What is next big thing? Mine bet it is an artificial intelligence.
What businesses are working on the frontier of the AI technology? It is companies that deal with recognition problems, like OCR, speech/image recognition etc.
With thanks to smartphones speech recognition market is already worth billions $ a year. And leaders in the field are Google and Microsoft. They are big, but there is one company that much smaller and ASR (automatic speech recognition) is their core business: Nuance Communications(NASDAQ:NUAN). They are already profitable and clients include Apple and Samsung. NUAN revenue is already well over a billion a year.
So what is AI? Please do not imagine talking heads like Siri or IBM WATSON that wan Jeopardy TV show. The fist real application of the technology is industrial automation. Think about robotic arms. They could do any work that human could do if you place human operator with remote control and camera vision system. But to program arm to deal with the task in the environment that have slight variations to it is virtually impossible without algorithms that are able to generalize and adjust arm movement according to the data from sensors that it never encountered before.
Think about thousands who work on Tesla Factory assembly. Most of them basically are doing repetitive tasks. But most of those tasks are virtually impossible to automate using current technology. Think about those hundreds of thousands employed by McDonald's who are cooking ~20 "dishes" menu making minimum wage... Etc.
So what the problem and how ASR could come into picture? Current approach to "teach" robotic arm to do the job is to guide it throw predefined trajectory and apply end of arm tool/tools when needed. Plus throw couple checks to make sure object arm working with is in place and correctly positioned. Current cutting edge tech: do the above but may be check for part variability(no two body in white have same dimension for example, there are few mm difference here and there) and run simple approximation algorithm that calculate some trajectory adjustments. But the more parameters robot programmer trying to cope with - the number of possible outcomes literally explodes. With two binary inputs you have to deal with 4 situations. With 30 binary inputs - with over a billion... And when parameters start to affect each other it is simply impossible to deal with them using conventional programming technics.
When computers deal with speech or scanned characters - they are given data that they never seen before. And each time they try to figure out what word been said or what character been printed. Until very recently most successful approaches were based on combination of statistical methods(Gaussian mixture models etc) with hidden Markov models(HMM). With very poor performance that could not be tolerated in industrial environment. But in a last 12 months MSFT and GOOG reported that they started to use neural networks for crucial part of speech recognition. With statistical methods one have to provide thousands of examples of how each word said. Big Data. But some most advanced NN recently showed ability to strongly generalize from what they have learned already. So after you teach NN first 100 words, you could teach it next 100 words providing it with fewer examples. In the nutshell this should basically allow more human-like programming of robots: you show human worker what he should do on assembly line couple times and he is ready to work... It might even allow less stupid talking heads on your phone - but transition from ASR or recognition algorithms in general to strong AI or AGI is not a trivial one. I think it best explained in Ray Kurzweil book "How to create a mind".
But do not make mistake: not all NN are born equal. It is laughable when another media reports about another NN with whatever billions or trillions neurons in it. NN advances are not about numbers of neurons, they are about math behind those networks. And math is mindblowing. If you not afraid of math, you could try to read
this for example. The company those guys created were bought by Google btw.
Bottom line. Companies that work on ASR actually working on the frontier of AI research. If they nail down good algorithm, NN or not, doesn't really matter, market for potential applications for technology is HUGE. Industrial automation and fast food automation is just a tip of an iceberg. They will have a hammer in their hand and everything would look like a nail to them. And NUAN very well positioned to benefit from such advances more, they are smaller and more agile. But the good thing about this bet, is that even if AI wont happened, NUAN still got a business that already bring them billion plus $$$ in revenue each year.
I'm long.
