Last week, the IgnitEd board hosted an Artificial Intelligence talk at Stanford’s Spiker Engineering and Applied Sciences Center with Remi El-Ouzzane, now a Vice President and General Manager at Intel Corporation. Intel purchased Remi’s Movidius last fall. The tag line for Movidius is, “Visual sensing for the internet of things.”
Remi answered many questions for the mixed audience of teachers, entrepreneurs, engineers, and Silicon Valley managers; and thrilled us with stories of what may be coming next from the Silicon Valley. Spoiler: self-driving cars may be a red herring, but we might want to reconsider the emphasis we are placing on teaching kids to code – the machines may just be able to do the coding for themselves pretty soon.
Of particular interest to me was Remi’s discussion of deep learning. It is a bit unnerving how the language we have developed for neuroscience is now being used to describe the structures for artificial intelligence algorithms that seem to be capable of thinking for themselves. Deep learning technology is a subset of machine learning and is used in some natural language processing applications that give us the creepy stuff like Google’s “Okay Google,” voice to text and search capability.
Deep learning is sometimes referred to as ‘neural networks,” and the magic (or sorcery – depending on your perspective) comes from a nonlinear process of action that results in outputs whose genesis can be impossible to trace. In other words, the engineers themselves sometimes do not know how the technology is getting the right answer. Did anyone else just say, “Skynet?” Applications for this tech are already in our daily lives (at least in mine). Some customer success robots, our favorite voice recognition tech, end even some new drug discovery is now relying on deep learning.
In education circles, deeper learning is a different concept – a non-technical one – but not entirely unrelated. In response to the No Child Left Behind, decade-long race to the bottom in which preparation for standardized bubble tests forced an overemphasis on rote learning, education leaders have reacted in recent years by emphasizing (again) the need for education that encourages young learners to employ logic, examine multiple perspectives, and evaluate sources. Deeper learning is a brand used as a counter to policy that encouraged procedure and memorization over creativity and critical evaluation.
Deeper learning is not a higher degree of deep learning. However, deeper learning and deep learning seem to have an interesting common root. Both result in outcomes that we recognize as superior, and both thwart a linear analysis that could identify the source of the magic.
The Deepest Learning
Educators in the room listened intently and patiently to Remi for about forty-five minutes, and then the room boiled over. A flurry of comments about the value of human judgment and the learner-teacher relationship bubbled up. They all begged the question, which someone finally spoke, “What will teaching look like in a world with deep learning?”
Remi stepped carefully here. He reassured the audience that we are not talking about tomorrow, and there will always be value in, and perhaps even increasing value in the teacher-learner relationship in a world with neural network algorithms and voice-recognition access to the internet. He did argue the likelihood that some jobs may be replaced by A.I. in the ten year term; paralegal, telemarketer, and perhaps truck driver among them.
The deepest learning of the night came for me when one of the teachers asked Remi, “What would you recommend for my soon to graduate students? What should they study in light of what you just told us about computers programming themselves?”
Remi’s response, “I think there will be a growing significance for the liberal arts.”