2015 : WHAT DO YOU THINK ABOUT MACHINES THAT THINK?

bruce_parker's picture
Visiting Professor, Stevens Institute of Technology; Author, The Power of the Sea: Tsunamis, Storm Surges, and Our Quest to Predict Disasters
Grandchildren, If-Then Statements, And Artificial (Natural) Selection

Grandchildren give us a second chance to observe and be fascinated by the learning system with which new little humans come into the world. Driven by an insatiable curiosity, they somehow make sense of the totally unknown environment into which they have been thrust. And the sheer delight of each new discovery, as they piece together this new world, reveals an inherent sense of humor with which they are also born.          

I think it is fair to say that no artificial digital machine will ever go through exactly the same delightful process as a human baby discovering the world. It is even possible that no artificial machine will ever approach the intelligence potential of a newborn human baby. In the natural world, after 3.5 billion years of natural-selection-driven evolution, only one species developed the ability to carry out abstract self-aware conscious analytical thinking. Do we really think we can short-cut the process and succeed on some comparable level?           

In humans it is not just the evolved curiosity and desire to understand the world that set us apart from the rest of the animal kingdom. It is also our evolved tendency toward social cooperation and communication which led to sharing and passing on learned knowledge (eventually leading to science and technology). How many genes must have mutated and been naturally selected to achieve the complex human brain with its curiosity and social bonding and communication capabilities?           

Can we really reproduce this with digital machines? Many believe we can by taking advantage of the ever-growing speed of computation of these machines. Computation power certainly allows these machines to make fast and accurate decisions, when those decisions only require large digital databases and (the equivalent of) many thousands of if-then statements to make the best choice among numerous possibilities. We know that with this brute force technique such machines can defeat chess champions, provide autopilots for jet planes for use during hazardous conditions, rapidly buy and sell stocks based on complex changes in the market, and carry out endless other functions. Computation power can also allow realistic looking imitations of human actions, decisions, and even emotions (mere technical puppetry really), but it may never produce true analytical thinking. A machine may be able to self-monitor what decisions it has made, but it may never attain human-like self awareness and consciousness.          

At least not without the right software. But how can we produce software as powerful as the genetically based software of our brains that took nature 3.5 billion years to produce?  We are so far from understanding the software of our brains. Some may talk of the efficient parallelism inherent in the brain's structure, but that is such an inadequate description of what our brains do. Parallelism in our computer operating systems and programs merely lets us do many things at the same time, admittedly in some very creative ways, but again that only is further increasing computation speed. Will we ever be able to reverse-engineer our brain—not in the sense of circuits/networks of neurons, which we are presently making strides in understanding, but in an overall "design" that would allow digital machines to think abstractly, have a sense of self, etc., in a manner similar to humans?             

Short of some incredible analytical breakthrough, our only other recourse seems to be to write programs that try to imitate the evolutionary process, taking advantage of our artificial machines' high-speed computational abilities (so that we might accomplish this in less than 3.5 billion years). We can create reproducing digital entities (programs that reproduce themselves) and give them mutations, but stimulating evolution toward eventually becoming a thinking machine is a much more daunting task. For this to work we must find a way to create a machine environment with a natural-selection-like driving force (which would actually be artificial selection) or some other motivation which would lead to the necessary changes. Can we make a machine "want" something in a way that would select for greater intelligence?

Any future advances in intelligence are more likely to be a result of what we will soon be able to do to the only thinking machines we presently have—ourselves.           

The natural-selection-driven evolution of the homo sapiens species stopped when humans created societies (families, tribes, towns, cities, countries), because now they could protect the weak, and survival of the fittest no longer drove a natural selection process. Humans with deficiencies that would have killed them could now live long enough to reproduce           

But now we are on the verge of being able to change the human species with genetic engineering. We will at some point try to enhance our intelligence by attempting to isolate the genes responsible for higher intelligence and greater analytical ability. And we will be able to spot those genes before we understand how they work—and long before we are able to correctly emulate them in digital programs. Artificial selection will change our genetic make up instead of natural selection.           

Our future is probably enhanced biological intelligence, not machine intelligence. And it is there that the dangers and/or benefits lie. We might, for example, select particular genes (or even create new genes) that we think will increase intelligence, while perhaps not really understanding how particular gene combinations work. Could we unknowingly begin a process that could change the best human qualities? While striving for higher intelligence could we somehow genetically diminish our capacity for compassion, or our inherent need for social bonding? How might the human species be changed in the long run? The qualities that got us here—the curiosity, the intelligence, the compassion and cooperation resulting from our need for social bonding—involve a complex combination of genes. Could these have been produced through artificial genetic selection? Could we lose them? Such worrying may not stop some scientists from deciding to use artificial selection. What will our grandchildren be like then?