Separating AI wheat from chaff

In the same interview, Schmidt said that the next generation of applications would be using machine intelligence platforms, fast networks, smartphones and tremendously rapid computational iterations. Right again!

Building machine intelligence differs from human-level intelligence. Mixing the two produces natural ignorance. Fortunes will be made by those who get this difference. Repeated failures and losses will await those who don’t.

Two types of errors can typically take place when it comes to any fixation like AI. We can drop the technological development path of the organization in pursuit of something vague but hyped like a mania that eventually proves having no value, or we can dismiss something tremendously beneficial because we do not take the needed time and effort to learn and apply it. Both are detrimental.

AI went through a false start in the late 1960s. The rough beginning was mainly due to striving for creating human-level intelligence, then applying that general-purpose platform in all instances. Hubert Dreyfus of the University of California, Berkeley, who sadly died last year, wrote a seminal book in 1972, entitled “What computers can’t do, the limits of artificial intelligence.” He predicted that the AI approach of that era was a dead-end. He was proven to be correct.

After significant hibernation, AI came back with a gusto toward the end of last decade. This time, the developments have been on a solid footing, with promising results.

To think about AI rationally, we first need to recognize incredible technological advancements in various fields of engineering and science that have taken place: computational power and data storage, robotics, natural language processing, image recognition, advanced algorithms for pattern discovery, autonomous system control and more. For good research and development reasons, breakthroughs came about in silos.

Now, the leading-edge AI work is dedicated to creating platforms that integrate advancements of various fields for delivering applications. Be it in health care, aerospace, social media, to name a few, AI-based applications are flourishing since they actually create tangible value.

But those applications are specific, designed for particular purposes. The AI application that can win Jeopardy cannot win the game Go, because it is not designed to win Go.

And that has been the reason for all the successes of AI we are witnessing now. The moment researchers gave up on creating digital version of human-level cognitive intelligence, and instead focused on applications, AI flourished. Experts separated the realistic wheat from the delusional chaff.

When I was an engineering undergraduate student at the University of Illinois, Urbana, the movie “2001: A Space Odyssey” was screened a few times a year on campus, even after many years from its first release. The entire campus movie theater would quiet down when Dave, the only surviving astronaut on a mission to Saturn, began shutting down the AI antagonist, HAL 9000, that had gone berserk. As Dave was turning off the computer, HAL said, “I became operational in Urbana, Illinois.” The entire theater would then burst into standing ovation for several minutes, showing pride in the university’s technological prowess.

The making of that science fiction movie in 1968 coincided with some of the outlandish AI claims of that time in the scientific and research community. That creating the digital version of human-brain functions was possible and within reach in few decades. In fact, the academic and industrial R&D of that era went off the rail, from which AI did not recover for more than three decades.

The huge detriment is when AI is treated like a commodity, with the misguided exclamation that if just a decent amount of AI is applied to any problem, that will solve that problem. That’s a lazy response, dodging the tough task of solving difficult technical and managerial issues. For sure, AI technologies that are becoming ever more powerful can solve many classes of problems. However, we are far away from creating an AI platform that can be applied to all categories and for all cases.

Until many technological breakthroughs happen over a long period of time, producing digital versions of human-level intelligence that can fully function cognitively, AI needs to be streamed for powerful applications.

In real estate, it is all about location, location, location. When it comes to AI, that’s application, application, application!

Jahan Alamzad is a management consultant. He lives in San Carlos.