Unthinking machines

By Robert Hickson 16/09/2014


It’s interesting how “Watson”, the computing system from IBM, is gathering a persona around itself (or rather one is being woven for it). “Watson”, which garnered fame several years ago by winning a game show, is now doing scientific research, among other noble things. Though it’s still a distant forebear of HAL 9000, so the cult of persona doesn’t seem warranted.

“Open the lab door, Watson”

“I’m sorry Dave, I’m afraid I can’t do that”

Researchers from Watson Discovery Advisor, Baylor College of Medicine & the MD Anderson Cancer Center have taught the operating system to analyse 70,000 articles [abstract] to find new protein kinases that may phosphorylate protein tumour suppressor p53, and so be potential sources of new cancer treatments.

The system is also starting to help clinicians diagnose illnesses, police departments analyse crime, and may soon assist firms analyse potential investments and provide other management advice.

Other similar machine learning systems (or “natural intelligence platforms”, one among several terms emerging to describe such systems) are also out there, but without the same cachet as Watson.

While these examples can be impressive, they are still essentially brute force computing, aimed at detecting patterns or solving more complex problems. Computers are getting quicker and more adaptable, moving from numerical to conceptual problems, but not really smarter. The next version of Siri isn’t going to start talking with you about its own views of a restaurant or what it likes or doesn’t like about that piece of art you are thinking about buying online.

Even attempts to develop quantum computers  – by Google  or others – are just ways of developing larger quicker hammers to crack tougher nuts.

While one third of judges thought “Eugene” was a 13 year old boy at a Turing Test earlier this year, it still didn’t pass the test. That test may soon be passed, but its still only one step on the way to creating true artificial intelligence.

As Douglas Hofstadter has pointed out, Watson and similar systems can’t think, or understand what they “read”. Hofstadter thinks that we won’t be able to develop artificial intelligence until we better understand how we ourselves think.

While a range of efforts are going into modelling the brain, and developing artificial brain tissues, we are still a long way from understanding thinking at physical and chemical levels.

Even if we do eventually figure that out, how easy would it be to programme those rules (we’d probably need a computer to help us)?

But do we need to create a computer that thinks like us? There are simpler ways of thinking that can probably be devised for a computing device. We’re not there yet, and will need to take a different approach than brute force computing.

In an article called “The moron and the manager”, published nearly fifty years ago (re-released, so to speak, recently by McKinsey), management consultant Peter Drucker noted that computers “force us to think and to set the criteria”. The best use of them, he suggested, is to assign them the routine or complex tasks freeing us up to think about the important decisions or issues.

Computers have helped move us from, in many cases, a scarcity of information, to, in many cases an excess of information. They haven’t though always be used in management, or other fields, to their best effect. Steven Strogatz worries that increasing reliance on computational approaches will have an even worse effect – killing off insight in the sciences.

They may help create knowledge, but we haven’t always used to to get to the next stage, wisdom.

While there is currently a lot of emphasis on training data scientists, as Drucker noted we need to be paying more attention to is developing better training and techniques for how we think. As computers become more powerful , and more widely used, that is going to be even more essential.

Its one thing to be able to fool a panel of judges after a 5 minute conversation, or to find potentially fruitful patterns in masses of information. Some of our most pressing problems – the middle east, terrorism, climate change, obesity, poverty – won’t be solved by artificial intelligence, but by political and other decision making.

 

As Sherlock Holmes said to another Watson once

“It may be that you are not yourself luminous, but that you are a conductor of light. Some people without possessing genius have a remarkable power of stimulating it.”

That’s how we need to think about how best to use Watson and the other computers being developed.