A while back I was giving a talk in Boston on the various challenges that remain in our quest for the development of systems that truly understand ordinary spoken (natural) language. At some point in the talk I reported on something that I have experienced firsthand: I often meet and converse with many so-called experts in Natural Language Processing (NLP). Experts that, as it has often turned out, have never heard of Russell, Frege, Carnap, Quine, Lewis, Kripke, Montague, etc., or have not heard of metonymy, intension, scope and reference resolution, and other challenges in the computational treatment of natural language. What got a chuckle was my remark that this is exactly like meeting a specialized physicist who never heard of Newton or Einstein, or of the laws of thermodynamics (and it is, by the way, exactly the same).
So what has happened? How could we have leading labs (both in industry and academia) graduating or nurturing so-called experts in language processing—‘experts’ that are indifferent to a couple of centuries of fundamental work by some of the most penetrating minds in logic, semantics, and formal languages? It seems now that one of the most difficult problems in computing science (i.e., NLU) is thought of as a ‘data’ problem, and thus it is a problem that can be easily tackled by pulling some machine learning library, downloading lots of data, training your ‘deep’ network on that ‘big’ data, and viola—you are another step closer to passing the Turing Test (or better yet, to passing the Winograd Schema Challenge!).
This is harmful to the field. Computing science is, after all (or at least in big part), a science. And so it is already a Data Science—thank you; not to mention that it is also an Information Science, and, ultimately, a Knowledge and a Cognitive Science.
The reason this trend is harmful to the field is that it turns computing science into computer and systems programming, while the latter are just our tools as computer scientists. While systems and software engineering are noble causes—and very stimulating and challenging at that—the science must also progress and it won’t if we ignore it. Solutions to major problems should not all be approached by finding some ‘approximate’ solution. Computer science has always been partly a science and partly an engineering discipline, so while many can do both, some of us should surely keep working on the science.
Walid Saba is Principal AI Scientist at Astound.ai, where he works on Conversational Agents technology. Before Astound, he co-founded and was the CTO of Klangoo. He has published over 35 articles in AI and NLP, including an award-winning paper at KI-2008.