4. The Intelligent Farm

• Kevin make a farm visit to see up close how AI is giving food producers higher crop yields with less environmental impact. In addition to cloud computing, there is edge computing where intensive processing takes place in real-time at the point where the data is collected. AI is also making energy production and distribution more efficient. It can monitor machinery of all types remotely and tell the repair people that something is about to fail before it fails. On the conservation front, it can monitor large tracts of land to prevent poaching and allow scientists to study species over huge territories. Even consumer appliances can benefit from AI as they become part of the Internet of things (IoT).
• The goal is to augment human ability. It can do things like making smaller farms more profitable. This results in farms in more communities feeding the farm to market movement and creating jobs that require more technical skills. Schools need to respond to this need, which includes retraining some workers. By some estimates, 65% of elementary students will end up in jobs that don’t exist today. There are currently 150,000 unfilled data science jobs. It’s difficult to predict exactly which new jobs will arrive, but they will come. It’s important that a great deal of AI development has been open source. This promotes the formation of a group of developers to extend this technology. Kevin also notes that the Trump administration has been friendly to promoting Internet access to rural areas and signed an executive order aimed at maintaining American leadership in AI.

Part 11: Where We’re Headed, and How to Get There – 5. AI: Why It’s Needed

• Kevin sees AI as the next steam engine, the key invention that powered the industrial revolution and created developed countries out those who adopted it. The leaders in AI will likewise benefit economically to a great extent. He emphatically states that AI is going to be the most powerful technology we/ve built to date for transforming zero-sum games into non-zero-sum games. It has the ability to create abundance to support basic human needs, accelerate human creativity and entrepreneurship, and help us better understand our selves and the world around us.
• We need to use it to help provide care for the elderly and facilitate the learning of new jobs for people who lose theirs to automation. The positive way to see this is that those people will have somthing better to do. It will also allow the US to bring more jobs back as it’s cheaper to make things automatically here. Scientists of all kinds are working on new uses for things like drugs and therapeutics. Another example is customer service. Many companies are already using AI to do tier 1 support. Humans like this as they get to do the more complex support and customers like it because they get better support. In short, as is it removes constraints and creates abundance it will help us solve our most important problems.

6. AL: What It Is (and Isn’t)

• Kevin uses the example of replacing cardiologists with AI that can take ECG data and diagnose a patient. Rather than putting the cardiologist out of business, this creates an abundance that lets one doctor serve far more patients. It allows access for people who previously couldn’t get this treatment. When you add watches that can gather heart data that can be analyzed by one’s smartphone, the abundance increase and you can pick up symptoms before it’s too late. Also, for machines to learn, humans must look at and label data so the machine knows what to look for.
• Technological change comes in fits and spurts and it’s hard to predict. (Kevin and I are still waiting for flying cars.) Here we see the concept of a platform that can support many new technologies. Electricity is a good example. It’s not much good as is, but it can do amazing things when used properly. Other platforms include the Internet, the PC, cloud computing, and AI itself, which is an attempt to get machines to do intelligent things. It has three phases. First is systems of reasoning where you articulate a set of logical rules to manipulate knowledge. Next is systems of learning where you develop algorithms that can learn to emulate intelligence from large volumes of data. Last is systems of simulations where we teach AI agents to emulate intelligence through simulations. Examples are programs that can play chess or drive a car. They have to be clever rather than using brute force to avoid overwhelming available computing power.

7. How Models Learn

• Machine Learning uses data to build models and then uses those models to do intelligent things. Models called classifiers use it to determine what class something belongs to. A spam filter is an example. First, the data has to be labeled so it can be used to train the model. This usually involves humans doing the labeling in large numbers. This is phase two of AI. Phase one built reasoning systems. Data quality is key. If your data is biased your results will be biased too. Deep neural networks (DNN) are loosely based on the human brain but are much smaller. They can recognize images or translate text.
• They model neurons, which are either on or off. They pass signals to other neurons that either turn on or off based on the input from all the other neurons attached. Forgetting is part of the process because if you forget too little, you may not learn how to generalize. If you forget too much you may never learn anything. An effort is underway to automate the experimentation that produces DNNs. Machine teaching is another approach. Here rather than label tons of data, you take known characteristics of something and teach the model what to look for. Unfortunately, the field of AI is so broad and expansive that no one can fully understand what is going on so beware of the experts.
• Emerging AI often finds it’s way to the Internet in the guise of free products or services. You often get something in exchange for information about yourself. It is not wise to pretend that there are no side effects to AI use that comes free of charge. Another issue is Artificial General Intelligence (AGI), which is indiscernible from human intelligence for arbitrary cognitive tasks. Estimates of its arrival vary from five years to never.

Share this: