Skip to main content

Meadway and Walker, the constraining economics of AI







There was an interesting debate on Novara Media between Michael Walker and James Meadway, which explored AI from a different perspective than is usually aired, certainly than in print and broadcast media. It involved the economic realities of AI.

Michael Walker is the Contributing Editor at Novara Media, and one of the better presenters of complex issues. 

Meadway is one an astute economist, the director of the Progressive Economy Forum his podcast, Macro Dose, can be found at patreon.com/Macrodose 

The Debate gets twisted


Walker was framing the debate from Hinton's much publicised resignation from Google, and his subsequent warning. The debate that followed was soon to change direction.

Walker: Geoffrey Hinton, a computer scientist at the University of Toronto, is considered one of the fathers of artificial intelligence (AI). He helped develop neural networks, a type of machine learning that mimics the way the human brain works. Neural networks are now used in a wide range of applications, from self-driving cars to facial recognition software.

In a recent interview with The New York Times, Hinton expressed his concerns about the future of AI. He believes that as AI systems become more powerful, they pose an increasing threat to humanity.

"Look at how AI was five years ago," Hinton said. "Now take the difference and propagate it forwards. That's scary."

Hinton is particularly worried about the potential for AI to be used to create fake news and propaganda. He believes that AI-generated content could be so realistic that it would be difficult for people to tell what is real and what is fake. This could have a devastating impact on democracy and society as a whole.

Hinton is also concerned about the potential for AI to be used to create autonomous weapons. He believes that such weapons could be used to kill without human intervention, and that this could lead to a new era of warfare.

"I fear a day when truly autonomous weapons, those killer robots, become reality," Hinton said.

Hinton's concerns are shared by many other experts in the field of AI. However, there is no consensus on what to do about the problem. Some experts believe that we need to develop international regulations to govern the development and use of AI. Others believe that we need to focus on developing ethical guidelines for the use of AI.

Whatever we do, it is clear that we need to start thinking about the potential risks of AI now. If we don't, we could be in for a very dangerous future.'

Meadway was having none of this line of discourse. Instead he stressed that we need to examine the problems that we are currently experiencing, arguing that the economics of resourcing AI development would have significant impacts in delaying any AGI like capable machine.

Meadway further indicated that the trade papers of the IT industry are demonstrating the increasing costs of AI and the scarcity of resources that's beginning to accrue. Here are just some snippets from articles that expand upon this point: 



'AI consumes more energy than traditional types of computing, and just one model's training can consume more electricity in a year than 100 US homes do.

According to a study article released in 2021, training GPT-3, a single general-purpose AI software that can generate language and has numerous applications, required 1.287 gigawatt hours, or roughly the amount of electricity used by 120 American houses annually. According to the same report, such training produced 502 tons of carbon emissions, or roughly 110 US automobiles' worth in a year. Although training a model has a significant upfront power cost, researchers found that in some circumstances it only consumes around 40% of the power used by the real use of the model, with billions of applications for popular programs flooding in. Also, the models are growing larger. The 175 billion parameters, or variables, that the AI system has learnt through training and retraining are used in OpenAI's GPT-3. Only 1.5 billion were consumed by its predecessor.'


Forbes, published March 22, 2023 Green Intelligence, why data and AI must become more sustainable

'Sanjay Podder, managing director and global lead of technology sustainability innovation at Accenture, says that the exponential growth in data and its increased energy demand could actually counteract and impede our global progress on climate change.

Right now, the AI community has adopted a “bigger is better” attitude regarding data and artificial intelligence – but that approach threatens to inflict major environmental damage in the future.
Tech experts will need to expend greater and greater amounts of energy to build increasingly larger models, with decreasing improvements in performance.'

A financial arms race, generating hype to generate profitability


Generative AI has been costly to build, but particularly to train for Alignment. Legal costs may also begin to accrue, quite rapidly too. What's been very apparent over the last 18 months or so is that these costs are hitting the share values of the major players. Hence the large number of layoffs by the tech giants we've seen recently. The real profits, over this time has been with the hardware companies. 

NVIDIA, who make the GPUs that run AI models, that are financially benefitting from the AI hypesters. The costs of the hardware to develop, maintain and run LLMs has been staggering. One of the reasons why Google recently announced the development of it's chips, so has Microsoft 

Share prices of the major players over a one year period.


Meadway concludes

Meadway: 'It's a bit like climate change. We can say "wow, what if this terrible thing happens in the future?" But it's already happening. These terrible things are already happening. We should deal with the terrible things that are already happening.

