This is why most if not all technology players are heavily invested in AI, with Google buying AI startup DeepMind in 2014 for $400 million. Microsoft Ventures investing in AI startups, Agolo and Bonsai. An Amazon even building their own Amazon AI entity with Amazon Echo being their first mainstream product. And this is not a game for technology giants only. Global investments in AI went from $0.6 million in 2012 to approximately five billion in 2016, and is estimated to reach $12.5 billion in 2017. So, more than 50 percent increase in only one year. This growth is expected to continue throughout 2020 when revenues will be close to $50 billion. More than 60 percent of executives in a joint MIT and BCG survey believe AI will have a large effect on businesses in the next five years with industries like telecommunications and financial services, probably leading the way. What does this mean for businesses? It means that if you are a business owner and you're not using AI, it is time to start experimenting and mainly, start understanding what are the potential applications for you. So, how do you do that? If you remember, we have decomposed AI into three fundamental elements; computing power, learning algorithms, and training data. Let's look at each one separately. Computing power. Computing power is becoming more and more a commodity. So, unless you are in a very specialized industry, you won't be competing there and you are likely to use off the shelf computing infrastructure potentially on a pay per use basis. Algorithms are everything but a commodity. They are a lively field of research and the limitation factor here is finding the right talent to keep up with this evolution. And I would argue that it's not only about data science or machine learning talent, it's also about upscaling the whole workforce to work in an AI enabled environments. Last but not least, the training data. It is more often than not a competitive advantage and business is done to keep its property. The key thing to understand here is that no algorithm can make up for missing or low quality data. Remember this very simple rule about AI, garbage in, garbage out. This also points us to one of the key challenges we face with AI. Even if we excel in all three fundamental capabilities, phenomenal computing power, best in class talent to keep up with advances in AI algorithms and very high quality data. To illustrate the challenge, let me share with you this philosophical riddle. If a self driving car has the choice between two courses of action. One, which will kill one pedestrian, and one where it will kill five. Which one should it choose? You felt like most people would want wanted to save the five and sacrifice the one. Now, what if the one is not a pedestrian, but the car owner in the front seat. Should the car sacrifice the owner for the greater good. If you say yes then you are coherent, but then would you personally buy this car. If you are like most people again you would not. And this is in case we cannot understand explicitly what choices the AI is making, which in most cases is unlikely. The most sophisticated AI algorithms are black boxes that learn from experience to make their choices, but no one is able to explicitly identify which input variable or which past experience led to which choice. This has serious ethical and sometimes legal implications. In 2015 for example, a study has demonstrated that AI algorithms can develop discrimination biases based on race or on gender. Let's say for example that you want to train an AI to help you select the best candidates for a given job description. If you based the training on the results of previous recruiting campaigns, you will likely be replicating the same biases that human recruiters might have against certain minorities. Or the algorithm might learn for example that female offers should be lowered than males, because that is still unfortunately the case in many companies today. Some of those dilemmas will likely be solved with improved understanding or better control of AI algorithms. But, some others will need policy and regulatory changes. So, what should you take away from this lesson. The rise of AI in the business world is powered by a combination of processing speed, learning algorithms, and availability of data. You should distinguish between General and Narrow AI. All AI you have seen so far is Narrow AI, except maybe in the movies. AI revenue is expected to grow more than 50 percent per annum throughout 2020. Early adopters that are able to upscale their talent pool and enforced data governance best practices will keep a competitive edge there. Even in the short and medium term, AI raises ethical questions that still require policy and regulatory answers.
