AI models can easily learn routine tasks that follow consistent patterns on which they can be trained using historical data. Yet other types of tasks or problems require an understanding of context of the situation or the ability to apply common sense reasoning to solve a problem. In these situations, AI models can struggle because of their inability to extend beyond the specific task for which they are trained or to understand broader context of everyday situations. Therefore, the use cases for AI systems fall into two categories. Automation refers to the idea of replacing humans with AI systems, to perform routine tasks or make standard decisions. Augmentation refers to supporting humans in making decisions or performing tasks rather than completely replacing them. Automation has been a long standing trend across many different industries, but AI significantly expands the scope of what can be automated. Now tests which involve images or understanding language can be automated using computer vision models or natural language processing models. A recent study by the global consulting firm Mckinsey of 800 different occupations found that 60% of those 800 occupations could have more than 30% of their activities fully automated by an AI system. The 60% included occupations which had physical activities performed and highly structured environments or activities involving data collection or processing of large amounts of data. Let's take a look at some examples of fields which are being heavily impacted by the use of AI to automate activities or even entire occupations. If you look warehouse logistics has been largely dominated by the use of automation, including intelligent robots and AI systems to manage the operations of large warehouses. Likewise, factory lines are becoming increasingly automated, even now including the activities of performing quality control, inspecting parts coming off the assembly line, through the use of cameras coupled with computer vision models to detect defects in products coming off the line. Field of transcription and translation is also now being impacted by AI models as companies increasingly use AI to transcribe speech to text rather than relying on human workers. Likewise, translation between languages used to be the domain of skilled human employees and they are now increasingly being replaced by AI models that can translate between one language to another. The field of customer support is also now seeing increased use of AI to replace human customer support workers with intelligent chatbots which are capable of understanding and answering questions from users. McKinsey estimates that 15% of the global workforce, or roughly 400 million jobs could be displaced by automation by 2030. Some job categories such as the ones we saw in the previous slide will significantly shrink or go away completely while other categories of jobs will see rapid growth. This will require an accelerated shift in the required workforce skills, a move towards providing employees with increased digital skills in computer programming, website design, digital marketing or even building machine learning models. An important benefit of this increase in automation and transformation in workforce skills Is that labor productivity growth is expected to rise from a level of roughly 0.5% in the years 2010 to 2014 up to an average of 2% due to the increased automation and the up skilling of our workforce. An alternative use of AI to automation is what's called augmentation or complimenting human intelligence rather than trying to replace it. There are a couple of key advantages of human computer collaboration, one is that we as humans and AI models have complementary skill sets. as humans we excel and understanding new situations, understanding context and causation, yet AI models are capable of processing vast amounts of data that we as humans cannot and distilling fine patterns within those large sets of data. When humans and computers collaborate, we also have human control over the process, meaning that the humans are in the driver's seat and can make use of Ai as a tool to complement their own intelligence when needed. In 1996 IBM's deep blue Ai system made headlines worldwide when it defeated the chess grandmaster, Garry Kasparov. This was a wake up call to society of the power that AI models now have to play games as complex as chess. Shortly after this point, a new field of chess emerged called cyborg chess. In cyborg chess, human players play against each other but each human can be complemented through the use of a computer model to assist them in playing the game. It's been found through studies that so called cyborg pairs or pairs of a human together with a computer system can outperform both the best humans on a standalone basis, but also the best AI systems on a standalone basis. The complementary nature of our human ability to apply common sense reasoning and understanding context, together with the computer's ability to process large amounts of data and learn patterns from large historical data have been shown to outperform AI systems alone, or even the best human experts alone. Let's talk about a couple of different forms of AI augmentation or approaches to using AI systems to augment human intelligence. The first approach is called triage, and the idea behind triage is that we're using an AI system as a first pass at something. When the AI system is able to make clear decisions one way or the other, it doesn't then need to go onto a human decision maker. Yet when there's a high level of uncertainty in the AI system's decision, when they're unsure about their decision, it's then passed to a human expert to make a final decision using his or her judgment. One example of triage would be the use of an AI model in insurance under writing. So determining the risk level and the appropriate insurance policy for different individuals applying for home or auto insurance. Machine learning models can be used to create a first pass to flag users that have a low level of risk and the clear them to go ahead with an insurance policy and also to identify users with a very high level of risk which the company may not want to underwrite a policy for. Yet users who fall somewhere in the middle would then be passed onto a human underwriting expert for further evaluation and to make a final decision. Another example of a triage application might be the use of a machine learning model for radiology. A model might be used to make a first pass, for example, looking for cases of pneumonia and scans of chest X rays and identifying either clear cases of pneumonia or no pneumonia but referring on certain cases onto a human radiologist for reading and interpretation. Another form of AI augmentation is what's called decision support. So in decision support, the AI model is providing its recommendation or insight based on data to a human expert but the human expert is taking that recommendation and applying their own common sense reasoning and context understanding to make the final decision. When we talked about the field of cyborg chest, this is the primary application of AI to support the decisions of the human players. Another example of decision support through the use of an AI model might be in investing where a machine learning model can process vast amounts of historical data about different stocks. But you still need a human investor to make final decisions by applying his or her common sense reasoning and understanding of what's going on within the broader market. Another application of decision support systems might be in determining medical diagnoses. Where an AI system could process historical data on different, similar cases, which have occurred in the past and the outcomes of different forms of treatment, and provide recommendations to a human doctor. But these would be cases where a human doctor was still necessary to make a final decision and diagnosis of the patient by being able to engage in dialogue with them and understand the full picture of their symptoms to make the final diagnosis.
