Mark Rodgers used to work in the oil fields connecting large pipes together. According to this Bloomberg article, his job was recently replaced by a robot, the Iron Roughneck. The machine helps tie together pipes and is becoming better with each generation. Beyond manual labor, many white collar jobs are being improved with AI powered software from accounting to engineering.
Artificial Intelligence is becoming smarter every day through techniques known as Machine Learning. To most people the term AI strikes fear in their heart as images from movies like iRobot, Ex Machina or the Terminator come to mind. Today these movies are still well within the realm of science-fiction, but the pace of development is picking up each year. You may have found yourself asking, what is AI? Or, more importantly, what is it capable of?
What is Artificial Intelligence?
AI is a very broad term and most generally refers to any non-biological entity or system which can acquire and apply knowledge or skills. The term generally refers to types of software programs, which allow for computers to perform advanced tasks.
Today computers are all around us from our cell phones to smart homes to cars. In many ways, they are already much “smarter” than we are with sites like wikipedia which hold more information than any one person could know or Google which can sort through billions and billions of pages almost instantly. One form of artificial intelligence is the calculator (often taken for granted because it was created over two generations ago). Consider, how the modern handheld calculator can do things no human can do (e.g. performing arithmetic on massive numbers in milliseconds). Since humans have a tendency to become well acquainted with novel forms of intelligence quickly, the term artificial intelligence is often pushed to whatever is outside the current capability. However, let’s not discount the forms of AI which are already commonplace.
Today, most software systems that can perform actions which require intelligence utilize a method known as machine learning. In short, machine learning is a field of study dedicated to building systems where the software acquires new insights directly from data rather than requiring explicit rule based instructions. Practically, it means the program is using an algorithm to solve many complex equations. An algorithm simply refers to the method used to solve these equations. Rather than writing explicit instructions developers or data scientists create the programs to solve these equations and give the system data to “learn from” or in other words solve for the unknowns within the equation.
The holy grail of artificial intelligence is known as General Artificial Intelligence (GAI) or Strong AI. The goal is to accomplish human-like application of knowledge and skills, where software could do “everything” humans can do from engineering to art to medicine. Defining, exactly what “everything” refers to is still a big unknown.
What is AI capable of in 2017?
There are no universal systems which can solve all problems and nothing that has a general understanding of the world. That is why all too often Siri or Alexa misunderstand what you’re asking, because they lack the context to interpret the words you’re saying. AI today is very very task specific and new software is needed for new tasks. The various types of problems can be broken down as follows:
These systems identify objects within images and many systems can outperform the best humans. For example, there are a plethora of programs which can detect various skin diseases or the software within self driving cars to identify the people, signs, and obstacles on the road.
Audio processing and generation
This is the software behind Siri, Cortana, Amazon’s Alexa, or Google Home which transcribes audio into text. There are also machine learning models which can generate text. The text is then processed by a natural language processing system to extract the intent of a statement.
Natural Language Processing and Understanding
These family of algorithms allow software to extract insights and information from text such as websites, books, social media or any written words. The latest can answer simple fact based questions, identify keywords and topics within a passage. There is an algorithm called Glove which builds a mathematical representation of the meaning of individual words. Advanced as these algos are, they don’t have a clear representation of what the words mean. There’s currently no way for the software to reason about ideas or understand themes within a text. This is why most voice assistants frequently misunderstand what you’re saying.
This class of algorithms are the most common and create predictions for everything from the stock market to whether or not you’re like a facebook post.
Logical Inference and reasoning
Being able to solve problems and deduce logical reasoning is an essential part of intelligence. There are a variety of programming languages and algorithms build to do just this. However, right now they can only handle basic logic and the real challenge lies in translating plain text into the appropriate format for them to be used.
The infamous IBM chess player or AlphaGo systems have been able to dominate world champions. There are some deep learning algorithms that can play all of the atari games and some teams working on beating the game Starcraft. Although very impressive they are all built specifically for playing specific games and can’t be used for much else.
