Artificial lack of intelligence: how smart is AI really?
Wireless Technologies

Artificial lack of intelligence: how smart is AI really?

This is a contributed piece by Gavin Fell, general manager UK at Exact


Thanks to machine learning it will not be long before robots and intelligent computers can do more than people. That’s the hype, anyway. According to some, this will mean that we can spend our future days relaxing on the beach; others fear that we will have to hide in the shelters. What will it be? And when exactly will it happen?

To understand how we can control Artificial Intelligence (AI), and whether we will even need to do that in the first place, it’s a good idea to look at where we are now. Realistically, artificial intelligence is not actually all that intelligent - at least, not yet. Let's take a step back before we assess the great promises that surround artificial intelligence.


The arms race between Google and IBM

All the major players in the tech world are investing heavily in robot technology. AI is already used for important everyday tasks like predicting the weather and to perform repetitive work in customer services or on the farm.

Currently the competitive battle for dominance of ‘next-gen’ AI is dominated by Google and IBM - both in terms of R&D and marketing. IBM invests a large proportion of its budget on Watson, its proprietary artificial intelligence platform and a poster child for the concept. Last year, the company announced that it has managed to imitate neuronal processes using computers. That would mean that computers are increasingly capable of imitating our brain functions; a next step might be that they can actually do a better job than people in highly complex areas.

Last year, Google's artificial intelligence platform DeepMind became world champion at Go - widely regarded as one of the most difficult games in the world because its moves are hard to predict and players rely on intuition rather than algorithmic rules. What fewer people know is that Google Translate entirely invented its own artificial language. Why? The ‘third’ language made it easier to create large numbers of translations. Nobody gave the system the command to do so; the computer invented it all by itself.


AI is still relatively dumb

While that's all very practical and clever, in general AI is still relatively stupid. We’ve all seen the videos of robots trying and largely failing to play football or walk upstairs, for example.

Even the victory of Google's DeepMind over mankind during a game of Go has to be put into perspective. In the end, this is not much more than probability calculations in a controlled environment. Interacting in a world full of unexpected impulses is something else altogether.

What makes us human, and intelligent beings, is much more than probability and analytical skills. In uncontrolled environments with unexpected twists we can improvise, based on our past experiences. Try doing that on the computer. Another important difference between humans and artificial intelligence is that the latter currently cannot identify objects from cold. All computers do is to discover patterns from massive data sets and draw conclusions from there.


We are on the verge of a major breakthrough

The crucial hurdle on the road to real intelligent computers is known as the ‘corner case’. Corner cases are defined as problems or situations which only occur outside of normal operating parameters. KUKA, a German manufacturer of industrial robots, illustrated this by having one of their robots play table tennis against champion Timo Boll. The robot built up a steaming lead, until Boll discovered a corner case - the fact that the robot couldn’t decipher his movements if they were unconventional or masked by other parts of his body - and exploited it to win the match.

In a game of table tennis or a weather forecast, we’re happy to let those corner cases pass. If a computer gets it right roughly 80% of the time, that's very nice. But when driving a car, we have to set that bar much higher. We recently saw a video of a Tesla car anticipating an accident on the motorway. Very clever, but as long as this is special, it might be better to let humans drive cars. To be better at driving than humans, the software of a self-driving car must be able to read its surroundings very carefully. The car should be able to tell the difference between a harmless newspaper blowing across the road or a pedestrian crossing suddenly.

According to Hod Lipson, engineering professor at Columbia University in New York, it will not be long before AI is capable of doing just that. Corner cases are slowly disappearing thanks to the huge amounts of data becoming available worldwide as fuel for AI. A good example is a race which recently took place between self-driving cars; while one car effortlessly dodged a dog, its opponent crashed. So it's safe to say that there is still quite a way to go in making self-driving cars truly intelligent.


Where robots already beat people

Although artificial intelligence is pretty smart in a controlled environment, its skills still very much depend on probability. Driving a car when kids are crossing the road is a different ballgame than playing chess. How do we make computers so smart that they can anticipate, improvise and make decisions? That is the central question in cracking corner cases; now developers have determined that this is the final hurdle, we can expect AI to improve very quickly.

To illustrate this tipping point, it is interesting to look at the ImageNet Large Scale Visual Recognition Challenge (ILSVRC). This is an international race to write the best picture recognition algorithm. Last year, the human error margin of 5.1% (which has been the Holy Grail of the contest) was improved on by computers for the first time. In everyday language; software is better able to determine what can be seen on photos than you or me. Using a similar algorithm, your car will soon have a smaller error margin when driving than you.

In summary, there is a need for nuance in how we talk about artificial intelligence. We are not as far advanced as some experts suggest. However, that’s not to say that a sudden breakthrough won’t happen.



Also read:

AI-as-a-service: IBM and Infosys lock horns but can anyone challenge Google


«Are digital subscription models the answer in the wake of fake news?


Box CEO: Transformation is more than putting software on top of traditional business models»
IDG Connect

IDG Connect tackles the tech stories that matter to you

Our Case Studies

IDG Connect delivers full creative solutions to meet all your demand generatlon needs. These cover the full scope of options, from customized content and lead delivery through to fully integrated campaigns.


Our Marketing Research

Our in-house analyst and editorial team create a range of insights for the global marketing community. These look at IT buying preferences, the latest soclal media trends and other zeitgeist topics.



Should the government regulate Artificial Intelligence?