dpullum Loc: Tampa Florida

Robg, I stand corrected. My mistake is that I was narrow in my thinking to consider only the application to photography AI. I recall when it started every company suddenly had to label things as AI.

Reading [ref below] about learning and deep learning I came across "Transfer Learning." "The first thing to remember here is that, transfer learning, is not a new concept which is very specific to deep learning" In my life of solving industrial science [commonly incorrectly referred to as engineering] problems I recall when starting out, I did well, but as time progressed I have a backlog of knowledge not necessarily specific to the problem at hand, but the concept and/or data from a previous problem could be modified and transferred to help solve the newer problem. Often the knowledge-bits were from multi-problems solving from the past. My phone linked me to experts who drew from their neuronetworks [brain] to add to my problem solution.

Explaining to others is helpful. I have a female friend in England who is absolutely not technical, she will ask me to explain what construction or repair or creation I am working on. [She wants details so she may learn, latest was changing her oven heating elements.] Having to explain things to her about my problems solving, I become more inventive and mind visualizing and as a result solve the problem as I speak.

Often in my real-life career as a technician manager, I solved problems in my sleep that I could not solve otherwise. Awake I was confined by the practical. My math professor teaching "modern algebra and matrices" said go to bed with the problem and wake at 2 AM and write down the solution and then rest well. I thought Advanced Calculus was bad... that course was worse.
Solving problems in your sleep:
https://www.inc.com/the-muse/albert-einstein-thomas-edison-your-half-asleep-brain-can-solve-problems-better.html

My first company purchased IBM PC, 198?, ran at 10 meg hz and had 128 meg of memory!!! In the beginning, floppies were big and held little. So at that level, we could do very little beyond basic programming with DOS. Deep-leaning was limited to supercomputers. Applications and computer time were not squandered on photography... which was very wet at that time. Now we have digital photos and the 0 & 1s can be worked on by deep learning.

If you have problems sleeping then read this practical "hands-on guide to transfer learning" !!!
https://towardsdatascience.com/a-comprehensive-hands-on-guide-to-transfer-learning-with-real-world-applications-in-deep-learning-212bf3b2f27a

Drbobcameraguy Loc: Eaton Ohio

I understand that it came up with a new proof but it found that solution by trying every possible combination of the data it had been programmed with. It didn't actually come up with an idea. We are a long way from true Al learning. As in I program a robot with words and meaning and we can have an actual conversation. I'm looking for the article I just read a couple days ago. It was very interesting and informative. Was not knocking Al or learning just explaining that the neural network capibilities we have now are a long way off of true learning and also presented the project they are working on now to try to reach the goal. It was a very informative and interesting article and I thought I saved it but apparently not. As soon as I find it I'll send you the link.

Robg

I’ll begin by accepting, for now, your premise, that Logic Theorist (LT) was essentially a search algorithm. It seems to me that your conclusion, that this is not human creativity, may be non sequitur. It is quite possible that search is a large, even major, part of intelligence and creativity. First, there is the statement, purportedly Edison’s, that “invention was 99% perspiration and 1% inspiration”. Similarly, many of the contestants on Shark Tank mention the many product iterations they went through to come up with a final recipe or consumer gadget. Those statements all imply that some form of search is part of the creative process.

Earlier in this thread there were several posts that talked about how some people solve problems while they sleep, waking up to an “aha” moment. This is not a universal phenomenon. There are some geniuses who have reported that kind of thinking, but there are others who have said they don’t have that experience. A very informative discussion of genius creativity (and the role of sleep) can be found in a book “The Psychology of Invention in the Mathematical Field” by Jacques Hadamard published in 1944 ( http://worrydream.com/refs/Hadamard%20-%20The%20psychology%20of%20invention%20in%20the%20mathematical%20field.pdf)

A reasonable hypothesis suggests itself – when we sleep our brain is far less busy processing sensory stimuli. Perhaps this frees it to carry out more extensive searches, leading to the aha moment.

