AI Isn’t Actually Intelligent

Thinking it is has lead to people dying

Shefali O'Hara

--

Photo by Steve Johnson on Unsplash

When I was at MIT in the 1980s, I worked at the MIT AI lab for a semester as an undergraduate research assistant through UROP (undergraduate research opportunities program). I later switched to the LCS (Laboratory for Computer Science) where I designed pattern recognition based code.

A few years after graduation, I bought a software package that allowed me to design machine learning experiments using C++. It was a lot of fun! Unfortunately I no longer have the software since it was on floppy disks…

Recently AI seems to be having a resurgence, which is great, but people seem to have unrealistic expectations.

What people don’t understand — an AI is only as good as its code.

What is AI and how does it work?

AI stands for artificial intelligence. The Turing Test, developed by Alan Turing in the 1950s, is used to measure computer intelligence. Turing is considered one of the founding fathers of AI, along with Marvin Minsky, John McCarthy, Allen Newell, and Herbert A. Simon.

The term refers to the simulation of human intelligence.

Having worked with complex simulations in the late 1980s and early 1990s, I can tell you that a simulation is only as good as the inputs you provide it. This includes AI.

When I designed simple neural networks, I experimented with teaching programs to differentiate between different objects based on specific criteria. The first, and simplest, program I wrote was one that differentiated between apples and oranges.

The two criteria I used were size and color.

The problem of course is that this was all information I fed the program. What if the apple was not red, but yellow or green? What if the orange was not perfectly round?

This is the problem with AI.

To teach the AI, you need to provide feedback. At first, the AI may produce many incorrect results, but over time it will be taught to become increasingly accurate.

In order to faciliate this process, designers must acquire and provide a large enough data set. They also need to create…

--

--

Shefali O'Hara

Cancer survivor, writer, engineer. BSEE from MIT, MSEE, and MA in history. Love nature, animals, books, art, and interesting discussions.