And in dealing with those terrible things, we can start to say "well, how can we put under some sort of public supervision and control the development of these particular data techniques and these machine learning techniques?"

I think when George said this, he was kind of right. It's a process of competition. Google is scared of Microsoft, so they have to charge at this and put loads of money and time and infinite resources into it. The process of competition is driving this in a chaotic way.

What can we do that starts to put some public control, some democratic accountability, as much supervision as we can? We can have a serious society-wide conversation about how we start to use these things. We can potentially use them for all sorts of really good, exciting purposes. The invention of new drugs, the invention of far more efficient ways to use the resources we have, is something that AI and big data could be particularly good at.

We can do that if we start to get in our heads that we need to challenge the structures that are investing in and applying and using these new technologies. And that means, at the very least, thinking about regulation now. Not because of the existential risk, which I can write a kind of science fiction story about how this might happen in the future. I've tried to give you a sort of economics version of why this is quite implausible given the technologies that we have.

But what we really have to deal with is that issue of corporate power. The fact that decisions over these things are made by a really quite a limited number of institutions globally, with very limited real public control and oversight. And if we don't start to deal with this, then the kind of problems we saw with the development of things like Facebook and all the other big tech, big data companies from the early 2000s onwards, they will reproduce themselves in a much more dramatic and unpleasant fashion through machine learning and the application of machine learning.

That to me seems like something to really focus on now. And potentially then think about existential risks, rather than the immediate existential risk of "oh my God, we'll have a computer that will run off and kill us all." The problem is capitalism. The problem isn't the computer directly.'

It was a sophisticated argument from Meadway: acknowledging the issues we currently face, including the lack of public accountability, the sorry state of regulation, economic and environmental constraints, alignment issues, and the intractable problem of capitalism. 

Watch the full debate.



Comments

Popular posts from this blog

OpenAI's NSA Appointment Raises Alarming Surveillance Concerns

  The recent appointment of General Paul Nakasone, former head of the National Security Agency (NSA), to OpenAI's board of directors has sparked widespread outrage and concern among privacy advocates and tech enthusiasts alike. Nakasone, who led the NSA from 2018 to 2023, will join OpenAI's Safety and Security Committee, tasked with enhancing AI's role in cybersecurity. However, this move has raised significant red flags, particularly given the NSA's history of mass surveillance and data collection without warrants. Critics, including Edward Snowden, have voiced their concerns that OpenAI's AI capabilities could be leveraged to strengthen the NSA's snooping network, further eroding individual privacy. Snowden has gone so far as to label the appointment a "willful, calculated betrayal of the rights of every person on Earth." The tech community is rightly alarmed, with many drawing parallels to dystopian fiction. The move has also raised questions about ...

What is happening inside of the black box?

  Neel Nanda is involved in Mechanistic Interpretability research at DeepMind, formerly of AnthropicAI, what's fascinating about the research conducted by Nanda is he gets to peer into the Black Box to figure out how different types of AI models work. Anyone concerned with AI should understand how important this is. In this video Nanda discusses some of his findings, including 'induction heads', which turn out to have some vital properties.  Induction heads are a type of attention head that allows a language model to learn long-range dependencies in text. They do this by using a simple algorithm to complete token sequences like [A][B] ... [A] -> [B]. For example, if a model is given the sequence "The cat sat on the mat," it can use induction heads to predict that the word "mat" will be followed by the word "the". Induction heads were first discovered in 2022 by a team of researchers at OpenAI. They found that induction heads were present in ...

Prompt Engineering: Expert Tips for a variety of Platforms

  Prompt engineering has become a crucial aspect of harnessing the full potential of AI language models. Both Google and Anthropic have recently released comprehensive guides to help users optimise their prompts for better interactions with their AI tools. What follows is a quick overview of tips drawn from these documents. And to think just a year ago there were countless YouTube videos that were promoting 'Prompt Engineering' as a job that could earn megabucks... The main providers of these 'chatbots' will hopefully get rid of this problem, soon. Currently their interfaces are akin to 1970's command lines, we've seen a regression in UI. Constructing complex prompts should be relegated to Linux lovers. Just a word of caution, even excellent prompts don't stop LLM 'hallucinations'. They can be mitigated against by supplementing a LLM with a RAG, and perhaps by 'Memory Tuning ' as suggested by Lamini (I've not tested this approach yet).  ...