Chapter 2 – Think like a Machine
1. The concept of using an artificial neuron was discovered in 1990. (Yes | No)
The answer is no. Warren McCulloch and Walter Pitts invented the first neuron and published a paper in 1943. Legend has it that at age 12 years in 1935, Walter Pitts, a poor child living in a bad neighborhood, was chased by bullies and sought refuge in a library. There he discovered Principia Mathematica by Bertrand Russell and Alfred Whitehead. Anyway, not only did he find mistakes in the reasoning, but also he sent a letter to Bertrand Russell! From then on, Walter was noted for his genius in mathematics. With Waren McCulloch, another genius, they invented the first neuron. It seems simple. But it's the result of sleepless nights. Just as the invention of the wheel appears simple, nothing better has been found to this day. This concept of a the neuron is the wheel of artificial intelligence.
2. Does a McCulloch-Pitts neuron require a threshold? (Yes | No)
The answer is yes. Adding up weights does not mean much if you do not have something to measure the value. It took months of work for McCulloch and Pitt to put this together. At first, time was in the equation, just like it is in our brain. But then, like Joseph Fourier (1768-1830), they found cycles that repeated themselves—periods that did not require much more than that neuron.
3. A logistic sigmoid activation function makes the sum of the weights larger. (Yes | No)
The answer is no. The whole point is to reduce the sums when necessary to have comparable numbers to work with.
4. A McCulloch-Pitts neuron sums the weights of its inputs. (Yes | No)
The answer is yes. It's only when you sum the weights that they make sense.
5. A logistic sigmoid function is a log10 operation? (Yes | No)
The answer is no. The sigmoid function is based on Euler's number, e, a constant that is equal to 2.71828. This number produces a natural logarithm. Leonhard Euler (1707-1783) discovered this in the 18th century with a quill—no scientific calculator or computer! Did you notice that the main mathematical functions used in artificial intelligence run far back in history? This aspect of the hype around what we think we have found now but has existed for decades, and sometimes centuries, will be dealt with in the following chapters.
6. A Logistic Softmax is not necessary if a logistic sigmoid function is applied to a vector. (Yes | No)
The answer is no. Calculating the sum of several numbers of a vector and then dividing each number by that sum gives a view of the proportions involved. It is a precious tool to keep in mind.
7. A probability is a value between -1 and 1. (Yes | No)
The answer is no. Probabilities lie between 0 and 1.
