I would argue that currently artificial intelligence (AI) is by far the most popular technology being discussed, developed, and implemented. AI is being infused in a wide variety of products and services. You see it in the movies, automobiles, appliances, search engines, mobile applications, sports events, and smart phones, to list a few. If you perform a google search for the words “Artificial Intelligence” or “AI” you will get approximately 5 billion results, whereas, if you searched against Lionel Messi, arguably the world’s best offensive footballer (soccer player) in the modern game, you will only see 166 Million hits, 30 times less content. To put that in context, football is the most popular sport in the world and has a worldwide fan base of 3.5 billion people.
Let’s define AI as simply anything that makes computers act intelligently; understanding natural language and learning, either on its own or with the help of humans. In the last few years, I am sure many of us have experienced or interacted with AI in some way; knowingly or unknowingly. If you have composed a text message or used Gmail to write an email, you may have been presented with a list of suggested words to choose from to continue the sentence being typed. You will notice that the word options are contextually appropriate and consistent with words likely to be used by you when composing a sentence. What about when you call a service provider and you are required to speak to the automated digital assistant who ends up requesting, receiving and processing the information you provided audibly? These are two very common and simple examples of AI at work.
Based on current technological advancements in AI research, development and implementation, AI technologies can be assigned to one of three types; Weak, Strong, and Broad AI, or you may have seen them referred to as Narrow, General and Super AI. Weak AI like, Alexa, Siri and chatbots, solutions that have a single purpose based on pre-programmed responses, are today’s most commonly available AI solutions. We are years away from seeing Strong AI and decades away from experiencing Broad AI, where systems will be born with simple knowledge, but will be able to interact with the environment, learning from those interactions and as time passes, gaining common sense, and developing language and knowledge based on the context of its environment.
Currently many companies offering and implementing AI solutions (see graphic) do not see AI as a replacement to humans, but more of an assistant. They have also redefined AI to mean augmented intelligence; assisting humans in performing tasks more efficiently by providing information when needed so that decision making can be accomplished much faster, based on the most current and complete data set available. AI can be seen as one of the key components in helping us to continue advancing the human experience. In saying that, the stewards of these technologies have a responsibility to develop them so they are fair and free of bias, or if there is bias, be transparent about that bias with anyone who directly or indirectly uses them. As AI technology improves and we move more toward Strong and Broad AI, we are going to see AI solutions integrated into most of the products and services we purchase or consume, heavily influencing our lives. The AI we are developing now is the foundation for the AI technologies to come. We need to make sure we are thoughtful and intentional with each decision we make along this journey to ensure AI is for good.
AI has the potential of being one of the most transformative innovations that positively impacts the human experience. For that to happen, we all need to do our part to be aware of our intrinsic biases and develop AI solutions that have Trust, Privacy, and Transparency as their cornerstone. If you cannot trust the AI to be completely honest with you, you cannot trust the advice or information it gives you. Bias in AI today is primarily experienced through the use of machine learning (ML) algorithms. These algorithms use data and this data is provided by humans. We know that humans are intrinsically biased. If training data is biased and does not truly represent the scenario that it’s modeling, then the results provided by the AI system will be biased and inaccurate.
Think about being a recipient of advice from one of the following professionals: Doctor, Financial Advisor, Attorney, Dentist, Loan Officer, or Judge, and these professionals are being advised by an AI system. Wouldn’t you want the professional to know that the AI system is without bias, or, if there is bias, to be transparent about it? What if you were hiring an assistant and your assistant is an AI system, what would you want or need to know about that system? At a minimum, I would want answers to the following questions: Where did the data come from that was used to train the model? What data elements were included or excluded in the data set used to train the model? What algorithm was used to train? Why was that algorithm selected? Who did the training? I am sure you can think of many other questions you would like answered before you feel comfortable acting on the information provided by the AI solution.
So whether you are inventing new AI capabilities, developing solutions with AI, a user of AI provided information or a recipient of advice being influenced by AI generated information, we need to remind ourselves that AI is developed by humans; humans have implicit bias, so we must push for AI to be unbiased and also be transparent, by sharing details about all aspects of the solution where bias can occur.
Garfield Vaughn
IBM Integration Architect