We Don't Need Another Mountain

  

There are mountains and hillsides enough to climb.

There was a picture symbolising AI in a LinkedIn post, and in the background was “Natural Language Processing”. Why is it such an insignificant and insipid little creature when, used properly, it is almost all of AI, excluding vision.

A natural language like English captures in a large but neat package

Propositional Logic
Existential Logic
Temporal Logic
Locational Logic

And all the object5s and operators you could ever wish for, and some you didn’t know you needed, like quantum entanglement..

Jack and Jill went up the hill – conjunctions can assemble groups of objects,

Jack fell down and broke his crown - groups of operators, groups of clauses, groups of chapters, groups of documents.

Operators can provide feedback, feed forward, amplification, inversion – again, anything you can think of, and more besides.

Sounds good, but isn’t it a bit complex?
The world is complex. If the machine does what you do unconsciously, the complexity is decreased.

When you hear “An Englishman, a Scotsman and an Irishman ran into a bar”, you immediately think of a drinking bar.

A bit more context - “An Englishman, a Scotsman and an Irishman ran into a bar on their travel plans to Vladivostok”. When you are describing complex things, the context may be on the next page, or hundreds (thousands) of pages away – you have to leave the meaning in limbo (after stripping away all the impossibilities – but if it is a new gizmo, you may have made a mistake and have to reverse your decision – it is no longer impossible).

Even the choice of POS can be difficult:

He turned on the light.             Adverb
The car turned on a dime.       Preposition

Don’t like the sound of this. Why can’t we have a single meaning for each word?

This is a natural reaction from people who are used to programming – every symbol having a single meaning makes life easy, but it also precludes involvement in more complex things (and things are getting more complex by the day).

You can have single meanings with something like Neurosymbolics (this is the mountain we don’t need – more like a foothill in comparison with English). The drawback is that the approach severely limits the complexity of the problems it can describe and the solutions it can describe back (some people may see this as a blessing). Its proponents unabashedly talk about Symbolic Logic and ANNs – one proponent of ANNs even boasted about “I love programming deep neural nets”. A problem – a “deep neural net” is a directed resistor network – its usefulness in AI is close to zero. Fortunately, ANNS are nothing like real Neural Nets, with their ability to change and reconfigure themselves. An area of interest talks about how the behaviour of young children around traffic is “unpredictable” – an ANN is not a useful tool in such a situation (How would you handle it?) The autonomous vehicle has to “keep an eye” on each child that is a risk – not easy when it is driving past a primary school shedding hundreds of students.

OK. Is there anything we can simplify?

Just a bit more complexifying. You were taught about Intransitive verbs, Transitive verbs, Ditransitive verbs. In reality, there are about a hundred verb forms – a favourite is BiTransClausal – I bet you any money your horse won’t win.

We can clean up some verbs so their actions are easier to understand.

Fred lent John $50.

We can change it to

Fred lent $50 to John

Fred’s free cash has dropped by $50 and he has acquired an asset, while John’s free cash has gone up by $50 and he has incurred a debt. We don’t know when the loan will expire (it gets more complicated when dealing with a bank, but the principle is to represent all the objects and operators involved – we are exposing the system to massive detail on defence materiel, but how else?).

Is this all vapourware?

We built a maths and logic system a long while ago – an undirected unbounded network of objects and operators. It was too hard for people to use, so we decided on the whole shebang of Semantic AI in 2000, using the bones of the maths system. We now have a vocabulary of 45,000 words, a hundred verb forms, 10,000 phrases, 100,000 senses. We are looking for something to use it on – it has to be hard to make the effort worthwhile – governments have hard jobs that they make a mess of  - legislation, materiel projects.

 

Orion Semantic AI