Abstract

Recent progress in artificial intelligence has led some to speculate that machine intelligence may soon match or surpass human intelligence. We argue that this understanding of intelligence is flawed. While physical machines are designed by humans to simulate human rule-following behaviour, the issue of whether human abilities can be emulated is not well-defined. We outline a series of obstacles that stand in the way of formalizing emulation, and show that even a simple, well-defined function cannot be decided in practice. In light of this, we suggest that the debate on intelligence should be shifted from emulation to simulation, addressing, for example, how useful machines can be at particular tasks, rather than deliberating over the nebulous concept of general intelligence.