Meta AI now available in NZ; vendors tout AI, but will it be inflationary and kill the planet?

It has been a very artificial intelligence-heavy week, with big players making announcements and releasing research touting the benefits the new change-everything technology might bring.

Meta, the parent company of Facebook, Messenger, WhatsApp, Threads, and Instagram has made its virtual AI assistant available in New Zealand in English, along with Australia, Canada and 10 other countries outside the United States. Unsurprisingly named Meta AI, the virtual assistant can be accessed through chat and search integration with the above mentioned sites and messenger apps bar Threads, as well as over the web.

Facebook will also have Meta AI integrated in the social network's feed, and it can generate images such as animated GIFs.

The free version of Meta AI is built on the new LLaMa 3 large language model, with 8 billion parameters. There is also a 70 billion parameter LLaMa 3 version, which compares to competitor OpenAI's GPT-4 that reportedly has around 1.7 trillion parameters, and which Microsoft is currently commercialising.

Meta also has a LLaMa 3 model in training currently, with over 400 billion parameters.

Parameters are the weightings and settings used to process the input tokens in AIs. The more the better is the rule here, although big parameter models require huge amount of computing resources.

Meta said the LLaMa 3 models will be available on a large number of platforms, including AWS, Databricks, Google Cloud, Hugging Face, Kaggle, IBM WatsonX, Microsoft Azure, NVIDIA NIM, and Snowflake.

On the hardware side, platforms by AMD, AWS, Dell, Intel, NVIDIA, and Qualcomm will support LLaMa 3.

It is possibly a good thing that Meta updated its LLM, as researchers have started poking holes in the earlier 7 billion parameter LLaMa 2, finding among other things that it has a high risk of hallucinating, can leak sensitive personally identifiable information (PII) and isn't very fair.

This week, cloud giant Amazon Web Services published the results of a survey it had commissioned from Access Partnership, suggesting AI will bump up productivity by almost 49%, which sounds very promising.

Obviously, as a provider of AI, AWS is keen to market the upsides of the technology so as to sell more cloud services, and its figures need to be looked at in that context. 

There is however an overlap between the survey and earlier independent research that suggests AI can boost productivity if used correctly to augment human (and yes, people will still be needed) capabilities, and to automate repetitive tasks.

There's an unexpected downside to all that though, because AWS talks about New Zealand employers being willing to pay the AI-skilled some 30% more. 

Why's that? Well, it could be easy money. As AI is a general purpose technology that doesn't require any special skills as such, unless you're poking deeply under the hood running LLMs on your own machines for example, could this mean driving say Chat-GPT, Meta AI and Google Gemini and adding that to the old CV will get you a pay hike?

We, and the 70% of boomers that they survey says are keen to learn AI skills, perhaps to supplement their pensions now that the housing market retirement plan is tanking, look forward to finding out.

One august institution that isn't so keen on various demographics discovering the pay-boosting joys of AI is the Bank of International Settlements (BIS). 

In a working paper, BIS researchers believe AI will raise productivity significantly, along with GDP if the tech is concentrated in consumer-facing industries. This is a bit unevenly spread across different industry sectors, but most will benefit from using AI.

With increased AI adoption, central banks would enjoy a "Goldilocks scenario", where everything is just right, and use of the technology could ease inflationary pressures in the near term.

However, remember the 30% pay bumps that people and companies anticipate will take place through increased AI adoption? That's bad, it's likely to cause inflation to pick up and remain elevated due to AI-induced demand.

Ergo, central banks would need to raise "policy rates" to dampen said demand. Is that still a "Goldilocks scenario"?

Economist Shamubeel Eaqub is sceptical about the BIS paper, pointing to prior change-everything technologies like robotics and automation which at one point were supposed to make humans redundant.

Eaqub said the paper was more of an academic exercise to create a framework to help people think about technological advances.

Much to think about, and it might all come to nothing. British chip designer Arm, which makes the processors you have in your smartphones is and which has gone in on AI boots and all, warns future models will continue to become larger and smarter.

In doing so, more computing power is needed, which means more demand for electricity. Arm quotes a figure of 460 terawatt hours being used to power data centres currently; in six short years, Arm thinks this will triple thanks to AI. 

That's more than the total power consumption of India currently, a huge demand increase that could make the best laid plans of AI men and mice go astray. "In other words, no electricity, no AI," as Arm said.

Now, Arm prides itself on its energy efficient chips. Whether or not AI data centre builders using the company's parts en masse would lower overall energy consumption, or if the whole thing becomes as futile as building new roads to ease congestion, remains to be seen. 

That AI-induced demand through anticipated benefits that BIS researchers are talking about might possibly not happen, as we continue to heat up the planet by burning fossil fuels, partly to power gigantic data centres.

A recent paper published in Nature by researchers from the Potsdam Institute for Climate Impact Research in Germany doesn't mention AI per se, but models an immediate future in which average incomes will drop by almost a fifth by 2050. It's not going to be an income reduction shared evenly by all countries either, and a stronger word than "catastrophe" is required to adequately describe what's coming up over the next bit of time.

That's due to the climate emergency, which is projected to cause US$38 trillion worth of damage and destruction each year by that time, and it looks unavoidable unless drastic emission reductions are implemented. It's not clear how helpful AI will be with that, although its use to improve sustainability has been suggested.