This second generation of artificial intelligence has entered the marketplace with promises of dramatic acceleration of learning, removal of monotonous tasks and augmentation of human capabilities. 

The speed at which AIs are learning is profound and likely hard to comprehend for most human beings. In contrast, while human beings still hold the monopoly on human intelligence, our current pace of learning, collectively, is so much slower. We are becoming intellectually extinct. 

Removal of monotony has shown up, but it also threatens entire – non monotonous – industries like photography, illustration, video and writing. Letting a new being out into a marketplace means it will find the areas of least friction to consume – these art forms are easy targets. That which makes us human is being consumed by the artificial. 

The augmentation of human capabilities has certainly shown up in a variety of places. But, on this one, perhaps the promise wasn’t set up properly. As a new member of society, artificial intelligence should enrich the whole of society. Augmentation is a frankenstein of the human species, advancement comes from enrichment. 

We are still morally responsible for our creation. These promises, while not all false, have early indications of less than adequate contributions, in the ways we might hope. The good news is there will be another generation of this form of intelligence and it should have anchors in deterministic methods vs probabilistic. 

The current large language models fueling this generation are probabilistic, which is why you can get different answers to the same question. Also why it might feel more like chatting with a politician than an actual human being. In contrast, a deterministic model or what is being called an Architectural Language Model, is anchored in human intelligence and consistency. You might say probabilistic models work in shades of gray while deterministic works in black and white. 

If you’re wondering, why does this matter, well it has more to do with what makes us humane. We are born with values, instinctually built into our human code and add more as we are raised by parents, family, friends and our communities. These values anchor us and disincentivize us from giving two different answers to the same question. Artificial Intelligence contains no such concept of values as a layer of how it interfaces with humanity. 

AI makes up answers when it doesn’t know them, and a timeout, dunce cap and five minutes in the corner of a classroom isn’t an option. Just like you can’t rely on a co-worker, family member or politician who rarely tells the truth, we can’t trust artificial intelligence. We need a filter, like the one you put on when you’re around young children, trying not to be the one who teaches the neighbor kid the F word. 

This same idea applies to artificial intelligence directed to help build enterprise scale software for highly regulated industries, banking, healthcare, insurance and financial services. The precision cannot be compromised by an overly zealous, optimistic or confident AI, it needs to be consistent and anchored in truth. This type of advancement will likely come from the next generation of Architectural Language Model advancements, like those used in Byggr.ai.

It will be good to see more attention spent on where AI can relieve greater areas of monotony, enrich human intelligence and help us achieve greater advancements for a humane culture.