A Summer Diary of Artificial Intelligence Posted on : Aug 31 - 2017

The summer of 2017 was the summer of artificial intelligence (AI), as developments in the field of AI and robotics picked up pace globally. On the technological side, Internet companies continued to invest in the development of AI technologies aimed at enhancing their products, while researchers explored modalities for addressing concerns such as privacy, ethics, and accountability in AI decisions. Countries elaborated strategic AI plans aimed to position them at the forefront of developments. Alerts on the risks of AI for society seemed to become more alarming, while parliaments and governments discussed the economic, social, and ethical implications of AI. More details on some of these developments in our Summer Diary of Artificial Intelligence…

Major breakthroughs in AI

AI algorithms involve judgments and decision-making, replacing similar human processes. But, while humans can explain why they make certain decisions, this is not (yet) the case with AI systems. Concerns about discrimination and bias in decisions made by AI systems push researchers to continuously work on identifying ways to make algorithms ‘accountable’. In one example of such work, researchers at the Massachusetts Institute of Technology (MIT) are one step closer to finding a way to determine why an AI system makes one decision over another (e.g. why a driverless car makes certain decisions while on the road). Having looked at how artificial neural networks process information, they found that individual neurons in such networks could be highly correlated to high-level concepts involved in decision-making (such as identifying patterns and concepts in an image). In practical terms, identifying the specific neuron in an artificial neural network that is responsible for a certain decision could help explain why the AI system made that particular decision.

Efforts are also being made to enhance the efficiency of AI systems, as well as to make them safer. Researchers from OpenAI and DeepMind have been working on an AI algorithm that learns from human feedback, in an attempt to address problems associated with the concept of reinforcement learning ‒ an area of machine learning that rewards AI systems if they take the right actions to complete a task. Researchers proposed a method in which the reward predicted is based on human judgement, which is fed back into the reinforcement learning algorithm to change the agent’s behaviour. DeepMind is also working on new technologies for empowering AI systems with imagination. In one such proposed technology, an imagination-based planner could perform imagination steps before taking an action (proposing an imagined action and evaluating it). Another proposal describes imagination-augmented agents, which could interpret predictions from a learned environment model to develop plans for decisions and actions.

In the field of business applications, Amazon has launched a new service that uses AI to help identify and protect sensitive data stored in Amazon Web Services (AWS). The service, called Amazon Macie, relies on machine learning to discover and classify sensitive data such as personally identifiable information, and protect it from breaches, data leaks, and unauthorised access. Access to such data is continuously monitored in order to detect any suspicious activity, based on access patterns. Detailed alerts are generated when risks of unauthorised access or inadvertent data leaks are identified. In addition, Facebook has explained how it combines AI and human expertise to ‘keep terrorist content off’ the social media platform. View More