Do You Believe ChatGPT Will Take Over The World? WHAT!!!

It is interesting when one is very knowledgeable about a field and listens to non-experts make predictions.? It is amazing when you watch the news and some reporter "tells" you about a subject but you can tell they have no clue on what they are reporting.

?So does a historian fare better?

Recently Yuval Harari has been on a lot of podcasts and shows where he has been warning about AI and specifically ChatGPT.? He is a great historian but no AI expert.

?First, let us understand what ChatGPT does well.? It is able, based on the terabytes of data that have been fed to it, to predict the next word or even paragraph.? It is generating those words based on probabilities of what it had seen before by being trained on them.

Is that general intelligence?

Let's talk a little about AI...

?If we pull back and focus on supervised learning we are simply asking the AI algorithm to predict a value, regression, or classify an input, classification.? So to keep it simple, the machine learning algorithm is simply finding a pattern that it matches in the data and it uses this pattern to look smart when predicting.? This is why AI is simply pattern-matching.

Unsupervised learning takes it to another level and asks the algorithm to group the data and then match the new data into one of those clusters.

(I am omitting Reinforcement Learning from this discussion)

When we pull back and look at this we simply see that a ton of data must be fed, processed (which is costly), a pattern determined, and ultimately that pattern used to predict the next value or group.?

So where is the general intelligence?

Those ideas are helpful BUT far from the general intelligence that humans are capable of exhibiting.? Yes we are limited in finding that pattern in 1000 dimensional data but our general intuition can not be modeled yet.

I recently listened to Andrew Ng, one of the leading AI researchers AND a person who has applied AI in his career, say that we are at least 30-50 years away from general intelligence. https://www.youtube.com/watch?v=5p248yoa3oE

30- 50 years… let us cut that in half 15-25 years.

So let us say in 20 years if we continue on this nosebleed linear improvement in AI we will reach GAI.? If then.

Being able to regurgitate facts is not intelligence by itself.? Being able to create is a sign of intelligence but it is not intelligence by itself.?

IMHO, a general AI system will be able to look at problems BY ITSELF, FIND AND use past data to understand the situation, COME UP with a creative solution, and lastly APPLY IT in the real world.? We as humans especially hold the key to apply it in the real world.?

The GAI would have to know where to apply it and without our instruction, it will not know.

I am not worried about GAI.? My PhD. Research 30 years ago was trying to extract rules from data so that better decisions could be made.? I wanted to prove that a person could learn from data without the help of an expert who can easily extract information and knowledge from that data.? Once in a while, I read a machine learning paper that talks about an idea I worked on 30 years ago as if it is the latest research.

So stop panicking… Enjoy the benefits of the new tools that will be created to help you.? Master how to use them and even create businesses that use them.?

Interestingly 30 years ago ML was thought of as the least promising area of research in AI.? It was too slow.? It took me 60 min to teach an algorithm when a point was inside a circle and when it was outside.? However, data availability and super-fast processors have cut that time to a fraction of a second…Moore’s Law.? I still think that the other schools of AI have not seen their moment in glory because of these data and computer processing improvements.? Some hybrid combinations will ultimately approximate the general intelligence these experts are attributing to a dump language prediction algorithm.

Maybe in 20 years?

My four cents...

?

??