive, even lightening, from a Trump action? I have long demanded examples where a Republican U.S. administration had major, palpable comparative positive outcomes for the nation and world as a whole, and not just benefiting narrow, conniving cabals. Certainly, when the Trump guys sold off the U.S. Helium Reserve to buddies, for almost nothing, those pals immediately jacked up prices for the element that’s utterly necessary for many medical uses, such as supercooled imaging systems. And quantum computers.  


Only now the deal might (accidentally and unintentionally) benefit us, as a major new source of Helium has been found deep under Minnesota. Like the recent phosphate discoveries in Norway, it may be good news for civilization… and bad for market-cornering cheaters…


…till we get the mother lodes from asteroids, that is.


So... what now?


Now YOU do your bit for the Enlightenment, for science and civilization. 


Check your voter registration and those of friends! Give or volunteer if you can. Wear this. And remember that earlier victories in the recurring U.S. Civil War always led to golden eras. 


Imperfectly!  But we moved forward, correcting errors made by every prior generation. Making new ones for the next to correct! But moving ahead.


Toward the stars.


5 comments:

Chris Heinz said...

I like "Age of Tsunamis of Lies".
I have been referring to it as "The Coming AI-Generated Bullshit Apocalypse."
;->

duncan cairncross said...

From the previous comments
Iran and the middle east
The USA caused the situation in Iran
First overthrowing (with British help) the secular democratic government
Then supporting a bloodthirsty dictator - to the extent that the only surviving opposition was the religious one (Mullahs)
The Mad Mullahs would have been in power for 5 years - ten at most - but the USA provided every despot's dream - an external enemy
First by sicking their other bloodthirsty dictator on Iran - most of today's Iranians lost friends and relatives in that conflict then by keeping the pressure on
Obama negotiated with the Iranian moderates - the days of the Mullahs were numbered
Then the Orange Cockwomble screwed up the treaty and stabbed the moderates in the back
The Mad Mullahs were saved! - and the moderates crippled so that renewing the Obama agreement became impossible

duncan cairncross said...

For todays post
The first stage will be not so much setting your AI against the others as USING your AI to find the patterns - to find where the news has been distorted

This involves using AI as a tool - and will work long before AI is smart enough to do it itself

Larry Hart said...

duncan cairncross:

long before AI is smart enough to do it itself


As long as AI is trained by treating every sentence on the internet as having equal weight, and then further trained by ingesting the output of other AIs, AI's intelligence is going to trend downward, not upward.

GMT -5 (Hugh) said...

David, regarding that Union Army cap, here is a picture of me as part of the Union Army at the 3rd Battle of Manassas (aka the 1974 National Model Rocketry Convention).

https://www.facebook.com/share/p/14Vt57Gum6/?mibextid=qi2Omg

I am carrying my Estes Egg-Roc for the egg lift competition. Sadly, my rocket barely cleared the launch rail. My dad and skipped the con for a day to drive into DC and be part of the crowd in front of the White House when Nixon left office.