The term Artificial Intelligence is becoming more commonplace in our everyday lives. How people react to AI depends very much on their actual interaction with it. People working in a knowledge-based industry see the value that AI can provide.

AI algorithms can take over repetitive tasks and perform them quicker and continuously learning how to improve results. This means that people are left with more time for the creative tasks which might result in more fulfilling working environments and conditions.

One of the most recent advances in AI is that of reinforcement learning – algorithms that are set to experiment and figure out how things are done without specific instruction placed by the programmer as is the norm. This type of learning algorithm was used by AlphaGo – a computer software that learnt how to play the game Go by playing against itself many times. Over time, the software was able to compete in a number of championships and it beat the best professional Go players in the world in 2016.

This was hailed as a technological breakthrough in AI because Go is a strategic game that, unlike chess, has a very large branching factor making it too complex to use traditional algorithms. Another aspect about the game Go is that often, even accomplished Go players are unable to say why certain moves are better than others. The analysis of how AlphaGo played revealed that it made certain moves which were never usually played, encouraging professional players to learn new moves and strategies by observing how AlphaGo played.

Researchers saw that the success of reinforcement learning in the arena of games can be transferred to other domains where it is difficult for humans to actually codify all their decision-making processes, such as self-driving cars. Such cars must be able to react quickly, drive smoothly without causing others to slow down. This is a rather complex task. Think about it – how do you realise that another driver is about to turn without using an indicator?

There are certain elements in driving and our decision-making that are based solely on ‘gut feelings’. The reality is that our brains are complex and capable of making estimates on the basis of our previous observations and experiences.

Reinforcement learning follows the same principles. Self-driving cars must be extremely competent at the task when it comes to interacting with human drivers and they need to be capable of predicting the actions and outcomes of other drivers. Reinforcement learning allows the software to learn how to drive in a safe environment through virtual simulations rather than on actual roads. And the more the software practises in a simulated environment, which is the equivalent of AlphaGo playing against itself, the better the self-driving car will be when it interacts with human drivers.

Dr Claudia Borg is a lecturer with the Department of Artificial Intelligence at the University of Malta.

Sound bites

• Researchers have shown that a slight modification to the way reinforcement learning rewards itself will allow humans to maintain the upper hand and control agent systems that are guided by AI. This is done seamlessly whenever a human interrupts an action that the agent is doing. When this occurs, the agent learns that the action in certain situations is not desirable and therefore should be avoided. This is also being applied to when agents need to communicate between themselves and negotiate actions between them.

https://www.sciencedaily.com/releases/2017/12/171204094950.htm

• Researchers have developed a robotic learning technology that enables robots to imagine the future of their actions so they can figure out how to manipulate objects they have never encountered before. In the future, this technology could help self-driving cars anticipate future events on the road and produce more intelligent robotic assistants in homes, but the initial prototype focuses on learning simple manual skills entirely from autonomous play.

https://www.sciencedaily.com/releases/2017/12/171204162335.htm

• For more interesting science news listen to Radio Mocha every Saturday at 11.05am on Radju Malta 93.7FM.

https://www.facebook.com/RadioMochaMalta/

Did you know?

• Driverless cars are able to recreate roadways in three dimensions. They must be able to recognise pedestrians, signage, other vehicles and traffic lanes in order to function well.

• The first autonomous car prototype was operated by Google in 2015.

• It has been estimated that by 2035 10 per cent of vehicles will be autonomous and by 2050 almost all vehicles on the road will be autonomous.

• Driverless cars could help eliminating issues of drunk driving, accidents due to texting and cell phone use, and other instances where drivers are distracted.

• Google’s self-driving car had an initial speed limit of 40 km/hr. However, now it generally observes the given speed limit and is even allowed to overspeed by 10 km/hr when other cars are also overspeeding.

For more trivia see: www.um.edu.mt/think

Sign up to our free newsletters

Get the best updates straight to your inbox:
Please select at least one mailing list.

You can unsubscribe at any time by clicking the link in the footer of our emails. We use Mailchimp as our marketing platform. By subscribing, you acknowledge that your information will be transferred to Mailchimp for processing.