Artificial Intelligence – Disillusion or Revolution?
Interest in Artificial Intelligence and related methods such as deep learning and machine learning are at an all-time high. Our news feeds are packed with sensational articles from mainstream journalists and various click-bait titles.
Back in July, Gartner’s latest annual Hype Cycle for Emerging Technologies, 2017 report placed deep learning and machine learning at the “peak of inflated expectation”.
So what is next for AI? Will it lead to a computing nirvana or will we instead find ourselves wallowing for years in Gartner’s “trough of disillusionment”?
Narrow AI vs General AI
When discussing Artificial Intelligence it is important to differentiate between “narrow AI” and “general AI”.
Narrow AI technology specialises in one area such as beating the world Chess or Go champion or creating self-driving autonomous vehicles. However, ask the same technology to recognise my handwriting and it wouldn’t know where to start.
General AI refers to intelligence that can reason, think, plan, learn, communicate and solve broad intellectual problems in the same way as humans.
The astonishing successes of Artificial Intelligence
Certainly the application of narrow AI is yielding stunning results. Here are some examples we hear about or use daily:
- Tesla’s autopilot
- Apple’s Siri, “OK Google” and Microsoft’s Cortana.
- Google’s Translate
- Amazon’ s product recommendations
- Facebook’s facial recognition and newsfeed
- Microsoft’s Office 365 spam and clutter classification
Outside of daily use and firmly in the “under development” camp, there are plenty of novel examples of narrow AI. For example, did anyone see the University of Toronto’s AI generated Christmas jingle from last year?
Ok, so it’s not great. The lyrics are perhaps something you might expect from a two- year old. But isn’t that exciting? If at the end of 2016 we have AI capable of generating music at the level of a two-year old and AI capable of driving a car at the skill level of someone far older? Where will this technology be in another year, or two, or five?
The achievements of AI in narrow or specialist fields are already revolutionising and disrupting our industries. There are significant new breakthroughs and applications occurring every day and many. Here are a handful of examples from the news recently:
- genetic deep learning to spot lung cancer before it is too advanced to cure.
- automated texture synthesis to turn pixelated photos into crystal clear images.
- ML and natural language processing to detect “fake news”
- AI that automatically responds to scam emails and wastes the scammer’s time.
There is no sign of narrow AI applications abating. Indeed, investment and advancement in AI is increasing at an exponential rate. There is a shortage of AI experts, although if Google are successful in developing AI to develop AI then perhaps computer scientists could be replaced as soon as transcriptionists.
The worrying implications of AI
In all the AI hype and success, it is important that we recognise the constraints of current AI solutions. Narrow AI operates within a constrained and brittle domain and its knowledge is often based on specific training sets. Among its attributes, it lacks the human-like qualities of general intelligence, culture, context, emotion and ethics.
Consider this AI solution which was trained to understand the difference between dogs and wolves. Instead of learning the differences between the animals it instead learned that wolf pictures include snow and dogs’ pictures included grass. Also recall Microsoft’s ill-fated experiment with “Tay” the teen robot who learned to became a Hitler loving, feminist hating sex maniac in less than 24 hours.
Our global infrastructure is increasingly complex and interconnected. As the adoption of narrow AI increases, its shortcomings present real risks. Is it only a matter of time before narrow AI knocks out a power station, collides autonomous vehicles or causes a global economic collapse? The possibilities are difficult to imagine and frightening at the same time.
A general AI breakthrough?
What about general artificial intelligence? Are we on the cusp of a breakthrough that would enable a computer to think abstractly, reason, plan and learn as a human does?
Arguably we already have sufficient cloud computing hardware to recreate the human brain. However, from a software perspective, we have some way left to go. We have very little idea how to engineer general intelligence, because we have very little understanding of how our own brains and intelligence work.
“Many people in AI believe that we’re close to [a computer passing the Turing Test] within the next five years.”
A giddy prediction and even with the benefit of four years hindsight, I don’t think there’s any chance of the Turing test being passed any time soon.