My unsolicited prediction for AI in 2024

Here is my prediction for 2024 in AI, written here to compare in a year.

LLM will scale to become a new interface, via natural written or spoken language, cloud based, not local.

The effect will be comparable with the passage from cards to keyboard, or with the passage from pc to smart phone.

I can’t fathom the social consequences. Maybe:

• the internet will be really dead this time. What will be the new name of these new content delivery networks? Edge cloud? I don’t know.
  Everything will rather be created on demand, or hallucinated so to say, because it will be cheaper and faster than caching.
• maybe not in a year, but soon enough about 30% of the jobs would consist in mediating between humans and AI, in both senses. What will be the name used, perhaps reinforcement learning with human feedback (RLHF)? Nobody who is afraid of losing the job because chatgpt will loose that job, they will reconvert to RLHF consultants, or another name.
• in academia, scaling replaces research. Much easier to scrap, AI aided, for ideas and scale what gives fast returns than have new ideas. Environment hostile to research as consequence, very rare new ideas, better than ever according to micromanagement criteria.

In my Gutenberg-net analogy from april-may 2019 I say about 2024:

“the Net takes a new direction, freed from the bad influence of the old generation influencers (2024)”

https://mbuliga.github.io/gutenberg-net.html

I don’t believe LLM are this, more to the contrary this is a damping mechanism which aims to preserve old ideas influence.

It is scaling, not creation.

Added: or, if I feel bold, I’d predict:

• AI rewritten human culture. Nobody goes to sources. Inconsistencies creep makes everything from before the AI era meaningless.
• General opinion: people are not so smart because “smart” BS is easy to generate. Identity crisis, no purpose in life for those gifted with small talents, hostility and hate for talented people.
• Conformal media AI written and generated. Nothing interesting happens in the media. Everything is addictive in the media.
• Everything interesting not on the net. The damping mechanism will create a huge blow, eventually.
• A new generation will try a new direction, now that the old direction became stalled, incomprehensible and hostile. No understanding for the old ways, in comical or tragical ways, depending on your age.

Local machines

Suppose there is a deep conjecture which haunts the imagination of a part of the mathematical community. By the common work of many, maybe even spread over several centuries and continents, slowly a solution emerges and the conjecture becomes a theorem. Beautiful, or at least horrendously complex theoretical machinery is invented and put to the task. Populations of family members experienced extreme boredom when faced to the answers of the question “what are you thinking about?”. Many others expressed a moderate curiosity in the weird preoccupations of those mathematicians, some, say, obsessed with knots or zippers or other childish activities. Finally, a constructive solution is found. This is very very rare and much sought for, mind you, because once we have a constructive solution then we may run it on a computer. So we do it, perhaps for the immense benefit of the finance industry.

Now here is the weird part. No matter what programming discipline is used, no matter which are programmers preferences and beliefs, the computer which runs the program is a local machine, which functions without any appeal to meaning.

I stop a bit to explain what is a local machine. Things are well known, but maybe is better to have them clear in front of the eyes. Whatever happens in a computer, it is only physically local modifications of it’s state. If we look at the Turing machine (I’ll not argue about the fact that computers are not exactly TMs, let’s take this as a simplification which does not affect the main point), then we can describe it as well as a stateless Turing machine, simply by putting the states of the machine on the tape, and reformulating the behaviour of the machine as a family of rewrite rules on local portions of the tape. It is fully possible, well known, and it has the advantage to work even if we don’t add one or many moving heads into the story, or indirection, or other ingredient than the one that these rewrites are done randomly. Believe it or not (if not then read

Turing machines, chemlambda style
http://chorasimilarity.github.io/chemlambda-gui/dynamic/turingchem.html

for an example) but that is a computer, indifferently of what technological complexities are involved into really making one.

(this is an animation showing a harmonious interaction between a chemical molecule derived from a lambda term, in the upper side of the image, and a Turing machine whose tape is visible in the lower side of the image)

Let’s get back to the algorithmic form of the solution of the mathematical problem. On the theoretical side there are lots of high meanings and they were discovered by a vast social collaboration.

But the algorithm run by the computer, in the concrete form it is run, edits out any such meaning. It is a well prepared initial tape (say “intelligently designed”, hope you have taken your daily dose of humour), which is then stupidly, randomly, locally rewritten until there’s no more reaction possible. Gives the answer.

If it is possible to advance a bit, even with this severe constraint to ignore global semantics, then maybe we find really new stuff, which is not visible under all these decorations called “intelligent”, or high level.

[Source:

https://plus.google.com/u/0/+MariusBuliga/posts/5z4UBwq4Y7G