Engineering AI: Perspectives – I think one of the most important things that I bring to people that surprises them about AI is the idea that it’s not magic. AI is not magic. I often have people bring data and say “can you do AI on this data?” And my answer is, “perhaps, but what do you want to know?” Because all AI does is extract information that’s already inside the data, and it uses the things that we do understand as engineers and scientists. It uses math; it uses statistics.
And it finds things that maybe aren’t easy for us to see, but they’re there. AI doesn’t create information out of nothing. It doesn’t magically find things that are unfindable. Instead it just uses the patience of computers to see patterns that are hard for us to see in this high-dimensional space of the data, of the complex data that we receive. Automated data-driven systems increasingly make decisions that affect our lives.
These systems often make use of machine learning and other artificial intelligence. And they make decisions like which advertisements or recommendations should be shown to which individuals? Does a set of medical images indicate that an individual has cancer? How likely is a criminal offender to commit a crime again? Yet these systems behave like black boxes. We do not know why certain decisions are being made.
My research seeks to enable data-driven systems to be accountable for their decisions. I envision a future in which, not only are these systems making predictions, but they are augmented with accounting tools that can explain why a specific decision was made, or why the system exhibits a systematic pattern in its decision making. AI is essentially deep layer neural network nowadays.
They’ve been applied in many, many different applications, and their getting lots of attention. But neural network research has started in maybe late 1980s, including lots of research done right here at Carnegie Mellon University. But what is different now is the convergence of three technology advances. Computers have become much faster, including things like graphical processing units.
Storage and bandwidth have become much better. And also it has become much easier to collect lots of data. When you put these three things together, now the neural network algorithms can not only compute faster, they have more data that they can learn from, and also you can store this data and retrieve it as needed. So this is what makes AI algorithms as powerful as they are today.
reference – Engineering AI: Perspectives
Web enthusiast. Thinker. Evil coffeeaholic. Food specialist. Reader. Twitter fanatic. Music maven. AI and Machine Learning!