Boston Dynamics and Rethink Robotics are two leaders in the space. Boston Dynamics latest version of the Atlas robot can do a backflip and run across varied terrain slowly. It’s clearly not as adept as people, but it does seem like they will be ready for many human level tasks. Rethink robotics specializing in “cobots” such as Baxter which work alongside people in a factory.
Where do humans stand out?
Given the wide array of top notch performance by many software algorithms or models, where will humans fit in? There are numerous doomsday videos and blog posts which highlight how machines are smarter than humans and how many jobs are going to be automated by AI, nearly 45% are eligible. There exists huge potential for this coming automation to improve our society increasing economic output and leisure time. How do we get there? How do we help people transform their skillset and transition to an ever involving workplace?
People are adaptable and can do a wide range of things. Although no one would think of building a skyscraper without the use of a crane or backhoe, people are still essential to the construction process. Same goes for AI, as it becomes more advanced, people will still be able to adapt and simply leverage AI to get more done, rather than be replaced entirely.
Here are skills which we have an unique advantage over machines:
Society is run by humans. We make purchasing decisions, business decisions and governmental laws. Robots can’t empathize with people and won’t be able to relate to what is worth doing, building or solving. For now, humans are best equipped to guide the ship.
Jobs: Entrepreneur, salesman, politicians, business analyst, etc
Humans have a unique ability to synthesize prior ideas into new ones. Our wide range of past experience and the way our mind can capture those ideas allow us to create novelty. There are software algos which can write art or music, but humans have the unique ability to communicate a poignant message which goes beyond a surface level understanding. Software will likely learn how to mimic these type of behaviors but is a long way from grasping them or creating truly novel and insightful messages or innovations.
Machines use algorithms and models to solve very domain specific problems. Machine learning is simply solving for unknowns in a long equation. Therefore, it is currently impossible for machines to solve problems vague, ill-defined or those which span across multiple subjects. For example, brainstorming why a new product didn’t launch very well in a new market or why a car broke down or how to modify the design of a washing machine to keep the motor from failing at 1000 hours.
Jobs: mechanic, engineer, financial analyst, etc
Fine motor movements
The human body is truly a masterpiece capable of a wide variety of tasks which are often taken for granted. The versatility of hands to type, adjust small screws, fix plumbing or tie wires together, make for a difficult challenge for any robot. The most advanced still have a long way to go and most can only handle course movements unless manually taught for a very specific task. A universal hand and system to identify, pick up and manipulate any object is still far in the future.
Jobs: construction specialities: plumbers, technicians, electricians, etc
Software lacks the ability to truly understand the meaning behind conversations or documents. Most programs use cleverly defined natural language processing tricks to match keywords or terms. The most advanced systems have the ability to learn rules and methods to answer complex questions which require first order logic such as the classic elementary school math word problems. For example, if Sally’s class has 10 toys and Sue’s class has 30 toys and they each have 15 students per class, how many toys are there per student? Although this is a very impressive accomplishment, these systems don’t have any intuitive understanding what the difference is between a toy, person, classroom or person.
Software doesn’t have any compassion for other people and can’t relate as to what a human’s perspective is in a given situation. There’s no intuition about what is painful, good, bad or anything in between. Decisions are made to optimize a very specific objective. If the people who design the system fail to include the objectives correctly or more likely forgot some very important but not obvious one, then it could result it a very bad outcome.
There’s no algorithm to identify how to motivate people. Therefore, leaders, managers and motivational speakers will all be in business for a long time to come.
Software can’t communicate at a 5 year old level yet and therefore doesn’t come close to sharing experiences or feeling to build rapport or true friendship. Therefore, salespeople, account managers and your friendly Starbucks cashier will likely have jobs for a long time to come. Software systems can mimic human conversation which could trick someone into feeling a connection, but until AI is autonomous and a self contained entity which experiences the world, it is unlikely to be able to truly connect with people.
Several other skills which software is not good at are: combining disparate sources of information to reach a conclusion, inductive reasoning, and dynamic tasks.
In short, AI systems are all around us and can perform a myriad of super human tasks, however there is no replacement for human creativity, abstract thinking or human connection. It would pay dividends to focus on building skills within these areas rather than memorizing facts, or routine manual jobs. The more repetitive it is the easier it will be to automate.