Total Recall: Human vs Machine Learning

Source: Deep Learning on Medium

Total Recall: Human vs Machine Learning

I recall sitting in class enthralled with the make-up of the human anatomy and physiology. The instructor was clever in her delivery and despite her thick Argentinian accent her message was clear. I was completely engaged. But it wasn’t until we covered the human brain that I got goosebumps. There was obviously an above average interest in this area. Then it happened again when we covered the nervous system. Needless to say I did well in that class. Fast forward, I entered the nursing field as a coachable nurse assistant but felt unfulfilled. So why did I enjoy human anatomy & physiology?

Oh that’s right, it was the brain and nervous system. Let’s say I wasn’t human but a specialized form of machine learning called deep learning. That meaningful piece of data- the sections favored- could’ve been learned and retrieved rather quickly using images in my brain. Although, I would’ve had to input a large volume of accurate data, clean it and train it to recognize those images. Could deep learning recommend what careers I’d likely be interested in? Yes. Could it determine if the brain and nervous system sections were favored? Most likely. But could it determine why the brain and nervous system was favored? No, and that’s where the line is drawn between machine and human intelligence. Deep learning goes deep to teach itself from its own data but doesn’t understand the underlying meaning of it yet. Whereas, a human during a brief conversation could surmise the why behind those sections being favored quite easily and might say, “Oh, you’re interested in how people are wired.”

Artificial intelligence requires data to make informed decisions. And so do we. Yet human data come in the form of memories stored in the brain for later retrieval. Although not all of our memories are real and accurate. We also have the ability to make decisions without having any data at all via our intuition. Intuition is commonly defined as knowing without knowing. But that isn’t the best definition for it. Because intuition is about knowing without data. You have to know it to perform it. So instead of knowing without knowing, you’re trusting without data. And it’s certainly a human differential.

AI’s deep learning differential is efficiency. It requires expensive hardware made up of metal, alloys and plastic along with GPUs (graphical processing units) for complex mathematical computation to make fast and efficient decisions. Whereas, our hardware is made up of natural organic matter designed to function in the natural world.

We’re in a league of our own. We can’t and shouldn’t want to compete with AI when it comes to repetitive and efficient tasks. The human brain is flexible and incredibly creative. For example, performing mathematical problems should take time to allow relational problem solving instead of problem solving based on speed and memorization of which we’ve been mostly taught. It’s been found math test anxiety correlates to working memory and performance. Fast just isn’t our niche. We simply don’t have the hardware to accommodate that type of processing. And it worsens with age.

Yet, aged research of 20 years ago is responsible for the advances we’ve seen in recent AI. Deep neural networks are increasingly getting better at object recognition tasks but fail miserably in simple ways that would never be found in human behavior. Meanwhile we’re behind in understanding the human brain and intelligence as it stands to still be the most complex, powerful operating system ever known.

Preceding the human brain, it was the study of primate brain physiology that provided the guidance on how to approach facial recognition in artificial intelligence. It’s the interdisciplinary approach of fields in neuroscience, cognitive science and computation that’ll be the driving force for continued AI advancement. So, what will it be for human beings? It will be unraveling what lies behind our abstract intuitive ability to tap into the brain’s complexity.

This powerful complexity is a core reason why we can anticipate customer’s needs during a conversation before the data proves it. It’s the reason why we can tell stories our clients relate to for understanding the utility behind the features and benefits of a product. And although we may not recall every memory exactly as it happened our ability to derive meaning from them is key.

In the era of AI, it’s human learning- human brain and intelligence that should be front in center- now more than any other time in history.

References

Deep Learning (Ian J. Goodfellow, Yoshua Bengio and Aaron Courville), MIT Press, 2016. http://www.deeplearningbook.org/

The Relationships Among Working Memory,Math Anxiety, and Performance https://www.apa.org/news/press/releases/xge1302224.pdf

The Memory of Illusion https://blogs.scientificamerican.com/mind-guest-blog/the-memory-illusion/

Turing++ Questions: A Test for the Science of (Human) Intelligence http://cbmm.mit.edu/sites/default/files/publications/Turing_Plus_Questions.pdf

Brains, Minds and Machines https://ocw.mit.edu/resources/res-9-003-brains-minds-and-machines-summer-course-summer-2015/introduction/MITRES_9_003SUM15_lec0.pdf

Mathematics Anxiety, Working Memory, and Mathematics Performance in Secondary-School Children https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4735424/

Hey there. I’m Nerissa Kelly and I implement change initiatives and whole brain training for the future of work. If you or your company is in need of my help, I’m happy to assist. You can reach me here.