Some Human Musings On Machine Learning

Throughout our semiconductor industry, there are examples of binary balance. By that, I’m not just referring to the 1s and 0s in binary code. This balance also applies to n-well and p-well device features or the deposition and etching of materials on a wafer.

This duality is present in our human makeup, too. We use both hard intellect and intangible feeling in recognizing challenges, finding solutions and living our lives. To date, the two main hemispheres of computing have been logic and memory. But if artificial intelligence and machine-learning systems are to truly think as humans do, won’t that require more than purely digital computations?

Beginning in the 1920s, John von Neumann applied his genius in mathematics across a wide spectrum of projects. These included working on the Manhattan Project to construct the first atomic bomb; creating the landmark von Neumann architecture for digital computers that store both programs and data; and developing the field of game theory, which many high-stakes poker players use today to deduce future outcomes and win tens of millions of dollars. With these and more accomplishments, von Neumann’s work has come to represent analytical, left-brained thinking, based on data and calculations.

But binary balance is present even in the way our minds work. There is also a right-brained approach – a more ethereal, instinctive way of perceiving things – that could be described as emotional intellect. It complements cognitive, rational decision making with more analog or interpretive ways of thinking. It takes into account human feelings and attempts to inform actions that are difficult to quantify. As an example, whereas von Neumann’s game theory is used to arrive at decisions through logical reasoning, poker players also gather information about their opponents by reading their body language and demeanor at the table. That’s the right brain at work.

Making machines that more closely replicate the way that humans deliberate is addressed in neuromorphic computing. Instead of strictly digital processing, neuromorphic chips assimilate analog information, which is then interpreted for shades of meaning. This opens the door for creating neural networks that are aligned with how we think.

A sort of precursor to neuromorphic computing is already widespread in our lives. When we shop or make a purchase from an online retailer, the interest that we express in that product is cataloged, grouped along with the interests of other buyers, compared with those buyers’ previous purchases and used to pitch us on buying additional products that people in the same demographic have bought. Emails and pop-up ads that proclaim “You might also be interested in …” demonstrate computing power being applied to get into consumers’ heads and influence spending patterns.

Similarly, machine learning can be applied in directing (or is it “assisting”?) people’s future actions. Databases are being used to predict consumers’ needs and to stock local inventories so that, as soon as you exhaust your supply of a certain item, your local store or distributor will know and be able to replace it by offering you same-day delivery.

The next step is to factor in product reviews from other members of your demographic group. This would allow retailers to make high-probability extrapolations about your level of satisfaction with the products you are currently using and the likelihood that you may switch to a similar product from a different producer. This educated guesswork will be based on “reading” your emotional decision-making processes. With this ability to predict future behavior, the continued dominance of poker-player computers is assured. Bet on it.

The state of the art in neuromorphic computing has not yet reached the point of precisely predicting all of our next moves. We are not (yet) living in the world of the Steven Spielberg movie “Minority Report,” in which savant-like “pre-cogs” can predict future crimes before they actually occur, enabling law enforcement to arrest criminals-to-be before they do their intended damage. But it’s intriguing to think about.

Would the fusing of digital processing and emotional intellect present a huge benefit, enabling our voice-command assistants like Alexa and Siri to better anticipate our desires? Or would it bring us one step closer to having the omnipresent machinery in our lives actually run our lives? One thing seems certain: If and when full-blown neuromorphic computing exists, it will be used.