What is the most common limitation for creating breakthroughs in AI?.
Answers
Answered by
0
AI is fundamentally unlimited when compared to human intelligence in theory, as far as we know. In practice, there are still many problems that artificial intelligence cannot perform for some of the reasons mentioned below.
Machine learning is limited by data, model complexity, and compute time.
There are three general limitations to ML models:
Data: you don’t have enough data (or, if meaningful patterns in your data don’t exist). Sometimes this can be remedied by introduction of additional data, and sometimes not. For example, if you are trying to model a historical event, you must use limited historical data.
Model complexity: the space of models which you can learn is to limited to fully model the relationships in the data. Increasing the model complexity is arbitrarily possible with representation learning and neural networks.
Compute time: the available computation resources is not feasible to accurately bring down the training error to reasonable upper bounds. This is limited to the computation resources available, like GPU cycles, and RAM.
Computation is limited by the laws of thermodynamics
Like all other matter in the universe, AI relies on computation, just as organisms rely on biological computation.
There is an energy cost of running a computation. If the energy available is not sufficient to perform the learning operation, then you’re out of luck.
Machine learning is limited by data, model complexity, and compute time.
There are three general limitations to ML models:
Data: you don’t have enough data (or, if meaningful patterns in your data don’t exist). Sometimes this can be remedied by introduction of additional data, and sometimes not. For example, if you are trying to model a historical event, you must use limited historical data.
Model complexity: the space of models which you can learn is to limited to fully model the relationships in the data. Increasing the model complexity is arbitrarily possible with representation learning and neural networks.
Compute time: the available computation resources is not feasible to accurately bring down the training error to reasonable upper bounds. This is limited to the computation resources available, like GPU cycles, and RAM.
Computation is limited by the laws of thermodynamics
Like all other matter in the universe, AI relies on computation, just as organisms rely on biological computation.
There is an energy cost of running a computation. If the energy available is not sufficient to perform the learning operation, then you’re out of luck.
Similar questions