Bing’s Disclaimer

 

Bing is powered by AI, so … mistakes are possible.

Bing’s dsclaimer set me thinking – why are we doing this?

Is it a way of using stupid AI to defang the threat of AI. We can say – AI isn’t so bad, it is no threat, look how stupid it is. We are not going to give it the keys to the kingdom (the nuclear codes).

Our company is interested in developing AI that doesn’t make mistakes – AGI. To do that, the AGI has to understand what words mean, and how words connect to each other,

An example – imprudent is defined as “Not showing care for the consequences of an action”

Why make it that – couldn’t we just say “Not prudent”? English is a living language, and words are pressed into service to convey meanings they were not created to handle. "Not" is a tricky beast - he is not tall doesn't mean he is short, just average (it can mean he is short - part of the trick).

Words accrete new meanings, and the existing meanings drift, or become out of date. And a human has to know how words fit together into larger structures. An LLM can’t be bothered with any of this – more exactly, it would come at a high development cost, when more profit can be had by using marketing to push an inferior product. See “It thinks like an expert” – this blog.

What would be the purpose of AGI – why are “no mistakes” important.

In no particular order:

Boeing 737 MAX – many dead through venality, incompetence, and greed. Layers of human-based protection found wanting.

F-35 Project – a convincing demonstration of how not to run a complex project.

Voyager – let’s send a message to point its antenna away from Earth.

The Robodebt fiasco – no authoritative, easily understood and activatable source. 

Humans make lots of mistakes – they have a Four Pieces Limit, and they have an unwillingness to change their thinking after about the age of 25 (the stomach ulcers affair was a good example of this, but humans turn conservative as they age, as a way of resisting having to discard shibboleths and rebuild their cognitive structure anew every few years as the world wags on).

Is it sensible to introduce, and spend billions on propagating, deeply compromised AI in the form of LLMs? No

Will humans do it anyway? Yes

Does AGI have a future? Let’s hope so, and let’s hope it is soon. It looks like predictions about Climate Change are already demonstrating the Four Pieces Limit – if it is too complicated to think about, we can’t think about it effectively, or not on a reasonable time scale.

We will know we are being serious about it when we put the same effort into teaching a machine the English language as we do with our children (teaching our children is much easier, because they have the same apparatus, and the teacher needs much less understanding of the language).

Link