My Thoughts on Current States of AI
With recent advances in computing power such as GPUs and distributed systems enabling massive parallel computations combined with big data for AI to learn from, there is no doubt that there will be an exponential growth in AI. Today, algorithms designed in neural networks running on GPUs are routinely used by applications running on cloud based architecture to perform sophisticated tasks such as face recognition in photos, and making intelligent recommendations. Here are some of my thoughts on current state of AI.
Autonomous Driving: This is a no-brainer. Every car being sold in the world will have this capability, in next decade or less. This is a narrow area of AI — with computer vision, controls and planning. There has been a lot of R&D work already completed for this technology over the last decade, and its just matter of making it robust, reliable and safe for consumers. There has already been quite a lot of effort by Google and other startup companies. Most recently, Tesla has demonstrated this technology to public and “Advanced Driver-Assistive System” is being shipped with Tesla Model S. BMW, Mercedes-Benz and Audi are jumping into action. Even Detroit automotive companies GM and Ford are trying to catch up. GM’s recent purchase of a self-driving car startup called Cruise for 1.3 Billion speaks for itself. However, it will never replace the manual driving for those who enjoy it. It will most likely be used for driving on highways, mostly long distance driving where conditions are easy to track for AI, and most human find it boring to drive. In addition, your car will have capability to automatically park for you and pull up to pick you up.
Personal Robots: There are companies already working on personal robots, such as Jibo, a “Social Robot for the Home”, and Amazon Echo. As exciting as it may sound, however, its more likely that those first generation of robots will fail in next few years. Certainly they are highly entertaining and cool devices, but, they do not serve real purpose and does not integrate well into our lives because it is location dependent. You can only access it where there is one, and you cannot take it with you everywhere. Perhaps more realistically, assistive social robot will be adopted first, such assistive care technology and robot in manufacturing — Rethink Robotics.
In next decade, personal robot will be available to you as AI software as a service model where it will be with you where ever you go. Your personal robot or assistance will be focusing on organizing your life — organizing data you generate, and make the data readily available to you when you need it, perhaps even before you know that you need it. Whether its in your home, on your phone, or in your car (potentially likes of apple car), and around public places, it will be accessible to you to assist you in organizing your personal information. It may be simple as understanding your calendar schedule, learn your everyday life pattern and recommend you solutions to implement them, such reorganizing your schedule, suggesting activities, and automatically suggest most efficient direction to places you want to go. You will able to simply ask question to your personal robot to find relevant information. Your voice will be main form of human to machine interface. There will be no need for keeping track of multiple organization software likes of Evernote, google calendar, apple notes and notepad etc. that scatters your data everywhere. You will be able to freely speak to it as if its with you all the time, be able to seamlessly ask any question about your schedule, remind daily tasks, and ask direction to places. It will also be able to recognize your emotions and recommend music or activities depending on your state of emotion and personal taste.
AI in Healthcare and Medicine: AI will be deeply integrated in our healthcare system without you even knowing it. IBM Watson is already able to provide recommendations to health care professional based on its learnt database to treatments. DeepMind (AlphaGo) is already planning to apply their reinforcement learning technology into assisting healthcare. Deep Genomics being headed by Prof. Brendan Frey at University of Toronto is developing new machine learning methods that can find patterns in massive datasets and infer computer models of how cells read the genome and generate biomolecules.