Episode 3 AI Never Sleeps! – Real-world business discussion about QRC AI deployments
Throughout these podcasts, we will focus on hands-on implementation of AI, deep learning, and machine learning with regard to (wrt) the actual engineering of ML solutions into real world enterprises and networks.
Wednesday I had the pleasure of meeting with a group of DC area C-level execs who are looking for some deep learning classifiers to add to their product lines. I have had some minor dealings with them in the past, so I know these folks are the real deal among senior executives who are both AI-savvy and know how to handle disruptive technology transitions for large enterprises. The CTO is sharp as a whip (Genius IQ, thinks out of the box, and knows BS and fluff when he hears it). Like many decision makers whose time is critical, he employs a very abrasive delivery as a way to cut to the chase.
Meeting basically went like this: (CTO:) Lloyd thanks for fighting DC traffic, welcome. So why MachineLrn and not the PhD body shops we know and love? Those guys have been working on AI for 20 years and have a large staff of PhD and PhD candidate students.
My Answer: No debating, they are a big shop of smart AI people. However, they have been heads down working on their own custom frameworks, hand coded predictive models, custom analytics/ analytic engines, and white papers tons and tons of whitepapers. (And usually training on made-up data corpus and cherry picked labelled data).
Based on my systems engineering experience, the most frequent question I get from these AI SMEs is “where can we get a representative corpus of data which reflects the real world within your enterprise?”
Essentially, I recommend leveraging the world’s largest internet company and the authority in artificial intelligence. Those other guys will fly in some PhDs and spend tons of billable time tweaking their inferior product. The end result is to slow roll AI deployment, wander down blind alleys are your billable expense, and ultimately play catchup with commercial AI developments.
BL: would you like to start transforming your company with QRC AI deployments, or to kick off an open ended study and pilots which make great lab demos but fail to smoothly integrate?
Why put this out there? This is real world! I didn’t plan this episode! I didn’t expect a call two days ago for a DC meetup. I didn’t expect such a polarizing question. Such is the urgency and angst among the CTOs that I should have expected it. They know from past sad experiences what doesn’t work, but they also appreciate that AI is here now. If you read my article on “Why AI will revolutionize computer programming”, it will be crystal clear that AI will completely transform the old concepts of MMI, while also create new knowledge by deep learning from enormous data carpi far too large for human processing.
If I can provide my own lessons- learned to someone else who can push through and get traction it was totally worth it. If you have better ideas (and you probably will if you read this far), please share them here.
I’m not anti-academic. Out of decades of brilliant fundamental research into adaptive filtering, later sigmoid and back prop for generic problem solving algorithms the researchers have delivered data science tools, algorithms and computer science breakthoughs. These developments emerge just in time to benefit from the confluence of data center cheap processing, specialize GPU/TPU processing, and mass (global) data storage.
Also, now is now the right time to start an AI QRC pilot, leveraging the enormous investment of current COTS AI technology. Do not wait for a whitepaper or your competition will get a head start.