Moving to the larger question, can AI be creative? For example, can AI compose music? Well it has! Check out https://www.youtube.com/watch?v=lv9W7qrYhbk which plays two short pieces, one composed by Bach the other by an AI program that learned from Bach's compositions how to compose like Bach. It turns out that AI musical composition and performance is already a rich area with lots of interesting results. Ditto for AI creations in other fine arts.

Returning, finally, to the premise that LT was a search algorithm. Since LT employed backtracking, to that extent I would have to agree as backtracking implies trying out different branches in a search tree. However, it most certainly did not try “every possible combination.” That would not have been possible with the computers of the late 1950’s! So there are other aspects to it, including how the branches in the search tree are generated and also the heuristics employed by LT to eliminate many of the “possible combinations”. The heuristics were intentionally modeled by Newell et al to mimic what they thought might be the human problem-solving process, namely, to hypothesize intermediate steps that, if successful, could be incorporated into a proof. For example, let’s say I’m trying to prove assertion A from a set of axioms, X. The proof, if I can find it, might be to generate a sequence of statements, each following from the preceding statement by applying one or more of the axioms in X. So my proof would be in the form of a series of statements B, C, D, E… A, where the starting point B is one of the axioms in X, C is proved by applying one of the axioms in X to B, etc., and the final statement is A, thereby proving A. LT uses this framework by generating a likely intermediate statement, say D, and then looking for, first, a proof that will prove A from D and then, second, looking for a proof that will prove D from B. The two proofs together, then prove A.

By the way, I hope you can find the article you referred to as I would be very interested in reading it.

Drbobcameraguy Loc: Eaton Ohio

I do understand that humans basically come up with solutions based on previous experiences. Basically we are programmed at birth and continually until we die. I am one of those people that never lets go of a problem until I find a solution. Lol. I have had several ah ha moments waking or just doing something completely different than the problem I was working on for days a month ago. Lol. Can't explain it but it's funny to me and my coworkers and family. I have been known to change the subject in the middle of a conversation with the solution to a problem from who knows how long ago. Lol ok enough about my quirk's. I still haven't found the exact article and I'm still looking. The link I have posted is similar but does not contain the information about how true artificial intelligence is being attempted and the solution to the shortcomings at this time. I started learning electronics in 1976. Have never stopped. Lol. It's a passion. Quantum physics has been my toy for the past few years. I'm very ignorant about quantum physics but keep learning. I have been saying for over ten years that our standard model of physics is flawed in some aspect and that is being confirmed over the last 2 years. I just hope that I'm still alive when the correction is found. Anyway I will keep looking. I'll find it eventually. I'll pm the link to you if nothing else. It was a very good read especially the possible solution for true artificial intelligence. The points you made are true. Is not our intelligence basically a set of data that we use to come to a decision? Yes I believe so. The computer is ahead of us in that respect because it doesn't allow emotions or feelings to cloud the sifting through the data. It's an interesting time. I wish I was 30 again. Lol

https://blogs.scientificamerican.com/observations/ai-doesnt-actually-exist-yet/

JBRIII

The only question I have is that if there are no emotions then what would be the driving force to do anything. At best, it would be a program driven by programming alone to do whatever. Seems to me that that would mean no sentience if nothing else. Maybe, as many/some, contend sentience is an allusion, but it at least a comforting one.

I agree with you on learning new things and aging, but I doubt that either of us would be anywhere but where we are in that regard in additional 60 years even if we had them. We now know much more about things today, personally and as a species, than we did even 30 yrs ago, and all we really know now is how much more we don't know.

I was just talking to my wife about the fact that we have 4 close friends and between us we have some 200+ person years in science and computers, 6 BS, 2 MS? and 1 PhD degree in fields from biochemistry, chemistry physics and computer science engineering, yet ask a question about copying a streaming program from a TV provider or can you charge an android device and attach an external device at the same time and no one knows. Way too much to learn and never ever enough time, but at least I can avoid being bored.

Drbobcameraguy Loc: Eaton Ohio

I hear you in the at least I won't be bored lol. That's why I have started my quantum physics adventure. If I live another hundred years I most likely won't know all the answers about the field. Lol