During the last year, we’ve experienced the escalation of social issues around artificial intelligence (AI), with Elon Musk leading the charge. Musk continues to advocate the idea that humanity is getting closer to a Skynet-like future – to many people’s concern. One of the very real and valid concerns is the idea that many existing jobs will be automated, thanks to AI.
But despite these two major concerns, there continues to be significant innovation and advancement in the field of machine learning (ML) and AI. And perhaps due to these concerns, there has been a significant focus placed on the transparency of machine learning and AI algorithms and advancements. This gives us more insights into what’s happening in the field, but many breakthroughs are not realised in the real world as they might cross the ethics line. In this blog, we’ll discuss some of the latest advancements in the field of machine learning and artificial intelligence and what these developments mean for our future.
In a paper published in Nature on 18 October 2017, Google’s Deepmind demonstrated the ability to master the game of Go, without human knowledge.
A bit of background
AlphaGo, the original, was the first computer program to beat a professional human player at the ancient Chinese game of Go. Following that, AlphaGo continued a winning streak to become the strongest Go player in history.
So what is AlphaGo Zero?
AlphaGo Zero is similar to the original version. The difference between the two is twofold:
- Whereas AlphaGo was taught how to play the game of Go using thousands of human amateur and professional games, AlphaGo Zero was not taught. Zero learned to play the game entirely on its own, playing games against itself, starting with entirely random play and developing tactics as it learned.
- AlphaGo Zero surpassed AlphaGo as the strongest Go player within just 40 days.
Why is this important?
The team at DeepMind, the world leader in AI research and the team behind AlphaGo, believes this has the potential to facilitate major scientific breakthroughs, drastically changing the world for the better.
AlphaGo Zero starts from a blank slate and learns for itself, without any human intervention. This is important because this can be applied to any other domain. If you have a computer that can learn based on zero data from the start – only logic and basic rules, the application of this computer is limitless. This is very important in situations where human knowledge is expensive, unreliable or simply unavailable. The DeepMind team are looking towards similar structured problems, such as protein folding, reducing energy consumption or searching for revolutionary new materials.
Self-driving cars are the future we’ve all been looking forward to and the application of AI that seems the most promising (and most imminent). But with two fatal crashes involving two leading companies in the field occurring within a week, you’ve got to be concerned.
Uber had been testing their self-driving cars in Arizona for between 13 and 17 months, when on 18 March 2018, one of the self-driving cars was involved in a fatal crash. By law, the self-driving car needs to be accompanied by a safety driver, who was indeed present in the car, but who failed to take action. The vehicle was traveling at 40mph or 64kph at the time of the crash. Footage released showed a woman crossing the road, who the car’s sensors failed to pick up. The accident occurred after dark, in Tempe, Arizona.
A week later, on the 23rd of March 2018, another fatality made news when a driver of a Tesla Model X crashed in California. The car was operating on Autopilot at the time and had given several visual and audio warnings earlier on in the drive, but no intervention was detected from the driver.
Does that mean we shouldn’t trust self-driving cars?
Not at all. But many companies are currently investing a lot of resources into developing a self-driving car, and all of these companies are looking to be the first to release a fully autonomous vehicle. History has taught these companies that whoever makes it to market first, is likely to gain a monopoly in what will be a lucrative industry in the future. (Click to Tweet!)
In this innovation race, and with so many companies talking about what is possible – it’s very easy for us to confuse what we currently have available with what is still being developed. Yes – what we’ve achieved thus far is impressive, but we should be careful of overestimating the technology that has been approved for public use and not confuse it for what has been achieved in labs and controlled and secure environments.
For example, while the Tesla autopilot feature is revolutionary and well-worth the hype, the company has said its Autopilot feature “can keep speed, change lanes and self-park but requires drivers to keep their eyes on the road and hands on the wheel, to be able to take control and avoid accidents”. Although being the frontrunner has its advantages, being ahead of the pack in this area is a big risk as a lot of the legislation and ramifications for “failure of the AI system” is untested in court and may result in significant financial consequences.
I’ll bet you didn’t think the GDPR has anything to do with ML or AI? It does, and with the GDPR coming into effect towards the end of May, it’s very pertinent.
What is the GDPR?
Some South Africans might not have heard about it, as it is an EU law. The GDPR, or General Data Protection Regulation, is new data protection and privacy law, adopted on 27 April 2016 which becomes enforceable on 25 May 2018.
You can read more about the GDPR here, but your business should be making these changes if you are storing any data on European citizens and residents.
What does it have to do with ML or AI?
The new law governs how companies like Facebook gather data on people residing in the EU, including facial biometric data. These companies now require consent to collect data in any way or to use facial recognition on photos. Many businesses are reviewing their data policies, security systems and products that are under development based on these regulations.
Beyond this, the GDPR refers to the use of AI and ML specifically in article 22(1).
“Any person has the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her.”
What does this mean for South-African companies?
Nothing – yet (unless your data contains information on EU citizens). But it’s good to be aware of, as many have touted the GDPR as the model on which all countries’ future data protection laws will be built.
And there you go – the top three developments in the AI and ML space you should be aware of. If you’re looking to keep up to date with South African insights into AI, ML and data analytics applications, subscribe to our blog notifications. We’ve also recently published a guide to using Machine Learning in your business, which you can download below.