Schroedingers Cat nailed a couple of important points and I need to expand it further.

I believe the important roadblock in the creation of an absolutely intelligent machine is the fact that human brain is still not decrypted. There isn't a holistic cognitive model that describes how the brain functions, reacts or take decisions. Even in case such a model is inferred, it still cannot be put into real use since human decision is based on knowledge and the environment (popularly called culture). So a machine being rational is hard to prove since human themselves cannot prove their rationality.

Technically, there are three problems

1. Language - Almost all programs use one or the other form of Context Free Grammar. Humans always use context while speaking.
2. Computationally it isn't feasible to do a simulation model similar to brain. Just assume a choice of 2 items among a set of 10 items. You have a possible 2^10 combinations. Think of 100 conditions and your computer would go into limbo.
3. Most of the AI engines built today are a far cry from those proposed by their forefathers. You must have read the AI: A modern approach. They are mostly built for a particular domain and aren't absolutely intelligent

There's hope though. I am just in the same situation (coincidentally I have subscribed for both machine learning and natural language processing, though all courses are indefinitely postponed). I could suggest learning the following

1. Artificial Neural Networks
2. Natural Language Processing
3. Probabilistic Models
4. Unsupervised machine learning techniques
5. Pattern discovery

I also suggest a change in your original model. Instead of searching Google to find the answers, it would be better you download the stack exchange data dump and start analyzing the patterns since those answers are given by true humans.

Feel free to ask more questions. There are more things to be questioned than to be answered in this field

Schroedingers Cat nailed a couple of important points and I need to expand it further.

I believe the important roadblock in the creation of an absolutely intelligent machine is the fact that human brain is still not decrypted. There isn't a holistic cognitive model that describes how the brain functions, reacts or take decisions. Even in case such a model is inferred, it still cannot be put into real use since human decision is based on knowledge and the environment (popularly called culture). So a machine being rational is hard to prove since human themselves cannot prove their rationality.

Technically, there are three problems

1. Language - Almost all programs use one or the other form of Context Free Grammar. Humans always use context while speaking.
2. Computationally it isn't feasible to do a simulation model similar to brain. Just assume a choice of 2 items among a set of 10 items. You have a possible 2^10 combinations. Think of 100 conditions and your computer would go into limbo.
3. Most of the AI engines built today are a far cry from those proposed by their forefathers. You must have read the AI: A modern approach. They are mostly built for a particular domain and aren't absolutely intelligent

And regarding your robot searching Google for the answers, its already possible by using the Google API's to search for answers but it won't fulfill your requirement. Google uses the PageRank technology for its results, which is more of a machine learning technique and is still far from true semantic search.

There's hope though. I am just in the same situation (coincidentally I have subscribed for both machine learning and natural language processing, though all courses are indefinitely postponed). I could suggest learning the following

1. Artificial Neural Networks
2. Natural Language Processing
3. Probabilistic Models
4. Unsupervised machine learning techniques
5. Advanced pattern discovery

I also suggest a change in your original model. Instead of searching Google to find the answers, it would be better you download the stack exchange data dump and start analyzing the patterns since those answers are given by true humans.

Feel free to ask more questions. There are more things to be questioned than to be answered in this field.