This article below is in response to the above articles.
Recently there has been a lot of excitement about AI. I believe this is the same mistake that was made in early AI research. The early pioneers in the field mistook simple programming algorithms that made a computer look smart for real intelligence. Marvin Minsky laid bare this falsity in early 80s and the field collapsed for a decade.
The history of AI is filled with booms and busts. In the early days people used smart exhaustive algorithms like Mathematica to look like intelligence but this failed. In the early 90s rule based learning systems were the rage and companies seemed to be on the verge of “useful AI.” This busted after a couple years. In the 2000s we had machine intelligence which was applying statistical mathematical techniques to data to draw out intelligent conclusions and pattern recognition. Again this busted after a few years. In 2010 we have neural nets. This boom has lasted a little longer and maybe it has legs but I predict it will soon run out of steam. There are some really fundamental missing pieces to this puzzle as I document in a blog on what is CNN and Deep Learning and what are the limitations. Here is that blog.
A human does not examine the cloud (his brain) for sequences of data and produce a result as a computation. Researchers are mistaking being able to produce fast computers and smart algorithms for actual learning again. More important the press has taken the misleading statements of some high profile individuals and is hyping them as it is wont to do.
It is not clear how “aha” moments happen. Sometimes the brain makes “leaps” of cognition where it pieces together an unbelievable number of past inputs and directs itself to find the “answer” by “thinking” and the process somehow has moments where “ideas” pop in to the brain. These ideas are beyond our understanding. It requires a consciousness in which there is directed thought that perceives the context and without thinking consciously about things somehow produces an answer out of apparently nothing.
Sometimes this happens in a dream where conscious thought appears absent. Yet possibilities are enumerated and eliminated without consciously doing so. This could be a form of pattern recognition but it happens without conscious thought.
Another example of this is in dreams I sometimes find that the dreams demonstrate dramatic examples of having been pre-planned. Things in the dreams happen earlier that later turn out to be discovered in the dream to be essential to the later events in a way that would have taken a lot of thinking to plan in advance. I have sometimes written computer code seeming to know in advance how many lines a certain amount of code will take which required a substantial amount of thinking in advance which I never consciously did. This happens with proofs and numerous thinking exercises where the brain seems to operate below conscious thought producing the result without obvious computation.
One could say this is simply the machinery of our pattern matching algorithms but if so the sophistication and complexity of this is staggering. It is hard to imagine how to replicate this with any algorithmic process.
What you have described above is learning where the domain of learning is known in advance. We set the algorithm of how to operate in advance. The human computer can take any form of input it seems conceptual or physical and process it to produce new conceptual models. It could be a game or trying to understand the universe, probing abstract math, designing complex computer systems. We are so far from having computers able to even start to start to start on problems like this. I think you trivialize the brain.
As an example of how little we know there is still after 50 years of searching found the basic means the memory. We have several possible places memory could be stored. We have several ways it could be done. The fact we cannot even locate where the data is stored after 50 years is perplexing and surprising. We have a LONG way to go. I realize our ability to do this is growing exponentially but our progress is zilch in the face of the amazing growth of our knowledge of nature, our tools and understanding. We have tried to build “smart” machines for 50 years but the best we can do is have smarter programs which know how to process information and algorithms faster. Our algorithms are faster, better, but the basic problem of learning is completely wrong. The way we learn to recognize patterns in faces, speech, etc.. are totally different than the way humans apparently do it and these things are frequently good but when they make mistakes the mistakes are awful and stupid. Humans rarely make those mistakes.
The funny thing is I was very depressed in early in life thinking we would build smart computers. Although it was my passion to want to do it the thought of smart computers scared me a little and made me feel kind of like it might be dangerous or worry about a lot of big questions. The lack of progress allowed me to forget those negative thoughts. It’s become apparent this is WAY harder than we thought 40 years ago.