Blog: “A tool for exploring deeper”
In his absorbing book, The Creativity Code, University of Oxford professor Marcus Du Sautoy describes his experience watching the international GO championships. Recalling that experience professor Du Sautoy writes;
“Ke Jie’s compatriot Gu Li, winner of the most Go world titles said, ‘Together, humans and AI will soon uncover the deeper mysteries of Go’.
For Demis Hassabis, the algorithm is like the Hubble telescope of Go. This illustrates the way many view this new AI. It is a tool for exploring deeper, further, wider than ever before. It is not meant to replace human creativity but to augment it.
Hassabis and his team at DeepMind created AlphaGo, a software application that defeated Lee Sedol, one of the world’s best Go players but for me the more startling achievement is the opportunities the work of the DeepMind team presents for all mankind. As a software consultant I would always encourage my clients to use information technology to “support” people, rather than as a “substitute” for people. Sadly, it was not the prevailing wisdom at the time.
During the 1980’s and 90’s Packaged application software was all the rage. In an effort to sell as much of it as possible, software sales and marketing people would position information technology and software as a way to reduce labor costs (i.e. people ). For transaction processing software, like payroll and accounting, their arguments were not far off but for more analytical applications like business intelligence (BI) or workflow management the “reduce unit labor cost” arguments were misleading at best and in many cases simply facetious.
On an individual scale, physical productivity depends on the difficulty of the task, the skills of the worker and his learning curve (the number of times he already performed the task and how he was guided by good “teachers / masters”). For example, think of an unskilled buyer assembling an IKEA piece of furniture for the first time: he’ll be looking to avoid mistakes more than optimizing the time to work. If he purchased a second piece, he will be much more productive (i.e. will need less time to assemble it). After a few pieces his productivity will level off with no further large improvements.
There are many business processes that require a great deal of creativity and problem solving to complete. They are not processes that lend themselves to the rigor of conventional technology. More often than not creative processes require a mental ‘leap’ not so much in the case of a dramatic change in the work that is performed but rather in the in the way tasks are arranged and rearranged to complete an assignment. Some tasks may require counting backwards or even sideways. These may be the tasks for which the collaboration between AI and human creativity may be best suited.
Think of an astronaut in a completely foreign environment in which processes that are generally accepted on Earth fail to behave as expected. What is the astronaut to do when he presses a “red” button expecting his motion to stop but instead it accelerates?The first thing he needs to do, perhaps quickly, is to experiment with the “red” button to understand how it operates in his new, unfamiliar environment.
That process of experimentation requires creativity. There are no thinking rules that say ‘apply more weight to the button and it will accelerate less’. The astronaut’s situation simply comes down to “trial and error”. Many times performing trial and error tests can be catastrophic. A human may die before he’s successful. This is where AI can be extraordinarily helpful. AI can wiz through literally thousands of trials and learn from it’s errors in a period of time a human being may not even comprehend. This is essentially how AlphaGo was able to defeat it’s human competitors. It could examine the Go board and evaluate moves and it’s opponent’s counter moves in literally minutes. Many fewer minutes and many more alternatives than any human could ever imagine.
We may not all be astronauts, facing life or death problems, but most of us face many problems for which we could use assistance developing and evaluating alternative solutions. That’s exactly how “a tool for exploring deeper” can be used.
That’s not the story Martin Ford tells in his award winning book, “Rise of the Robots, Technology and the Threat of a Jobless Future”. According to Ford, As technology continues to accelerate and machines begin taking care of themselves, fewer people will be necessary. According to Ford, artificial intelligence is already well on its way to making “good jobs” obsolete…The result could well be massive unemployment and inequality as well as the implosion of the consumer economy itself. Quite dire indeed.
Ford and those with similar ideas believe the benefits we may have received from advancing technology represent a “Goldilocks economy’ which is warm enough with steady economic growth to prevent a recession but not so hot as to push it into an inflationary status. Ford believes that a Goldilocks period ends when technology advances to the point it’s widely substituted for labor and wide-scale unemployment results.
I don’t mean to sound pollyannish but there are distinct differences in the futures described by professor Du Sautoy and Mr. Ford. It’s important that we understand those differences and come to an understanding of the future we’re likely to face so we can prepare for it. Will we face a:
- A Du Sautoy future in which AI is used as support for human workers or
- A Ford future in which AI is used as a substitute for human workers
My bias is toward the Du Sautory future but it’s a bias and a Ford future could certainly come about largely because U.S. business schools indoctrinate their students, the managers of most large businesses, with the idea that their job is to reduce unit labor costs and the most efficient way to do that is to substitute capital ( a.k.a. technology) for labor. As a consequence most workflows are designed to resolve problems of efficiency, not as “a tool for exploring deeper”.
Managers are not taught to enable employees to “explore deeper”. They’re taught to “produce less expensively”, which means that a Ford future is more likely than a Du Sautory future.The only way for this to turnout differently is for the production of products and services to become more reliant on human creativity than machinery and the champions of efficiency are unlikely to allow that to happen.
For the efficiency experts human workers will always be just one of the four factors of production to be assembled and reassembled in the most efficient arrangement possible that produces the product or service their business offers for sale. That arrangement rarely, if ever, requires “tools for exploring deeper”. Rather it requires technology that drives down costs as units of production increase.
Variable costs, like labor increase as production increases. However, substituting technology for labor changes costs from variable to fixed and fixed cost per unit decline as production increases. So managers trained in conventional cost accounting will always replace labor with technology. They have very little use for products or services that require “exploring deeper”. Therefore they have very little use for AI.
Going forward work and work flows need to be dramatically redesigned separating tasks that are best performed by machines from those best performed by human beings. Once that’s done Artificial Intelligence (AI) applications can be designed and built to enable human beings to “explore deeper” while designing and building other applications that maximize the efficiency of related machinery.
The ability to think creatively has long been understood as a specifically human skill. If AI were to truly crack creative thought it would depart from being an extension of code written by a programmer to thinking on their own and going beyond playing a 4,000 year old Chinese board game.
Above on the left is the result of a river in Tubingen painted to be stylistically similar to various paintings, including J.M. Turner’s “The Wreck of a Transport Ship,” Van Gogh’s “The Starry Night,” and Edvard Munch’s “The Scream.” Next to it is a comparison with Van Gogh’s actually Starry Night. This is not to suggest the computer paints with the same serendipity and spontaneity that a human being might but even the greatest human artists spent endless hours touring art galleries collecting information on themes and techniques before they put a brush to canvas. That’s not unlike all of the data available to machine learning algorithms before they put their own brush to canvas.
Both Leonardo, on the left and Michelangelo on the right kept volumes of information they used in preparation for the creation of their greatest works. Just think of how much more “productive” they may have been if they had a machine learning application collecting all of that information on their behalf, although I really doubt either would have cared.
Only the much lesser talented ones among us need those kind of tools for exploring deeper!