Artificial intelligence is now indispensable

Love it or hate it, you can’t ignore it: Artificial Intelligence (AI) and its impressing advances are in the spotlight nowadays. All too often attitudes to this are polarized: AI is going to bring the end of humankind versus just let AI research develop on its own, it will magically overcome all issues you can think of!In the following two articles we aim to provide a more balanced perspective. We shall look at current researcher predictions regarding AI advances and achieving the technological singularity. We then discuss current challenges for AI and autonomous robots. In the end we jump to the future and take educated guesses regarding the impact of the singularity and what could cause a robot uprising.

The risk with AI is that we miss today’s issues by focusing on those coming in a few decades. Social network bots are influencing our public debate today. Same goes for models that are fed biased data and in turn learn that the man is a surgeon and his female equivalent is a nurse. Big data, wearables and internet of things represent a huge challenge to privacy as we knew it. Autonomous systems such as vehicles bring complex moral and accountability dilemmas. Autonomous weapons are on the way and multiple high-profile open letters have been submitted to the UN, asking for their ban. As we approach AI outsmarting humans, we will deal more and more with job loss, rethinking education and ultimately our role in the new environment.

Background

The pace of research and breakthroughs in the field of machine learning, as well as in broader AI, has accelerated in the past years. Deep learning just to mention one development, has brought/enabled impressive improvements in a wide range of fields, from natural language processing to computer vision and artificially generating realistic images or texts and has influenced most machine learning applications. This has in turn brought closer applications such as self-driving cars and drone-based shipping. Even a few years ago it seemed like a far stretch in the future that an algorithm would beat a human expert at game as complex as go.

Similar to the past technological revolutions, AI brings its own sets of challenges, pitfalls and misuse potential. Before we come to the situation that AI outsmarts humans there will plenty of issues to deal with along the way, while Terminator-like scenarios create most fear through the assumption that might have to deal with those issues right now, which our current understanding and current state of the society.

There is already a question to define what „robots smarter than humans“ actually means. Grace et al [1] carried out a survey of researchers at NIPS and ICML, two of the most reputable machine learning conferences, regarding the timeframe of robots overtaking humans. They introduced the term „High-level machine intelligence“ (HLMI), which is „achieved when unaided machines can accomplish every task better and more cheaply than human workers.“ While most research up to now focused on the logical type of intelligence, there are attempts tackling other types of intelligence as well and we can expect more work in the future focusing on enabling robots to, for instance, understand human emotions and adapt their behavior accordingly.

The authors distinguish between tasks and occupations being automated, or in other words between the possibility of all granular tasks being executed better by robots, and the actual combination and implementation of these tasks that make up occupations nowadays to be executed by robots. Thus, while the questioned machine learning researchers believe there is a 50% chance of achieving HLMI in 45 years, the complete automation of all human jobs is to be achieved with a 50% likeliness in 120 years. Certain jobs such as translating languages and truck driving are seen as more likely to be automated in the next ten to fifteen years, while others such as surgeon and math researcher are forecast to take more than 30 years.

to be continued

References:
[1] https://arxiv.org/pdf/1705.08807.pdf

Autor: Dr.-Ing. Irina Diaconita