Re: Societal impact of mega AI, and consolidation of power

Hi David,

On 5 Aug 2025, at 18:13, David Booth <david@dbooth.org> wrote:
> 
> Interesting work and commentary on the impact of AI on society:
> https://ainowinstitute.org/publications/research/executive-summary-artificial-power
> ...
> There is something particularly pernicious about the intersection of unfettered capitalism, authoritarianism and big AI, which will inevitably be used to further accelerate developments that are increasingly harmful to society and our fragile planet.
> 
> I am wondering what we can do, to help prevent these downsides.


This indeed a question of societal values, and encouraging debate that shines a bright light on the upsides and downsides of different approaches, so that we don’t sleep walk into an even worse dystopia that we’re in now.

For example, if we envisage a future with a smaller human population (as suggested by today’s falling birthrates), how can we exploit AI to give all of us meaningful and productive lives?  This implies a choice in where automation is appropriate and where it is not, along with the need to redistribute the revenue raised from automation to support jobs involving direct human to human contact, e.g. social care, education, healthcare, sports, theatre, etc.  This should be determined by open and widespread discussion of what kind of society do we want to live in, including the environment around us and our stewardship of nature. Today's populist politics won’t deliver this, and we need to usher in a new era of enlightenment.

For a technical perspective, we should focus research on dramatically reducing AI's need for huge data centres along with their vast demands on energy and water.  For this we should a) work on techniques to better mimic human learning, memory and reasoning, and b) work on techniques that mimic the power efficiency of the brain. For the latter, a strong candidate is hybrid photonics using optical waveguides in place of electrical conductors.

This can be fabricated using silicon on insulator, with germanium on silicon for infrared emitters and detectors, oxygen or nitrogen doping for memristors, along with integrated conventional CMOS circuitry. Today’s advanced chip fabrication techniques are capable of the nanometre scale regime needed for quantum dots, and other photonic components. It is likely that this will require the development of biologically plausible alternatives to back propagation + gradient descent.

If we are successful in these aims, powerful edge-based AI systems will need just a few watts for continual learning and reasoning, undercutting the need for vast data centres, and democratising AI in a way that is reminiscent of how personal computers reduced the dependency on mainframes.

Best regards,

Dave Raggett <dsr@w3.org>

Received on Wednesday, 6 August 2025 08:23:21 UTC