Artificial intelligence: my take

AI is much in the news, and as I struggled to grapple with the plethora, I realized that I needed to develop a framework that will help me position this barrage of hype and the occasional nugget.  Herewith is my attempt. Hopefully it will be of some interest, use or amusement.

My definition:

AI is an improvable black box relying on various inputs to achieve a defined objective, and embodying feedback loops which create continuous improvement.  AI is a bit of a misnomer. There’s really not much artificial about all that is going on [except probably stock market claims], and continuous improvement seems a pretty narrow definition of intelligence – but, I quibble.  We‘re stuck with the AI title.

The Computing Engine:

 

There’s a lot of really sophisticated computing going on within those Black Boxes, and what began years ago with simple decision trees has moved on to deep learning by observation and exploration.  Convoluted neural networks. Stuff like that! Huge resources are committed to advancing AI capabilities for use by corporations, government agencies and academia. Processing power and speed has opened up real time calculations that were unthinkable five years ago, so we are certainly in the early days.

Inputs:

 

The inputs available to AI systems are improving exponentially – virtually anything measured.  Sensor technologies see, hear and touch [probably I just have never heard of the work done on taste and smell];  GPS provides spatial awareness. Data input no longer needs to be precisely defined and regimented. And it is all now available over the Internet, feeding our ubiquitous boxes.

Outputs:

This combination of computational power, data availability and real time results offers opportunities for almost any repeatable process to be improved by AI.  Every day, we read of new applications: autonomous vehicles, machines that read mammograms better than physicians, precision farming, robots that assemble online orders, a virtually endless list.  And all of this built on the capability to continuously improve the computational algorithms as results are evaluated and looped back into the AI decision-maker. AlphaGo played itself one million times before it was ready to defeat the human Grand Master.

Benefits and Costs:

 

The results and implications of the global application of AI technologies are unfathomable – yet will become more visible and impactful every year.  Mostly to the good.

But these benefits will not be shared equally – raising serious moral and ethical issues.  Many today jobs will disappear – so quickly that retraining is an unlikely prospect for those many displaced from routine or repetitive tasks.  Industries, professions, regions, countries will all be differentially impacted. Inequalities – already so pronounced – will probably worsen further, and social tensions will rise.  We will need wiser governance than currently on offer.

 

Some people fear the day when machines will be smarter than humans.  Like HAL, Arthur C. Clarke’s shipboard computer, machines will attempt to assume control.  Interesting science fiction, but our AI technologies are focussed on doing a specific task better and better.  It probably would be a bad idea to use the Go champion to read mammograms, and if the robot grocery picker was driving an 18-wheeler, few of us would be driving the 401.  They will stay in their specialized roles.

Until they don’t.  When and if the technologists achieve the holy grail of a “generally intelligent” AI, perhaps we can look forward to the day when politicians will be displaced by intelligent entities.

Artificial intelligence: my take

AI is much in the news, and as I struggled to grapple with the plethora, I realized that I needed to develop a framework that will help me position this barrage of hype and the occasional nugget.  Herewith is my attempt. Hopefully it will be of some interest, use or amusement.

My definition:

AI is an improvable black box relying on various inputs to achieve a defined objective, and embodying feedback loops which create continuous improvement.  AI is a bit of a misnomer. There’s really not much artificial about all that is going on [except probably stock market claims], and continuous improvement seems a pretty narrow definition of intelligence – but, I quibble.  We‘re stuck with the AI title.

The Computing Engine:

 

There’s a lot of really sophisticated computing going on within those Black Boxes, and what began years ago with simple decision trees has moved on to deep learning by observation and exploration.  Convoluted neural networks. Stuff like that! Huge resources are committed to advancing AI capabilities for use by corporations, government agencies and academia. Processing power and speed has opened up real time calculations that were unthinkable five years ago, so we are certainly in the early days.

Inputs:

 

The inputs available to AI systems are improving exponentially – virtually anything measured.  Sensor technologies see, hear and touch [probably I just have never heard of the work done on taste and smell];  GPS provides spatial awareness. Data input no longer needs to be precisely defined and regimented. And it is all now available over the Internet, feeding our ubiquitous boxes.

Outputs:

This combination of computational power, data availability and real time results offers opportunities for almost any repeatable process to be improved by AI.  Every day, we read of new applications: autonomous vehicles, machines that read mammograms better than physicians, precision farming, robots that assemble online orders, a virtually endless list.  And all of this built on the capability to continuously improve the computational algorithms as results are evaluated and looped back into the AI decision-maker. AlphaGo played itself one million times before it was ready to defeat the human Grand Master.

Benefits and Costs:

 

The results and implications of the global application of AI technologies are unfathomable – yet will become more visible and impactful every year.  Mostly to the good.

But these benefits will not be shared equally – raising serious moral and ethical issues.  Many today jobs will disappear – so quickly that retraining is an unlikely prospect for those many displaced from routine or repetitive tasks.  Industries, professions, regions, countries will all be differentially impacted. Inequalities – already so pronounced – will probably worsen further, and social tensions will rise.  We will need wiser governance than currently on offer.

 

Some people fear the day when machines will be smarter than humans.  Like HAL, Arthur C. Clarke’s shipboard computer, machines will attempt to assume control.  Interesting science fiction, but our AI technologies are focussed on doing a specific task better and better.  It probably would be a bad idea to use the Go champion to read mammograms, and if the robot grocery picker was driving an 18-wheeler, few of us would be driving the 401.  They will stay in their specialized roles.

Until they don’t.  When and if the technologists achieve the holy grail of a “generally intelligent” AI, perhaps we can look forward to the day when politicians will be displaced by intelligent entities.