Hyperbole

Hyperbole bubbling over

There’s been a lot of coverage in the media surrounding comments on AI from figures like the entrepreneur Elon Musk and physicist Stephen Hawking. High-profile predictions of doom can be fun and help shift papers / attract clicks. From the perspective of a company focused on Applied AI they aren’t particularly useful and serve only to promote further misconceptions about the field.

The very term Artificial Intelligence is misleading and doesn’t differentiate between the two major divisions: Strong AI and Weak (or Applied AI). Strong AI is the attempt to create automated systems that “think”. Whereas Applied AI focuses on systems which operate within restricted target domains (e.g., playing chess, or recognising speech). They are not mutually exclusive, but neither are they dependent upon one another.

Strong AI research can, at times, verge on Neuroscience (with experiments trying to simulate a brain at the level of synapses firing electrons). However, equally, it can become very philosophical; asking questions about what constitutes knowledge or intelligence? The Chinese Room is one oft-quoted thought experiment. The natural world has always been a source of inspiration for new developments, however, this doesn’t mean we need to slavishly ape it. 

Take flight as an example. The desire to fly has captivated mankind since time immemorial. You’ll have all seen woodcuts of misguided medieval monks jumping from high places covered in feathers, or sepia photos of prim Victorians pedalling flapping contraptions off of piers. You’ll no doubt also have noticed that commercial aviation involves surprisingly few feathers and very little flapping. Powered flight has unlocked the skies for humanity but in a different fashion from the natural examples.

Diagram of Alexandre Goupil's Flying Machine 1883

(Image from http://www.flyingmachines.org/goup.html)

Similarly AI has the potential to revolutionise many aspects of the modern world without any requirement for “thought” or “consciousness”. At Airts we primarily focus on combinatorial optimisation, trying to find the “best” answer to a decision problem. There’s lots of cute examples of decision problems (such as solving Sudoko puzzles) but we look for situations where there’s real benefit to be had from automating a previously manual process.

A recent successful application of AI has been for the maintenance scheduling of the Hong Kong Subway. The system is capable of planning the work more efficiently than humans, resulting in the actual maintenance crews having more time to perform their work and reducing the downtime of the overall network. Double win!

(Picture from https://www.flickr.com/photos/mtaphotos/6941656505/)

(Picture from https://www.flickr.com/photos/mtaphotos/6941656505/)

Imagine all the situations where day-to-day life could be improved via better scheduling? Do you want the waiting time for your operation to be dependent on someone’s spreadsheet prowess? Regardless of how good they are they’ll realistically only have time for one or two attempts. An automated system can assess tens of thousands of candidate schedules a second. 

An interesting blog post explored this idea in the context of teaching pottery to school children. One group were told they had a semester in which to make the best clay pot they could. The other class were told to make 50 pots and they’d be graded on their best effort. The group making multiple pots produced far superior work. Rather than fixating over getting the “perfect” pot, they had practised enough to discover what worked and what didn’t. 

(Image from https://www.flickr.com/photos/bridgmanpottery/174255827)

(Image from https://www.flickr.com/photos/bridgmanpottery/174255827)

Just as the teacher decides which pot was the “best”, so for an AI system we still decide which of its attempts is best. Capturing the requirements, translating them into a form that a computer can interpret, providing it with the tools to construct a schedule / pot, determining a scoring criteria all remain in our control.

AI will certainly impact on the way we work, but that’s not necessarily a bad thing. Computers can perform arithmetic faster and with fewer errors than humans. This hasn’t caused the demise of the accountant. Roles and professions change over time. Personally I’m fine that my surgeon won’t also give me a haircut…

The next time you are worried about an impending AI apocalypse, remember that the computer isn’t some Machiavellian machine plotting our downfall, it’s more like a roomful of children trying to win our approval by making pots!

Posted in Artificial Intelligence and tagged , .