Is AI Actually Getting Dumber?

The received wisdom is that we are headed for the Singularity – where the intelligence of machines exceeds that of humans. But the betting is that we are headed for the Modularity – we will know and understand a lot more about many more things but understand even less about how to control our overall environment.

Nothing Like a Good Old-Fashioned Collapse

Here’s the deal. All the AI stuff is coming from ChatGPT 4. That is based on the so-called large language model approach (LLM). But that model is now enthroned as the fount of wisdom. The returns to this approach are now on the way down. The more you add to it the less you get incrementally. In other words, AI will get smarter but less and less. We’ve already reaped the main benefits and we shouldn’t expect the ones to come in future to beat out the ones that have already landed.

The AI luminati are calling this phenomenon “model collapse”. You can check it all out here if you aren’t too bummed out to still try. So, what does this mean for the future of AI?

I guess no-one ever really figured out how AI would progress except that it would be probably dramatic and fast. But the ChatGPT case suggests that AI might actually do the opposite, that is proceed in stepwise fashion each time it finds a new seed to grow.

ChatGPT is the first of the seeds and on my thesis looks like it will end there, albeit after making a dramatic contribution to the growth of general intelligence. So, there will be other seeds and other steps. Some that come to mind are:

New GPT Seeds?

Quantum behaviors let’s call it QuantumGPT in which bodies of disparate knowledge are quantized to bring about dramatic steps in generatingnew insights into quantum models of what is going on in the world.

Spatial data: We’re seeing the sudden emergence of VR/AR headsets with huge increases in functionality. I think that going to result in another step in spatial insights, as in SpatialGPT.

Mathematics: I think we’re on the verge of several steps, all centered on different fields of math which will generate their own GPTs for new approaches to reality. This includes our concept of time, thus RiemannGPT.

Topology: Actually, topology is just one of the many GPT spinoffs we can expect. It seems to me that topology and the math of dimensions is going to lead to new approaches to AI and reality, as StringGPT focused on how topologies lead to new models of the world and how to predict it.

DNA genomics: Genomics is just shorthand for the many types of GPTs that will emerge from the genomics fields. This is a whole new role of its own that will open up new approaches to AI and prediction as in DNAGPT.

Neuronal Connectonomics The use of connectonomics to squeeze dramatically increased levels of cognition out of a brain that we might feel now is at its peak, but probably still has several new peaks to ascend as in ConnectGPT.

More General Intelligence, Less Smarts?

So, my perspective is that ChatGPT has had its day in the sun. It will continue to make advances but not at the rate of the last couple of years. That role will be picked up by the GPTs I have set out above.

Each of these will have its own type of epistemological approach, its ways of identifying new types of data and how these are defined and measured. Each of the subfields will generate its own GPT and in turn these will each generate its own GPT revolution which will lead to another quantum of advance until the field shakes out and – dare we say it – becomes mature.

The result is a constantly advancing field of AI based on different frameworks and GPTs. Each will generate new ways of predicting the future and doing useful things. But when one runs out of puff the new ones will emerge to advance the field in different directions.

The explosion of human knowledge into new and different subfields will lead to new types of GPTs that will in sum increase the overall amount of general intelligence available to everyone, even though it has emerged from the different GPT subfields.

Modularity is Coming!

But I don’t believe that more general intelligence will lead to the Singularity. I think that instead it will lead to what I call the Modularity. That is the state in which we are all better informed about the meaning of the parts while becoming less insightful about the whole.

In the era of Modularity, we will be even more prone to destroy our overall social and physical environment since they can’t make the modules work together. In other words, Modularity is where we are right now, just a lot worse.

We will see the dystopian future even more clearly, but we will be even less equipped to do anything about it.

Per ardua ad astra

#qaXpJ9jh7EfK5Ct