Progress and Artificial Intelligence?

Current battles over Artificial Intelligence (AI) are tied to the idea of “progress.” For the past century or so we have taken it for granted that progress is a thing. Progress is the idea that we can build on past knowledge to create new technology. Which in turn allows us to acquire more knowledge and more technology.

This led to the idea that each generation would be better off than the previous generation. But this is a very recent idea. For most of human existence, quality of life did not change much with each generation. People awaited the afterlife for a better life.

Elders were revered for remembering what little was really known. Ancient Greek philosophers Hesiod, Empedocles, and Demosthenes felt that society was actually in steady decline from a previous golden age.

Some societies kept calendars with long term cycles. Calendars have monthly and annual cycles governing planting, harvesting, and cultural festivals. But the Mayan calendar had a long cycle of about 5,125 years. The Hindus had nested “yugas” – discrete units of time equaling 4.32 million years. 1,000 of these comprise a mahāyuga – one daytime in the life of the creator god Brahma – accruing all the way up to Brahma’s entire lifetime (Maha-Kalpa) of 311 trillion years.

They could conceive of vast time periods while maintaining the idea that there is no actual progress. Many cultures have the idea of this historical cycle: Order ⇑ prosperity ⇑ decadence ⇑ disorder ⇑ replacement ⇑ renewed order.

With progress comes pushback. The Luddites famously smashed weaving machines in a vain attempt to protect the jobs of hand weavers.

I have been mystified about why religions don’t seem to show progress. Some new religions deliberately seek to return to “traditional” values rather than move forward.

Think of the Make America Great Again (MAGA) cult. The key word is “Again.” As if there was some time in the past where everything was Great. And we just need to restore that past. Oddly, they are trying to “restore” laws like abortion bans that never existed in most of human history.

Progress is disruptive. People lose jobs. The question is how to solve these real problems and not create imaginary enemies and even bigger problems.

There is much talk now of regulating AI. And the far right and the far left openly express Luddite views of destroying data centers. I have been following the up and down progress and dead ends of AI since I was a teen in the 1970s. For the past 25 years I have followed Ray Kurzweil’s roadmap of how AI will affect society. His predictions have been very accurate and we are on track for far more dramatic change.

I very much doubt that traditional “regulation” of AI can be much of a thing. Past technology could be regulated in visible ways. Nuclear materials and nuclear refinement can be tracked. Enforcing rules is still a challenge, as North Korea and Iran have shown. But at least we know what it means to have a regulation.

In the case of AI Large Language Models (LLMs), massive data centers are needed in order to “train” them. But it does not take a lot of processing power to use them. The “weights matrix” (functional core) of a major AI can be stored on a thumb drive. And the model can then be run locally on a laptop. How can you regulate that?

Some fear that AI will be used recursively (a continually perpetuating self- improvement loop) to write new code for better AIs. Is this a bug or is it an essential feature? Kurzweil sees this as the key feature that transcends the idea of exponential growth. We will now have double exponential growth: exponential growth begetting exponential growth. This leads to his repeated talk of a “singularity” where this rise becomes so steep that it literally changes everything.

There is talk of banning this recursive code writing. How is that possible? Again, we are already to the point where these models can be run on a laptop. Major AI companies can be held liable for harms, but this has limits.

Talk of AI Optimism or Pessimism is not helpful. We need to face AI Reality and guide it toward the best possible human outcomes. No, it won’t be easy.