The Deep History of Deep Learning… Part IV
LISPs, Languages, Dialects, and Battles
When we last left our story, the AI winter had begun and the music had died. Two years before the music had died, Fortran had been invented. The two facts are probably not related, nor are either very important to our story. However a year after Fortran and a year before ‘the music died’, LISP was invented. This will be very important to our story.
LISP was one of the earliest programming languages, second only to Fortran which we will not mention again. It stands for LISt Processing language which is both important and boring, all at the same time. It has nothing to do with speech impediments, which is both boring and unimportant.
LISP, unlike older Perceptrons, can perform an XOR. This has nothing to do with exorcisms or decepticons, but really should. This is just a long, but hopefully interesting way to note that after many dialects (nothing to do with Dr. Who), LISP ended the AI winter in 1980. Dialects are just incarnations of LISP. By 1980, Standard LISP was invented and suddenly the AI hype was back! This also represents an answer to some odd online Death Battle fan art.
LISP is also symbolic. It is code built from code. It has garbage collectors. And any number of other traits that make it an excellent language to build AI in. Remember — AI is layered, iterative, and adaptive. When code can be used to build code — iterative layers and adaptation follow.
Early AI saw success through connectionism, now LISP provided the upper hand to computationalism. Connections gave way to symbols. Unfortunately, for however hard the guys at MIT tried (and they tried quite hard), winter would return. It took less than a decade.
Part of the problem was funding. With very little to show in the way of returns, Artificial Intelligence was still heavily funded by DARPA. In 1986, Jack Schwarz, a Yale alum, took over the reigns of DARPA. In 1987, he wrote a book titled The Limits of Artificial Intelligence. He then promptly defunded it. Well, that would certainly limit things. By 1989, he was gone. In an interesting and not completely unimportant correlation, so was the Soviet Union.
DARPA funding had been at the heart of the first AI winter. Well actually, it was ARPA at that time. This time however, the private sector would play a role as well. The 80’s had seen a market bubble form in LISP Machine companies. They were all the rage, much to the delight of all those MIT guys who believed “everyone” could and should learn LISP. As perceived infinite markets often due, the LISP market collapsed in 1987. Two years later, so did the Japanese Stock Market.
With no public or private funding, no threat from the Soviet Union, and no more competition from Japan — the second AI winter began. In another interesting but likely not so important coincidence, Japan’s lost decade had also begun. In our final article, part V, we will examine the final few decades of Deep Learning. Stay tuned.
