AI: The New Overlord?

Every epoch in the digital era has arrived with a furore of how it will change society, business and investing. We can all remember the pundits providing commentary on how much things would change with the personal computer, the internet and the smart phone. And they did.

But this is the first time I can remember new tech being discussed as an existential threat to our species.

Recent developments in Large Language Models such as ChatGPT from Open AI are significant enough that some of the world’s most qualified researchers are making a plea for us to stop, or at least pause, before things spiral out of hand. It puts us in a quandary, because if anything has become clear in recent years, it’s that our legislators are severely lacking when it comes to their understanding of exponential technologies.

he line of questioning at the recent Tik Tok congressional hearing in the USA was telling. There’s a reason the top 6 technology companies in the USA have a market cap of over $US9 trillion and represent seven of the 10 biggest companies. So far, Big Tech, has been allowed to do whatever it wants.

A Rare Event

The Open Letter published by The Future of Life Institute requesting we put the brakes on all AI research is a rare event. From recent memory, we’ve only ever had such a moratorium on human cloning and germline modification.

This time, they are less worried about the modification of our species and more about creating another; a new species superior to us in every measurable way and many ways which we’ll fail to even understand.

As a long-time technologist, I think it would be prudent to pause and take heed of the letter. But first, we all need to understand the different types of AI, what they mean and what they do, as most people lack that context.

Today’s article will be less about investing, and more like an explainer. I feel it is important.

Types of AI

At its simplest, Artificial Intelligence is a field which combines computer science and robust datasets to enable problem-solving. AI is achieved by creating algorithms and computer programs that are capable of learning from and adapting to data and experiences.

It’s different to industrial machinery, as the software uses AI to ‘think’ like a human and perform tasks on its own. The AI chooses what to do, based on a set of instructions. A machine simply performs a pre-determined task upon instruction. This context is important so that we can understand the different types of AI and their implications. To date there hasn’t been much delineation between them, which at this point is very important.

There are 3 clear types of AI’s:

• Artificial Narrow Intelligence (ANI)
• Artificial General Intelligence (AGI)
• Artificial Super Intelligence (ASI)

Artificial Narrow Intelligence: ANI represents the large majority of AI’s we use in modern society. These are generally rule-based AI’s which decide something based on an If This, Then That Protocol (IFTTT). A simple example of this is would be Google maps which generate for you the quickest way to get somewhere based on distance, traffic conditions and estimated traffic conditions during the journey. A self- drive car also operates under this doctrine, as does something like search engine results. We could even add image recognition, and deep fakes to this list. They are described as narrow because they typically operate in a vertical. The only context they have is within a specific framework. These AI’s do improve, iterate and learn, as their dataset gets bigger. But they are limited to a specific realm and they are mostly probabilistic in nature. Most of these ANI’s outperform humans, but they lack nuance, are imperfect and can’t really be applied to different tasks.

Artificial General Intelligence: Chat GPT falls into this description and is the reason why some people are becoming very concerned. AGI’s take intelligence and make it horizontal, meaning they can be applied to a number of, and potentially unlimited, contexts – much like biological beings can. An AGI is capable of performing any intellectual task that a human being can do. Such systems can reason, learn, plan, understand natural language, recognize patterns, and solve problems. This is what Large language Models (Such as ChatGPT / Bard et al) can do, and the reason they can do it is that they’ve learned via the entire gamut of human experience, in what is our species’ killer app – Language. All they need is the correct input. AGI is intended to be a more flexible and adaptable form of AI that can adapt to new situations and environments in the same way that biological beings can. AGI systems are capable of generalizing knowledge and taking skills from one domain to applying it to another. Already, ChatGPT has shown a number of emergent capabilities and intelligences it was not designed for. These were surprises to the developers themselves. The fear is that this will lead to a form of self-awareness.

Artificial Super Intelligence: ASI’s are different, and the major risk we face as this ‘unauthorised’ AI experiment continues. It refers to a hypothetical form of artificial intelligence that surpasses human intelligence in every way and is capable of designing and improving its own systems, even beyond human comprehension. It has the potential to develop its own objectives, agendas, and organise the factors of production for its own purposes. It can go beyond instruction and become a new form of living entity, which can spawn new improved versions of itself, and even organise hardware to build out any physical manifestations of itself which it may require to perform tasks against its own desires – whatever they may be.

ASI is often considered to be the ultimate form of artificial intelligence, and could potentially revolutionize or, end human civilisation as we now it.

Didn’t mean to scare you.

The Singularity

The idea of the Singularity, as proposed by Ray Kurzweil, is based on the concept that once ASI is achieved, it would rapidly accelerate technological progress and fundamentally transform human civilization. According to Kurzweil, the Singularity would mark the point at which technology becomes so advanced that it is no longer possible to predict or comprehend the future. Kurzweil believes that this is both inevitable and irreversible — a technology black hole of the unknown. He currently predicts this will happen by 2045.

Externalities

As Alan pointed out in his Weekend Briefing, humans are typically slow to respond to risk of externalities. This was true with climate and fossil fuels. Often, we are unaware that they will even exist and pose a threat. This is not one of those times. If Kurzweil is right, global temperature could be the least of our worries.

Once of the biggest challenges we face is that we now live in an era of democratised technology. No one needs government approval to commence their own AI development lab, or Manhattan Project style budgets. But the implications could become nuclear. In addition, our geo-political environment is one in which we can sure that the Chinese Communist Party won’t pay heed and press pause even if we democratic countries do. 

While it could be that we never generate an ASI (I’m still waiting for my fully autonomous vehicle) this is a time when prudence is desirable. As the Open Letter mentions, the last thing we want is unelected tech leaders putting our civilisation on the line. What I’d much rather is courageous leaders using the powers they’ve been granted to make decisions no individual can.