What are the current best estimates as to what year artificial intelligence will be able to score 100 points on the Stanford Binet IQ test?
However according to Kurzweil it's late 20s:
Early in this decade, humanity will have the requisite hardware to emulate human intelligence within a $1000 personal computer, followed shortly by effective software models of human intelligence toward the middle of the decade: this will be enabled through the continuing exponential growth of brain-scanning technology, which is doubling in bandwidth, temporal and spatial resolution every year, and will be greatly amplified with nanotechnology, allowing us to have a detailed understanding of all the regions of the human brain and to aid in developing human-level machine intelligence by the end of this decade.
Not going into details of Stanford–Binet test, but just looking at wikipedia page it shows many subtests like knowledge, reasoning, verbal tests etc. Most of the efforts in the artificial intelligence today is directed into research of specific areas like computer vision, natural language processing, machine learning, but also combination of fields like implementation of self driving cars.
Within every field there are still other subfields and problems that are not solved yet. For example, development of human-like natural language processing (NLP) is necessary for intelligent agent to pass any verbal tests, or even non-verbal tests that requires processing of sentences of human language. Famous test that tests intelligence by asking questions in natural language and expects answers in the same form is Turing test. NLP still struggles with many (basic) human skills like listening, speaking, parsing and forming sentences. No one knows when we'll have system that can do these things as good as human. Since this system is crucial, but also far from human-like it's likely cause of delay in developing AI that passes intelligence test. Are these problems AI-hard? Do we need to develop strong AI to solve them?
You can look at speech and listening as interfaces used for expressing and affecting inner processes of human brain. Same goes for other senses like eyesight which is being approximated by computer vision. One could say that we only need to develop convincing mimics of human senses and incorporate them in one big system that will become first human-like AI. That is the minimum requirement. I doubt this will be achieved in this century.
What truly defines intelligence is brain activity. Since it's really complex and one artificial neuron is not equal to one neuron in brain, increase in computation power will not necessarily help achieving human-like AI. Also recognizing such system by mere intelligence test is questionable. For now it's only philosophical discussion but by the time we are able to design such machine I think we'll also have better understanding of human brain. Someone in 2100 might not read this answer on quantum computer with integrated AI OS powered from fusion reactor in his self-flying car, but will probably have many systems that help him in everyday tasks far more than we imagine today.