Minds, Machines, and Intelligence: A Conversation with Eric Horvitz

Behind the code delivers yet another great program.

12:30 “When I started getting into Decision science. I started getting interested in both decision making on uncertainty but also about what do we do with limited resoners if we built a limited system, how could it be optimized to do it the best it could. I had this sense that…um… and I still do very much that we can understand a lot about intelligence, biological intelligence, human intelligence if we push really hard on this notion of …um... trying to do the best one can with limited …um… computational recourses. Limited time for example. Um… that this would explain the pressures of competition and minds would etch out …a, brains with certain properties that would do their best under restraints of various kinds.”

How does a limited time restraint improve decision making? Have there been minds culled under study that do better than others with limited computational recourses? What other resources are there that are tracked besides time restraints? I would like to see the data. 

Can some one give an explanation to describe what directed acyclic graphs are? Other than what is already written on a wiki. Thanks

Thx for posting this Charles, it was very enjoyable and thought provoking.

I think once a lot of the technical, decision making and intelligence issues relating to machines and robots are resolved, we will have to grapple with the philosophical and ethical issues that will affect mankind going forward.

I like how 'think' is recursively defined   "meaning: generate thoughts autonomously"  What constitutes is a thought?  Also, how would you know if something was perceiving vs. simply registering (I see neuron C7 firing) seeing as perceiving is a purely subjective phenomena.  If it is purely subjective, then that's the end of it, you'll never "know" unless you are that subject.  Also, just because something "thinks" doesn't imply it has or will ever have emotions.  Something could be intelligent (say good at recognizing patterns and predicting what's next) but it doesn't have to "feel" one way or another about it. Actually I don't think we all know how emotions impact decision making.  That's some pretty cutting edge research that's currently ongoing. There's a great book called "How We Decide" that explores this concept.  Also, another book, "On Intelligence" is a great one to get you thinking about the 'intelligence algorithm' and how it is distinct from emotions.  It is quite possible that emotions are the auto-generated proto-thoughts of our reptialian brain...

I wouldn’t be surprised if artificial intelligence some day surpasses the intelligence of humans. Already some components of computers are out performing the human mind. Consider some of the large data bases and the ability to retrieve the information stored on them.

We maybe in the same ocean as Mr. Horvitz but he is in a different league than us. I try to read some of the papers he has published. They go way over my head. http://research.microsoft.com/en-us/um/people/horvitz/uai2k1.htm Wish I could understand better the subject of ‘Reasoning under limited computational resources’. The subject and level of knowledge needed to correspond about such things is a little intimidating. But I like that Mr. Horvitz shared the story about listening into neurons and trying to decipher the code behind intelligence that way. I think it displays his curiosity. Something I think we all need.

When I think, I realize I am at a disadvantage due to a lack of knowledge or lack of information regarding unknown variables. It’s amazing that I can come to any kind of conclusion at all when making a decision. Some people have a hard time walking down the street because they have so much information to process and yet not enough information to complete their inquiry. The majority of us make decisions when we relatively have very little information on the subject. We make our decisions in ignorance. Our minds make the best decisions possible based on the information at hand and then roles the dice. 

I think it may be just as advantages to figure out how to harness the human minds ability to deal with uncertainty as it is to give those same abilities to computers.

I do think we can make machines that reflect our ability to think. Abstractly: if the equations they make are able to be matched against each other over and over again. Like the rolling of dice until it strikes upon the most logically feasible result. I may have just over simplified a very challenging question. Kind of like when I was a kid and I thought all you needed to make a computer game was a card board box and a picture drawn on paper.

Great, Thank you!

 

Great video! @Minute 13:30 there are interesting insights on AI progress in mid 80s in healthcare & machine disgnosis plus the collapse of AI expectations due to requirements of modularity, maintainability & extensibility issues with new facts introduction (that made AI systems fragile).