Comments 3
anything to do with intelligence and wasn't just a search/optimization problem
Intelligence IS search/optimisation problem. 99% of it at least.
It is not like competition between human and ML. It is all about solving problems. You right about trasfer learning — it is necessary for solve complex problems. And it is fast growing area in ML.
I definitely think that any physical process can be framed as optimization. Optimization not necessarily generates knowledge useful for other problems.
I like Schmidhuber's definition of intelligence as compression efficacy. In case you are not familiar, according to his theory of artificial curiosity the more one can compress data, the more one can understand it. Example of increase in compression efficacy: 1. raw observations of planetary orbits 2. geocentric Ptolemaic epicycles 3. heliocentric ellipses 4. Newtonian mechanics 5. general relativity 6.? Under this view, compression of data is understanding, improvement of compressor is learning, progress of improvement is intrinsic reward. To learn as fast as possible about a piece of data, one should decrease as rapidly as possible the number of bits one need to compress that data. If one can choose which data to observe or create, one should interact with environment in a way to obtain data that maximizes the decrease in bits — the compression progress — of everything already known. Here, understanding, learning and interacting with environment are framed as optimization problems.
In case of living beings intelligence might be less like optimization/search problems from CS and more like optimization of functionals from physics.
Intelligence is more about adapting/transfering old knowledge to new task than it is about compressing experience into heuristics to predict outcome
I would say, intelligence is about getting that knowledge. From scratch, not from someone other intelligent who already knows it. Then it is used as kind of input information and this gets an ability to apply it to new tasks.
Playing Go can't be considered a non-standard problem for AlphaGo after playing 30 million games.
Good point)
Biologists define intelligence as ability to find non-standard solutions for non-standard problems and distinguish it from reflexes and instincts defined as standard solutions for standard problems.
I think, "non-standardness" is not defining characteristic. Reading (as getting information from text) or translating are quite standard tasks but they require capabilities of intelligence.
minimize the difference between its models and reality.
Perception and cognition are both attempts to create accurate models that match the world, thus minimizing free energy.
I have same opinion, but I consider it not in terms of free energy but in terms of infromation.
I have the article about this on Medium (on Habr in Russian).
In particular:
"Intelligence is an ability of an information system to build a model of environment by input information."
"Curiosity is a desire to know details of some model element."
Progress and hype in AI research