Don’t mistreat your machine, it may just get its own back

Should we be wary of AI to come, wonders Prof Chrisina Jayne

In 1950 Alan Turing, in his article “Computing Machinery and Intelligence” proposed the question: “Can a machine think?” This question has fascinated computer scientists and influenced Artificial Intelligence (AI) research over the last decades.

AI could be described as machine intelligence, or a computer system that exhibits similar behaviour to a human being. AI has recently enjoyed a lot of attention in the media. According to Gartner Inc., AI is the top technology trend for 2017.

AI technology is becoming increasingly part of our daily life. There are numerous examples such as Google search, Amazon recommendations, Apple Siri, and self-driving cars. Governments and industry across the world are investing in AI research and applications to secure competitive advantage and economic progress.

Recent research breakthroughs in machine learning algorithms, the abundance of data and computing power have accelerated the development of AI even more. Machine learning algorithms are part of AI, providing the ability to learn from data and experience without the need to be programmed.

Major tech companies, such as Microsoft, IBM, Salesforce, Facebook, Google and Baidu, have bought start-ups and hired top researchers and software engineers. Google, for example, acquired the AI startup Deepmind in 2014 for £400 million, while Salesforce has recently acquired the AI company Metamind. Deepmind is the company that devel- oped the Alphago software, which defeated the world champion at Go – Lee Se-dol. Go is considered much more complex than chess and a real challenge for AI in games. The reason for this success to be considered so important is that Alphago employs a machine learning algorithm that can be applied to any task regardless of the specific domain. Metamind applied the state-of-the-art machine learning algorithm called “deep learning” for “Natural Language Processing”, which is now embedded into the AI Salesforce Einstein platform. Deep learning techniques have fundamentally changed how computer-based systems recognise faces, objects and speech. Nowadays systems utilising deep learning work as well as humans and are continuously improving their performance.

The ultimate goal of AI research is to create so-called Artificial General Intelligence (AGI) that can act as human and learn just like a human can learn any task. The recent progress in machine learning and neuroscience make the prospect of achieving human-like AI a real possibility.

According to a Microsoft Academic Search conducted in 2013, the top 100 authors in artificial intelligence predict that there is a 90 per cent probability of achieving AGI by 2070. This suggests that the top experts in the AI field do not doubt that there will be human-like AI by the end of this century. What does that entail? Will that type of AI result in improving the life on our planet and solve the world’s problems?

Demis Hassabis, CEO of Google