Monday, December 18, 2017

WHY ARTIFICIAL INTELLIGENCE IS BASICALLY DIFFERENT FROM HUMAN INTELLIGENCE



Recently, Dubai announced a minister of Artificial Intelligence. The 27-year-old Omar Bin Sultan Al Olama is the UAE's first State Minister of AI. A remarkable move by UAE Vice-President and Prime Minister Shaikh Mohammad Bin Rashid Al Maktoum, who said: "we want the UAE to become the world's most prepared country for artificial intelligence". The role focuses on future skills, future sciences and future technology and offers Dubai a chance to be ready for an AI-driven future. The move by Dubai is a great showcase of vision and dare to take risk, since artificial intelligence will increasingly define our future.

In fact, I would recommend any country that takes digitalisation serious to follow Dubai's example and appoint a minister of Artificial Intelligence. Even more, I believe that every organisation should employ a Chief AI within their organisation to understand how AI will impact the business and how it will affect organisations and societies. Why? Simply because Artificial Intelligence will be fundamentally different from human intelligence and understanding this can bring you a competitive advantage and can help us mitigate risks.

Why AI Will Be Different

Intelligence is "the complex expression of a complex set of principles", consisting of various interdependent subsystems all linked together. Intelligence exists because of evolution and it enables humans to model, predict and manipulate reality. This unique ability allows us to reason backwards and forwards from a mental image and reason on possible (un)desired future outcomes. Evolution created our intelligence, but evolution is an inherently dumb process. It achieves its goal by trial and error and does not possess the foresight capabilities that we humans have.

In fact, the unintelligent process of evolution has resulted in flaws in human intelligence. Due to various constraints such as food availability, trade-offs with other organs, biological materials it is possible that our brains have not evolved in the most optimised way. Artificial Intelligence, on the other hand, will be developed by (artificial) intelligent beings/things who do possess the foresight capabilities that evolution misses. As such, it is better capable to understand which materials and what processes work best for intelligence. In addition, it can apply the knowledge of others to improve (initial) designs and as such improve artificial intelligence. Consequently, artificial intelligence will look and act in ways unfamiliar to mankind today and, hence, requires a different approach.

Examples of Unexpected Actions by AI 

The best example to show how this works is AlphaGo Zero, the new and improved version of the AlphaGo algorithm developed by Google's Deepmind. While previous versions of AlphaGo had been trained using human data, this version of the algorithm is completely self-trained, without human intervention or historical data. A mind-blowing feature, especially because within three days of teaching itself how to play the game, it beat the version that beat Lee Sedol in 2016. It did so using a new form of reinforcement learning where the algorithm became its own teacher. It played against itself, hence it always had a component of equal strength, and learned from every move it made. While the previous version that beat Lee Sedol already surprised the world champion with moves he had never seen or thought about, the new version applies completely unknown strategies that it had taught itself. As a result, after 40 days it surpassed all previous versions of AlphaGo and, arguably, became the past Go player in the world.

Unexpected moves and strategies as developed by AlphaGo Zero increasingly start to appear in AI development. For example, AI developers from Google recently build algorithms that had to compete for scarce resources, resulting in increasingly advanced strategies to beat the component. Another project from Google Brain developed algorithms that created new encryption methods, unlike any seen before developed by humans, to protect information from other neural networks. Finally, Facebook recently developed two algorithms that created its own secret language, unsolicited and used advanced strategies to get what it wanted. Consequently, Facebook had to shut down its algorithms.

AI Policies to Understand Advanced Artificial Intelligence

These four examples clearly show that Artificial Intelligence is increasingly becoming more advanced, especially when not limited by human knowledge. In this process, it is adopting new tactics or strategies that humans have not seen before or could not imagine. In addition, as AlphaGo Zero shows, removing the need for human intervention and/or historical data makes algorithms much more powerful and efficient, so it is likely to see more of such developments. Bringing us ever closer to the holy grail of Artificial Intelligence; Artificial General Intelligence (AGI). These AI systems have autonomous self-control and self-understanding and the ability to learn new things to solve a wide variety of problems in different contexts using unknown tactics and strategies.

Although AGI is still years, if not decades, away, the more advanced AI will become, the bigger the impact on organisations and societies. The examples of Google and Facebook show that when that happens, humans might not always understand what happens or how an AI arrives at a certain strategy or outcome. Therefore, if we wish to remain in control, and not be surprised by increasingly advanced AI, organisations and countries should follow Dubai's example of appointing a senior leader responsible for Artificial Intelligence. Those leaders can then form overarching committees, similar to the Eurogroup that holds informal meetings of the finance ministers of the Eurozone to exercise political control over the Euro, to establish guidelines and policies in terms of AI development. Although such AI regulation should not hinder technical developments, it should also advocate measures to ensure safe AI and to help organisations and societies understand the impact of a fundamentally different form of intelligence.

No comments:

Post a Comment